You may notice some images loading slow across the Open Collections website. Thank you for your patience as we rebuild the cache to make images load faster.

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Supply-oriented macroeconomics and the Greek economy : an empirical model Papatheodorou, George E. 1989

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


831-UBC_1989_A1 P36.pdf [ 12.36MB ]
JSON: 831-1.0098267.json
JSON-LD: 831-1.0098267-ld.json
RDF/XML (Pretty): 831-1.0098267-rdf.xml
RDF/JSON: 831-1.0098267-rdf.json
Turtle: 831-1.0098267-turtle.txt
N-Triples: 831-1.0098267-rdf-ntriples.txt
Original Record: 831-1.0098267-source.json
Full Text

Full Text

SUPPLY-ORIENTED MACROECONOMICS AND T H E GREEK ECONOMY: A N EMPIRICAL MODEL By George E. Papatheodorou B. A. Hons. (Economics With Mathematics) University of Sussex A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES ECONOMICS We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA September 1989 © George E. Papatheodorou, 1989 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Economics The University of British Columbia 2075 Wesbrook Place Vancouver, Canada V6T 1W5 Date: 2 $ £ £ r r ^ e e ^ _ / f Abstract Post-war Greek economic history is interesting and challenging to understand and model. Despite a promising start in the race to catch up with industrial economies, it has been losing ground in recent years. Since 1974 Greece has experienced business cycles and serious stagflation, a productivity slowdown and balance of payments crises. Until then, rapid growth with high investment rates and virtually no cyclical behaviour had been the rule. World economic events, domestic political events and the oil price shocks are expected to have contributed to the change, but to what an extent is a subject of controversy. This thesis attempts to analyze the structure of the Greek macroeconomy and to investigate such issues of economic history and related policy issues, through the estimation and simulation of a supply-oriented macro model. At the core of the model lies a supply block. Production structure is described by a long-run production function in capital, labour and energy, that defines normal output. A behavioural capacity utilization equation explains deviations of actual output from normal output in terms of unexpected sales, costs and inventory changes. Factor demands represent partial adjustments to desired factor levels derived consistently from the production function. Alternative hypotheses on output determination are tested against the null hypothesis of variable factor utilization but fail to reject it. An expenditure block covers imports, exports and consumption. The estimated in-come elasticity of exports is higher than the one for imports, to an extent that could justify optimism about improvements in the trade account, as long as the exchange rate stays close to PPP and Greek growth rates are comparable to the rest of the world's. It ii appears that recent balance of payments crises are mostly due to the collapse in unre-quited transfers. There is strong evidence of consumption smoothing through time. In the price and wage block one clear message that emerges is that Greece is a small open economy very prone to imported inflation. Estimated indices of real and nominal wage rigidity are about average for OECD countries. Estimates of the full-employment wage depend strongly on assumptions, and it is difficult to blame "excessive" real wages for continuing stagflation. Aggregate energy demand is split among different forms by interrelated quantity share equations, in order to endogenize energy price and energy imports. Government finance enters very simply. The monetary and international finance sectors are ignored for institutional reasons. Finally, simulations of issues of economic history and policy are carried out. Political instability in 1967 (the colonels' coup) and 1974-75 (the Cyprus crisis) is shown to have reduced the productive potential of the economy and caused an economic downturn, the latter being much more serious. The oil price shocks are shown to have caused lower output and capital formation, higher inflation and a deterioration in the trade balance. Fiscal and devaluation shocks have the usual effects, with devaluation fuelling inflation. The long-term out-of-sample stability of the model depends strongly on the specification of exogenous variable paths. Two other general conclusions are: 1. The Greek economy has slow speeds of adjustment to changes in equilibrium quan-tities, that have been exacerbating the problems from international and domestic supply and demand shocks. 2. The observed slowdown in productivity growth can be attributed to cyclical forces rather than a secular decline, although more observations may eventually refute this preliminary conclusion. in Table of Contents Abstract ii List of Tables ix List of Figures xi Acknowledgement xii 1 Introduction 1 1.1 A Short Historical Perspective 2 1.2 A Selected Literature Survey 10 1.3 Outline 24 2 Output Determination: Background, Specifications and Results 26 2.1 Background 26 2.2 Production Structure and Output Determination . 28 2.3 The Production Function: Nested CES and CES-Cobb-Douglas 35 2.3.1 General Characteristics 35 2.3.2 The Outer Function 37 2.3.3 The Inner Function 40 2.3.4 Parameter Estimates and Technical Change 42 2.4 Output Equation: Estimation Results 52 3 Output Determination: Alternative Hypothesis Tests 56 iv 3.1 An Exogenous Utilization Rate . . 56 3.2 A "Keynesian" Alternative 58 3.3 The Lucas Alternative 61 3.4 Is Money Neutral? The Barro Hypothesis 65 3.5 VAR: The Structure-Free Alternative 67 4 Derived Factor Demands 69 4.1 Derivation and Specification 69 4.2 Empirical results 72 4.2.1 Investment 76 4.2.2 Labour 80 5 Imports, Exports and Consumption 90 5.1 Imports 90 5.2 Exports 98 5.3 Consumption 104 6 Prices and Wages 119 6.1 The Main Wage-Price Block: Wages, Output Price and Consumption Price 120 6.1.1 Output Price . 120 6.1.2 Consumption Price 124 6.1.3 Wages 125 6.1.4 Wage Rigidity and the "Warranted Wage" 130 6.2 Absorption Price 134 6.3 Export Price 136 7 The Energy Block and the Rest of the Model 139 7.1 The Energy Sector 139 v 7.1.1 Quantities 140 7.1.2 Prices 146 7.2 Government Finance 148 7.3 Housing Investment • • • • 149 7.4 Money Markets and International Finance 153 8 Simulation Results 156 8.1 RMS Runs and Choice of Specification 157 8.2 Political History as Economic Shock: Cyprus Crisis and Coup 161 8.2.1 The Cyprus Crisis 161 8.2.2 The Colonels 168 8.3 Oil Price Shocks 172 8.4 Out-of-Sample Behaviour of the Model: Prediction Runs 173 8.5 The Long-Term Effects of Fiscal and Devaluation Shocks 180 9 Conclusion 186 9.1 An Overview of the Model 186 9.2 A Summary of Results 189 9.3 Some General Conclusions 191 9.4 Topics for Further Research 192 Appendices 195 A Data Sources, Problems and Methods 195 B List of Variables and Parameters 202 C Estimated Equations and Identities 209 vi C.l Supply Block 209 C.2 Domestic and Foreign Spending 213 C.3 Prices and Wages 214 C.4 Government Finance 216 C.5 Energy 216 C. 6 Balance of Trade 218 D Translog-Based Results 219 D. l The Production Function 219 D.2 Translog-Based Alternative Estimated Equations 230 Bibliography 233 vii List of Tables 2.1 Energy Demand Equations, Inner CES Function: Summary Statistics . . 44 2.2 Nested-CES- and CES-Cobb-Douglas-based Output Equations: Summary Statistics 53 3.3 An Exogenous Utilization Rate as Alternative Hypothesis (Dependent Variable: InQ) 57 3.4 From Keynesian Model to Factor Utilization (Dependent Variable: InQ) 60 3.5 From the Lucas Equation to Factor Utilization (Dependent Variable: In Q) 63 3.6 The Barro and Darby Equations (Dependent Variable: InQ) 66 3.7 Unstructured VAR versus Factor Utilization (Dependent Variable: InQ) . 68 4.8 Investment Equation, Partial Adjustment (Dependent Variable: Ine/Kne) 81 4.9 Investment Equation, Error Correction (Dependent Variable: ln(7n e//ne_i)) 82 4.10 Labour Demand, Partial Adjustment (Dependent Variable: In Lne) . . . . 86 4.11 Labour Demand, Error Correction (Dependent Variable: ^(Z-ne/^ne-i)) • 87 5.12 Imports as Buffers (Dependent Variable: lnM n e) 92 5.13 F-Tests on Alternative Restrictions 94 5.14 Alternative Import Equations (Dependent Variable: In MNE) . . . . . . . 95 5.15 Export Equations (Dependent Variable: \nXne) 101 5.16 Consumption Equations (Dependent Variable: Cpc) 108 5.17 Consumption Equations with Inflation and Interest Rates (Dependent Variable: C^) 112 5.18 First Difference Consumption Equations (Dependent Variable: ACpc) . . 116 viii 7.19 Energy Quantity Share Equation Parameters 142 7.20 Energy Quantity Share Equations: Summary Statistics 142 8.21 RMS Errors (%), selected variables (1) 158 8.22 RMS Errors (%), selected variables (2) 159 8.23 Cyprus Crisis Shock, % Deviations from Baseline, Dummy in pq (1) . . . 163 8.24 Cyprus Crisis Shock, % Deviations from Baseline, Dummy in pq (2) . . . 164 8.25 Cyprus Crisis Shock, % Deviations form Baseline, No Dummy in pq (1) . 166 8.26 Cyprus Crisis Shock, % Deviations form Baseline, No Dummy in pq (2) . 167 8.27 Coup D'Etat Shock, % Deviations from Baseline (1) 170 8.28 Coup D'Etat Shock, % Deviations from Baseline (2) 171 8.29 Oil Price Shocks, % Deviations from Baseline (1) 174 8.30 Oil Price Shocks, % Deviations from Baseline (2) 175 8.31 Out-of-Sample Simulation, Levels (1) 178 8.32 Out-of-Sample Simulation, Levels (2) 179 8.33 A 10% Fiscal Shock, % Deviations from Baseline (1) 181 8.34 A 10% Fiscal Shock, % Deviations from Baseline (2) 182 8.35 A 10% Devaluation Shock, % Deviations from Baseline (1) 184 8.36 A 10% Devaluation Shock, % Deviations from Baseline (2) 185 D.37 Translog Production Function: Coefficient Estimates 223 D.38 Translog-Based Output Equations: Summary Statistics 224 D.39 TL Production Function: Allen-Uzawa and Own-Price Elasticities, Se-lected Years 226 D.40 Additive Separability Tests, TL VAR. IFU model 228 D.41 Weak Separability Tests, TL VAR. IFU model 229 ix List of Figures 1.1 Profits as a Fraction of the Capital Stock 3 2.2 Log Energy Demand, Actual and Vintage-Based Prediction . 43 2.3 Labour Productivity Index and In ir 46 2.4 Labour Productivity Index and l n 7 r , Catchup Hypothesis 47 2.5 Labour Productivity Index, Actual and Cyclically Adjusted 49 2.6 Log Output, Actual and Normal-CES 51 2.7 Output Growth rate, Actual and Predicted, CES 54 4.8 Investment Equation (Partial Adjustment) 77 4.9 Investment Equation (Error Correction) . 79 4.10 Labour Demand Equation (Partial Adjustment) 83 4.11 Labour Demand Equation (Error Correction) 85 5.12 Import Demand Equation, Output-Based 96 5.13 Import Demand Equation, Aggregate Demand-Based 97 5.14 Export Equation, w. Igap and Rel. Price of Exports 100 5.15 Export Equation, without Igap, w. Profitability Variable 103 5.16 Log Non-Energy Imports, Exports and Trade Balance 105 5.17 Consumption Equation, Keynesian . 106 5.18 Consumption Equation, Habit-Persistence 109 5.19 Consumption Equation, Habit-Persistence w. Inflation Variable 113 5.20 Consumption Equation, Permanent-Income w. Inflation and Nom. Inter-est Variables . . . 114 x 5.21 Consumption Equation-lst Diff., Habit-Persistence w. Inflation Variable 117 5.22 Consumption Equation-lst Diff., Permanent-Income w. Inflation Variable 118 6.23 Output Price Equation 121 6.24 Consumption Price Equation 123 6.25 Wage Equation 126 6.26 Three Measures of the Wage Gap 132 6.27 Absorption Price Equation 135 6.28 Export Price Equation 138 xi Acknowledgement I gratefully acknowledge the help, guidance and inspiration I received from my supervisor Professor John F. Helliwell and my committee members Professors Philip A. Neher and Ardo Hansson, and departmental examiners Professors Keizo Nagatani and John Cragg. I would also like to thank Alan Chung, who drew on his modelling and data management experience for many helpful comments. Thanks are also due to Bob McRae for the simulator program that he developed and I used. I would also like to thank my colleagues at the Centre of Planning and Economic Research, and especially Professor Petros Livas and Dr. Panos Koutsouvelis for invaluable advice on the Greek macroeconomy. Thanks are due to the UBC Economics Department (faculty, staff and students) in general, and the Graduate Advisors Professors Don Paterson, John Weymark and Ken White in particular, for putting up with me through the years it took me. I am grateful to my mother- and father-in-law Jon and Elsie Visel for generous finan-cial support, and my parents Evangelos and Eleni Papatheodorou, who have pushed and exhorted and waited for a long time for this. Last, but by no means least, my heartfelt thanks to my wife, Robin Visel, who had to put up with me and provided me with support, motivation and inspiration all these years, and my daughters Elli and Georgia who, by arriving while this thesis was in gestation, made it all the more difficult and all the more worthwhile. To them this thesis is dedicated. xii Chapter 1 Introduction The purpose of this thesis is the estimation of a small macro model of Greece, with a substantial supply block and the possibility of energy applications; this model is put to use in order to explain the recent economic history of Greece, to gain insights into the country's macroeconomic structure, and to provide a framework for macroeconomic and energy policy analysis and forecasting. The basic modelling strategy used is the one behind the country-specific supply blocks of the OECD INTERLINK industrial country model, aiming for results that are as directly comparable as possible. The Greek economy is interesting and challenging to model. Its post-war economic history (described in section 1.1) poses interesting questions both about economic events and about the economic structure underlying the behaviour of the economy. In the rest of this chapter I argue that many of these questions can only be adequately addressed in the context of a supply-oriented model, something that to my knowledge has not been attempted so far. In addition, as I argue in the rest of this chapter and the thesis in general, Greece lies on the fringe of industrialized nations. Its level of economic and social development, the state of development and level of efficiency of its markets (especially financial markets), the degree of integration with world financial markets, and the impor-tance of supply-side constraints all place it in a position between industrialized countries and higher-income developing countries. The execise of applying specific macroeconomic techniques developed for industrial countries to a small, semi-industrialized open econ-omy, the difficulties involved, the limitations that become apparent, the insights gained 1 Chapter 1. Introduction 2 and the policy implications that may follow should be of broader intrest than the country-specific results obtained. Such an attempt to model the Greek economy, its successes and failures, might provide insights not only into the Greek economy, but other economies at a similar or lower stage of development. Despite the generally accepted importance of supply-side constraints in developing countries, most economic analysis of such countries (including Greece, as described in section 1.2) has focussed on financial flows, foreign exchange policies, monetary matters, and demand-side influences. This focus may of course be related to the relative abundance of financial (domestic and international) and Balance of Payments data, and the relative scarcity of reliable supply-side data. As it becomes clear in this thesis, incomplete or suspect data make the modelling tasks more difficult and throw some doubt on the results. On the other hand, a supply-oriented model reveals the presence of important economic relationships which remain hidden (or are misspecified) under previously used analytical frameworks. Before proceeding further with the outline of this thesis, let me provide some historical background on the Greek economy and put my research in the context of existing research. 1.1 A Short Historical Perspective Greece is a small country on the Mediterranean in southeastern Europe. It has a popu-lation of under 10 million, and a surface area of approximately 130,000 km2 (somewhere between Belgium and Austria), a good proportion of it distributed among scores of rocky islands. It is mostly mountainous and resource-poor. Fish stocks in the Mediterranean are mostly depleted, the mountains have been deforested for centuries, and few minerals are plentiful. Of metals only bauxite is found in relatively large quantities, and of fossil fuels only low grade brown coal is plentiful. Agriculture was until recently the major occupation, but tillable soil is scarce (just over 20% of total surface area) and often of Chapter 1. Introduction 3 raar Figure 1.1: Profits as a Fraction of the Capital Stock poor quality. The size of the agricultural labour force has shrunk from 54% of the total in 1961 to 28% in 1984. Urbanization has not brought commensurate industrialization; Greece seems to have moved from subsistence farming to a service economy. By 1984, agriculture, forestry and fishing represented 14% of GDP, mining 1.4%, manufacturing 24%, construction 5%, and services the remaining 56%. The Greek economy is one that is difficult to classify. It is not quite developed enough to be grouped with the Western industrial economies (with a GNP of US$ 33.5 billion and a per capita income of US$ 3,500 in 1984, it is the poorest Western European country with the exception of Portugal). Nevertheless it is not a less developed country either. A small country with a large Volume of international trade (exports and imports were 21% and 30% of GNP respectively in 1984), it seems a textbook example of the small open Chapter 1. Introduction 4 economy. While predominantly a market economy, it has a strong streak of "statism" in the form of pervasive, if not too effective, government intervention on all levels of economic life. It is an economy that emerged in 1949 from 10 years of destructive war, occupation and civil war with little infrastructure left intact. Since then, a predominantly agricultural economy has become heavily urbanized (if not heavily industrialized): over a third of the population lives in or around the capital, and a full half in the 3 largest cities. It had been growing at impressively high rates (8% annually on average) and with high rates of capital accumulation between the early 1950s and the early 1970s. It has since been plagued by low or negative growth, rising unemployment and two-digit inflation that has only recently been brought under 20%. Average aggregate profit rates1 have been squeezed from highs of almost 30% in the early 1960s to 5 or 6% in the 1980s (Figure 1.1), and investment has slowed dramatically as a consequence, as will be shown in Chapter 4. Changes in the social and political climate may have something to do with the above; the dictatorial regime of the colonels (1967-1974) helped keep wages down and profits up, slowing the inevitable erosion of (by Western European standards) abnormally high profits. The military regime fell in the summer of 1974, in the midst of the Cyprus crisis, general mobilization and the threat of war. There was substantial economic and social dislocation just before, during and after these events, as is evident in the time paths of all macro variables, especially employment (due to conscription) and investment (due to political uncertainty). 1974 is of course the year following the first oil price shock and the beginning of a world recession and rapid inflation, as well as a watershed in the conventional wisdom and focus of macroeconomic theory. 1974 also marks a new era 1 Profit is defined as a residual, net of the imputed wages of the self-employed and of depreciation, but not of interest costs. The profit rate is profits as a fraction of the capital stock. Chapter 1. Introduction 5 in the evolution of the Greek economy. By now Greece belongs firmly in the Western camp of cyclical economies; uninterrupted growth and low inflation have given way to stagflation, with a much higher inflation rate than the OECD average. In addition a chronic balance of trade deficit has developed into frequent Balance of Payments crises with the recent collapse in unrequited transfers, and the budget deficit has been around 16 to 18% of GNP for several years. To anyone with some knowledge of Greek history, these post-war economic events would be understandable in terms of broad features of the Greek economy and society, and the general policies of governments. Despite its largely agricultural nature, Greece has been in a real sense a "dual'' economy, with a mercantile capitalist class emerging under Turkish rule to take advantage of opportunities in shipping, commercial and fi-nancial services in the Mediterranean. This class has traditionally had an opportunistic mentality, which involved no national loyalty (in the absence of a nation to be loyal to), and no commitment to the undertaking of long-term investment projects. Short-term, high return ventures were the rule, and in light of the total absence of security of life and property under a despotic foreign regime such behaviour is understandable. This class of entrepreneurs was the only group with the requisite skills to undertake industrialization on a large scale, when that became possible. There was no incentive for a change in attitudes; security of fife was guaranteed in independent Greece, but the rules of the economic game kept shifting,2 sometimes catastrophically for those who trusted the governments of the day, and changes of governments led to repudiation of prior commitments. Thus industrial development had to be fuelled by favourable conditions: cheap labour, cheap and plentiful credit, trade protectionism, and, when applicable, •"Frequent wars, Axis occupation during World War II, and the 1944-1949 civil war made even survival uncertain. The rules of the game I refer to are not the ones Western macroeconomists are accustomed to, like monetary rules and tax reform. I am talking about repudiation of the national debt, hyperinflation, expropriation of property, suppression of labour unions and other drastic measures. Chapter 1. Introduction 6 guaranteed markets. None of these conditions are holding any longer, and profits have dropped dramatically. Until a new class of Greek entrepreneurs emerges that can operate under profit conditions more in line with international levels, and until the rules of the economic game become credible and stable enough to make them willing to do so, a return to the high growth rates of the 1960s may not be possible. Such a class is emerging mainly in services, but it still has the opportunistic attitudes described above, and it is still developing the skills for larger undertakings and, especially, competing abroad. Another engine of growth that may be gone forever is residential construction: urbanization has reached its limits, population growth is very low, and the housing stock has an extremely low depreciation rate. The fairly recent phase of Greek economic history described above has seen an in-crease in emphasis on macroeconomic modelling, and a movement away from a preoccu-pation with development planning. The trouble with this tendency, however interesting and timely, is that it has been characterized by predominantly demand-side approaches; the developmental approach was, at least, concerned with supply-side limitations to in-creased economic prosperity. I refer to "the trouble with" the new reality in the direction of research because, unfortunately, Greece has not solved its supply-side problems that limit growth and development. This does not mean that supply factors are irrelevant in industrialized countries; on the contrary, their relevance is increasingly recognized by macroeconomists. In the case of Greece, however, supply factors are, if anything, more important. Greece has been a rapidly growing economy with high rates of capital accu-mulation (until recently) and a strong trend towards urbanization and participation of women in the labour force. Coupled with this rapidly changing structure was relative lack of excess capacity, scarcity of skilled labour and specialized capital, and dependence on imported capital and intermediate goods. Thus its level of output (and consequently Chapter 1. Introduction 7 most key macro variables) could be expected to depend on, and be limited by, the avail-ability of factors of production, technological know-how and foreign exchange. In view of the above, the predominance of conventional Keynesian macroeconomics is a situation that calls for improvement. Given an economic history such as this, the purpose of this thesis emerged as an investigation of the structure of the macroeconomy of Greece in order to provide explanations for medium-term phenomena and tools for medium-term forecasts and policy analysis. I propose to do this through the estimation of a model with a sufficiently well-specified production structure for the purpose, a model that tries to integrate supply-side concerns with a serious consideration of the role of aggregate demand. The grander and, perhaps, more interesting and challenging questions of so-cioeconomic change that form the background to recent economic events, are not partic-ularly amenable to quantitative analysis on an aggregate level. Even more importantly, the future changes that are coming from integration with the European community can hardly even be guessed at yet: they are likely to entail a regime shift that no amount of recalculation of expectations can model. Such questions are forced into the background (but not forgotten), while I conduct the exercise of fitting a supply-oriented macro model to the time-series data set representing the Greek economy. They will have to be taken up again in the future, perhaps using different analytical techniques. The issues to be analyzed are numerous. Some of the questions I attempted to answer are the following: • How have energy prices affected energy use, investment, employment, productivity, inflation and the trade deficit? • Was the recent observed decline in productivity growth secular or merely cyclical? • What is causing persistent unemployment and inflation—is the nominal and/or real wage rigid, is it higher than the "warranted" or full-employment wage, is the Chapter 1. Introduction 8 process of wage and employment adjustment rapid or slow? • More generally, is the Greek economy as "ossified," as unresponsive to market signals and as slow-adjusting as casual observation, evidence of market failures and the country's general reputation suggest? On the other hand, policy issues that are easily amenable to analysis through simulation but have not been attempted yet would include: • has energy policy (e.g. taxation and pricing) been consistent with, or successful in, the pursuit of macroeconomic goals; and • what can we expect to happen to imports and exports in the medium term under different exchange rate and trade policies. None of the issues outlined above can be adequately analyzed without emphasis on supply. Although many pose empirical questions that can be answered using reduced-form equations, no understanding of the transmission mechanisms at work is possible from such equations, since they may describe any number of observationally equivalent economic structures. My primary aim is to uncover as much as possible of this underlying structure and the transmission mechanisms involved; forecasting success is a longer-term goal that may require more work than contained in this thesis, and more data than I had available. As will become apparent in the literature review and in subsequent substantive chapters, an underlying production structure is essential for properly defining the macro tools involved, and it provides much needed (if incomplete) consistency for the estimated behavioural equations. In addition, given the close to total dependence of the Greek economy on imported oil, and the severe impact (on output, employment, investment, growth and price stability) of the two oil price shocks, closer attention to energy and its interaction with relevant Chapter 1. Introduction 9 macroeconomic variables is necessary. A model that puts special emphasis on energy at the macro level would provide important insights into the problems of the Greek macroe-conomy. In addition, such a model could be an important contribution to the energy planning and development process, given the all-pervasive nature of government inter-vention and control in the economy. The range of energy-related policy options that can be exercised is so wide, that a coherent model that can analyze the macroeconomic im-pact of alternative energy-related policy decisions could be very useful. More information on the extent and form of government intervention in energy issues will be provided in the chapter on energy. Finally I would like to add that the financial and capital markets in Greece range from the imperfect to the non-existent. The banking system is completely controlled by the government either directly (through controlling interests in the two largest banks) or indirectly through 1. a maze of primary and secondary reserve requirements and 2. direct setting of the amount and allocation of credit and of interest rates. Thus the government controls both the money supply and the interest rate, which is set at below market-clearing levels, and, in the resulting chronic state of excess demand, decides who is allowed access to credit as well.3 Thus conventional ways of modelling the monetary sector are not applicable, since the credit and capital markets do not work properly and official interest rates bear little relation to the overall true cost of financ-ing. The domestic credit market certainly does not finance long-term investment, with the exception of loans allocated on largely political criteria. Working capital is provided with inventories as collateral, and the interaction of inventory stocks, the credit market 3The workings of the banking system in Greece could be understood in terms of artificial scarcity for the purpose of political control. Such a view poses interesting questions of political economy, not addressed here. Chapter 1. Introduction 10 and the cash flow of firms would be an interesting subject of research. In this sense the banking system operates as in, say, 19th century Western Europe, with financing of long-term projects avoided as much as possible, except again with official backing. The opportunistic mentality of entrepreneurs described earlier in this section closes the feedback loop, or, to use a normative term, the vicious circle. Under these conditions, self-financing, access to politically-brokered credit, and financing from abroad become important, but in all three cases the domestic interest rate is not particularly relevant. The actual magnitudes of credit from these three sources are presumably important, and the recent precipitous fall in investment may be related to the collapse of unrequited transfers from abroad, but with the exception of housing investment the effects are not quantifiable. Foreign financing very likely includes illegal, semi-legal and generally unre-ported activities, which is a possible reason for my failure to quantify its effects through officially reported figures. The fall in gross profitability is probably a better guide to the availability of self-financing than officially reported profits, and its effect is significant. The importance of the availability of official credit would probably surface in sectoral investment studies, but does not in the aggregate. Under these conditions the monetary sector and its effects on the real economy cannot be modelled meaningfully and will not be featured in my model. Given the a priori arguments above and in the absence of any evidence showing that fixed investment depends significantly on the (administered) in-terest rates (the most important transmission mechanism between the real and monetary sector in countries with functioning financial markets), this omission is, I trust, not as serious as it seems. 1.2 A Selected Literature Survey There is general agreement that aggregate supply deserves a central role in Chapter 1. Introduction 11 macroeconomic models, especially when such models are used for medium-term analysis. The more traditional demand-oriented models represent supply factors chiefly through price and wage equations, with the unemployment rate providing the main measure of unutilized supply potential, sometimes supple-mented by output relative to some measure, usually exogenous, of potential output. In "new classical" models, the level of output is supply determined, but little attention is given to the supply determination itself. This quotation from Helliwell et. al. [59] (p.76), sets the agenda for my proposed thesis, as well as accurately reflecting the state of macroeconomic research on the Greek economy. Until recently, the supply side of macroeconomics had been neglected in industrialized countries. While supply-side considerations are certainly important for these countries, it is my belief on a priori grounds, set out in the previous section, that the Greek macro-economy is even more supply-determined. The concentration on demand-side analysis may be excusable in Western industrialized countries, where excess capacity and unemployment (or its possibility) have been a fact of life for decades. In the case of Greece, however, conventional demand-side macroeconomics is even less appropriate. On the other end of the spectrum there are the "new classical" or real business cy-cle models with their emphasis on markets that clear continuously, full information and rational expectations. The assumption of continuously clearing markets is arguably the cornerstone of this approach. In the case of Greece, however, institutional constraints that pervade most aspects of economic life, plus many underdeveloped or even non-existent markets (the capital market being a case in point), make continuously clearing markets very hard to accept. In addition, every test I perform and equation I estimate points towards the same conclusion: the Greek macroeconomy is very often in disequi-librium, and prices and wages are not fully flexible; in fact every estimated speed of Chapter 1. Introduction 12 adjustment is very low by OECD standards. With regard to expectations, I tried to im-pose forward-looking rather than adaptive ones to the extent that it was accepted by the data; for example, the wage equation has a better econometric fit with forward-looking than with adaptive expectations of inflation. Rational expectations in their familiar form are associated with specific predictions and policy recommendations; these predictions and policy recommendations are grounded not so much in the expectations formation process itself, but the strong assumptions about markets that clear and policies that affect economic conditions through well-known transmission mechanisms. In the case of Greece these transmission mechanisms are far from clear, and often do not exist at all. For example, the existence of a short-term link between the rate of change in the money stock and inflation, whether through the real balance (Pigou) effect or the interest rate (Keynes) effect is, by all empirical evidence, quite doubtful.4 The issue of whether expectations are rational or not is less important than the nature of such transmission mechanisms and the existence of clearing markets. Thus I decided to use a pragmatic approach towards expectations: my first priority was to investigate the structure of the Greek macroeconomy with as much consistency as I considered cost-effective, given the tremendous computational burden of truly consistent expectations. Any structural model without fully rational expectations or continuously clearing markets is of course vulnerable to several criticisms, from the specific charge of lack of consistent microfoundations to the Lucas critique against structural macroeconometric models in general. These issues can be sidestepped (but not completely avoided) by taking the Sims [94], [95] line that no structure is the safest structure. The proliferation of empirical work based on reduced forms, or, to use their modern name, autoregressive models (and statistical algorithms for VAR, ARMA, ARIMA and VARIMA techniques) 4In the longer run such a link must of course exist, and the extent to which budget deficits are monetised is important. The latter issue has become much more important since the end of my sample, when deficits started growing out of control, and will have to be dealt with in any updates of this model. Chapter 1. Introduction 13 parallels similar developments in other social sciences (behaviourist psychology and cog-nitive science being the prime examples). This development towards prediction devoid of the complexities of explaining reality represents the reassertion of positivist method-ology. The methodological debate about whether theory-independent models are useful and whether a theory-neutral "observation language" is possible is far from over, how-ever. There are logical consistency problems involved in using "atheoretical" models to impose prior causal restrictions on theory and to evaluate policy, which are succinctly described in Cooley and LeRoy [29]. Beyond this, "black box" models that purport to predict without explaining anything are not particularly useful for my stated purpose, which was precisely to understand something of the structure of the Greek economy and to explain some aspects of its complex recent history, something that had to be done with a structural model. In any case, the structure-free models' lack of parsimony and of prior restrictions did not seem appropriate for a small (and less reliable than usual) data sample such as mine. I should add that I did not simply reject these alternative approaches on a priori grounds, but I tested them as alternative hypotheses in the explanation of the most crucial macro variable, output. The results, which are reported in Chapter 3, go a long way towards vindicating my decision. The same "duality" mentioned in the quotation can be discerned in Greek macroeco-nomic research. On the one hand, most macro research is of the conventional Keynesian demand-determined type, in which supply considerations enter only through the price and wage block; the unemployment rate is used as the main measure of unutilized po-tential, and is supplemented (in the face of rather serious measurement errors in the un-employment rate) by an exogenously determined industrial capacity utilization variable. Examples of this approach are Tsoris [100], Prodromidis [86], Garganas [44], Avramidis et. al. [8], Dimitriadou and Kouzionis [36], Spanos and Skordis [97] and Koutsouvelis Chapter 1. Introduction 14 and Karadeloglou [66].5 I should mention at this point that Tsoris [100] and other (unpublished) studies have attempted to estimate aggregate production functions (Tsoris used the Cobb-Douglas form). The problem with all such attempts is that the production function is not really integrated with the rest of the macro model, and in all cases the exercise did not give meaningful or statistically successful results because of the assumption that the economy operates on the production function (i.e. at normal capacity) at all times, thus ignoring important variations in the intensity of factor utilization. The problem of the lack of reliable data, especially labour data (examined in more detail below) also contributed to the poor performance. On the other side of the divide, there has been a substantial amount of estimation of small monetarist reduced-form models, almost exclusively at universities outside Greece and outside the mainstream of government-sponsored research. After a long period (be-tween the mid-60s and late 70s) of focussing on estimating money demand functions and modelling the monetary approach to the Balance of Payments, supply considerations started entering this strain of research. Two illustrative recent examples are Apostolakis [5] and Alogoskoufis [3]. Apostolakis estimated an aggregate translog cost function with labour, capital and money balances as inputs. Although his results are not comparable with mine (he uses a cost function, different inputs, fixed factor utilization and possibly unreliable labour data), his is, to the best of my knowledge, the only aggregate function that tries to explain output from the supply side, incorporates technical change, and has statistically significant and informative published results. As such, they may provide an instructive comparison. I find the emphasis on money balances as an input into the production process intriguing: my results point very strongly in the direction of money 5 None of these models, apart from Tsoris, have been formally published or are even in final form, and most explicitly do not authorize quotations, so my references to them are necessarily brief. Chapter 1. Introduction 15 neutrality. Alogoskoufis, on the other hand, estimates a three-equation reduced-form model that can be called monetarist and "new classical" at the same time. His model consists of three equations that determine output, the price level and the trade balance. Output is derived from an implicit Cobb-Douglas production function with labour, imported inputs and "productive government expenditure" as inputs (capital stock and technical change are only represented through a time trend). Wages are exogenously determined to stay on a trend path and labour input is fully adjusted ex post to real wages. A Lucas-type unexpected price effect is also added, and in the end the reduced-form aggregate supply depends only on price of output (actual and expected), the exchange rate, foreign prices, and government expenditure. Since the last three are exogenous, and prices depend only on the money supply (through a simple quantity equation) the structure of the supply function is greatly simplified. Finally a balance of trade equation is added, depending on domestic output and prices, the exchange rate and foreign prices. The predominant thread through the results and their policy implications is that all effects flow through the demand for money and, consequently, the price level and the terms of trade. I have very strong doubts about how useful an estimated demand for money function is in the case of Greece (given the chronic excess demand and credit rationing), and about how strongly monetary variables are connected to the rest of the economy: the mechanisms that link money with output, prices and the exchange rate are often simply not present, and my results bear this out. A third strain of research which I shall not concern myself with (although it is supply-oriented in a different sense), is input-output analysis, to which substantial government-sponsored research effort is devoted.6 6Another broad type of modelling approach that received some attention in the 1960s is neoclassical growth models. They have long been out of fashion, despite their simple and elegant way of characterizing the growth process through only a few variables. Chapter 1. Introduction 16 Finally, to go outside Greek research organizations and scholars, there is, to my knowledge, no macro model of the Greek economy that does justice to the supply side. The OECD, which has undertaken the most ambitious multi-country modelling project that incorporates a substantial supply block, has not yet finished the supply block for the group of small OECD counties that would include Greece. The paucity of supply-oriented macro research on the Greek economy, both domes-tically and internationally, is understandable given the paucity of reliable data. An instructive example is that the OECD has until recently had no capital stock series for Greece, not to mention aggregate energy use or energy price data. Probably the biggest impediment to supply-oriented research is the lack of reliable labour data: the Greek government did not gather, report and transmit to international organizations any em-ployment data until 1977 (with the exception of the census years 1961 and 1971), despite a universal but diversified social security system that could have been a goldmine of information. The only sector of the economy that has relatively good data (though with discontinuities and missing years) is manufacturing, and large-scale manufacturing at that. Thus many econometric investigations of issues ranging from the structure of the financing of investment (Tsoris [101]), economies of scale (Nicolaou [82]) and labour de-mand (Kintis [64]), to macro issues such as wage-price-employment dynamics (Economou [38]) have focused on manufacturing. So have estimations of production or cost functions for the purposes of analyzing factor substitution and technical change (Lianos [68] and [69], Kintis [64], Caramanis [24]). These data problems have of course confronted me too, but I have solved them in various ways, mostly by using data sources unreported to international organizations. Access to such data sources would have been difficult, but for my brief status as govern-ment researcher. The problem of the lack of directly observed labour data for the missing years I solved in the least unacceptable way by using OECD estimates, having satisfied Chapter 1. Introduction 17 myself that they were derived in a way that did not compromise their usefulness in pro-duction and factor demand functions. (More on data sources, problems and solutions in Appendix A). As for the modelling of energy issues, it has not gone beyond the micro level. There have been several attempts to estimate energy demand functions (Samouilidis [92], Rigas [89], Vlachou and Samouilidis [102], Diavolitsis and Balfoussia [35]) or to predict energy demand (Samouilidis [92]). Some of these ([92], [102]) use what I know to be unreliable data, and the rest use short time series. The reason for this is that they all disaggregate by sector of the economy, and the only reliable energy data prior to 1971 were aggre-gate data. None of these energy demand functions are aggregated across energy forms, and only in [102] are demands derived from underlying production and cost functions. Cavoulacos and Caramanis [27] are an exception in that they estimate factor input de-mands for capital, labour, electricity and liquid fuels derived from translog cost functions. Their data, however, are from manufacturing only, are possibly unreliable and excessively disaggregated (17 subsectors), but the results are instructive for comparison, especially with regard to the issue of factor substitution and complementarity. Thus it can be safely said that this attempt to estimate a supply-oriented macro model of Greece with energy applications, and to elevate supply factors to a central position while still paying due attention to the demand side, is an exercise which has never been undertaken before. As such it hopefully extends the limits of supply-oriented macroeconomics to a semi-developed country, with some useful theoretical, empirical and policy insights. In the remainder of this chapter I shall briefly attempt to put my model in the context of the relevant macro theory and similar empirical research on other countries. Rather than report on the theory behind Keynesian demand-determined macroeco-nomic modelling, I shall briefly comment on how existing Greek models of this kind Chapter 1. Introduction 18 fit within international experience. Models that treat output as demand-determined without reference to supply had been (and still are) very common, both in the case of national, stand-alone models, and linked multi-country models. For example, in all the linked national models comprising project LINK (for a review see Ball [11]), output was essentially demand-determined. In the case of Greece, two models were used in project LINK: the Centre of Planning and Economic Research (CPER) model (Prodromidis [86]) and the Bank of Greece (BoG) model (Garganas [44]). Output is demand-determined in both, and inventories are determined by a flexible accelerator system. The Bank of Greece model, in particular, was very disaggregated when it came to government fi-nance and the Balance of Payments, but had no real role for supply factors, even in the limited sense of using some capacity utilization measure in the price equations. In addition, wages were exogenously determined. I have no information about the current BoG model, since it has not been published and I had no access to it, except that output is still demand-determined. Finally, even the most recent of CPER models (Koutsouvelis and Karadeloglou [66]) treats output as completely demand-determined, although it uses an exogenous measure of industrial capacity utilization in the wage-price block. In recent years, however, supply-side factors have become increasingly prominent in macroeconomic literature and estimation. The introduction of supply factors has taken different forms. On one end of the spectrum lie "New Classical" models in which output is totally supply-determined. Sargent's [93] model determines output in terms of current and lagged employment plus a time trend, while the Lucas [71] supply function explains non-trend changes in output in terms of deviations of the price level from its expected value. Other New Classical models embed the Lucas supply function in reduced-form equations that explain the price movements themselves in terms of unanticipated changes in monetary and other policies (Barro [13], Darby et. al. [31]). The problem of the New Classical approach is its failure to explain stagflation, a Chapter 2. Introduction 19 failure it shares with traditional Keynesian and monetarist models: a typical shock would tend to move output and prices in the same direction in all three cases. The cause of this problem is to be found, in the New Classical case, in the fact that the underlying production function is assumed to apply continuously. The stagflationary experience of the 1970s and 80s is undoubtedly related to the increasing focus on the supply side, and the eventual attempt to combine supply and demand factors; stagflation could not be explained by demand factors alone, or by out-put being completely supply-determined. A large body of literature has arisen, which is designed specifically to explain the phenomenon of stagflation. In almost all of this literature (reviewed extensively in Helliwell [55]) a specific supply or production function forms the foundation of the models. Most models attach special importance to the role of energy and other raw material prices. Demand-side factors are often introduced in the form of demand variables in the derived equations for output or productivity growth. However, the applicability of demand-side influences is limited by the predominant as-sumption that the production function holds continuously. A seminal work within this general approach is Bruno and Sachs [22], and its principal result is that stagflation was mainly due to downward inflexibility of real wages. A common element in this literature is the definition and estimation of the "war-ranted" or full-employment real wage, and international comparisons of economic perfor-mance in a stagflationary period are often made on the basis of real and nominal wage rigidity. Of course the full employment wage cannot be properly defined without at least an implicit production structure, since it depends on the responsiveness of employment to relative factor prices. Given that the warranted wage is often (and certainly in the case of Greece) sensitive to the underlying assumptions, an explicit production structure is a necessary condition for a meaningful measure of the wage gap, otherwise it is hard to accept any apportioning of blame for high or persistent unemployment on "too high" Chapter 1. Introduction 20 wages. My findings on the wage-price-unemployment block are put in the context of this literature, and I refer to it in more detail in Chapter 6. Another attempt to introduce supply-side considerations in macro modelling was an OECD Secretariat model of the major OECD countries, as explained in P. Artus [6]. A three-factor, vintage, nested-CES production function was postulated, and by assuming cost-minimization by firms, consistent factor demand equations were derived for employment, investment and energy use. A concept of potential output was also derived from the production function. Actual output was demand-determined, however, although the gap between actual and potential output was used as an explanatory variable in the price equations. Thus, in this model, supply factors entered the consistently derived factor demands and endogenized the capacity utilization variable that drives the price block. However, it was still essentially a demand-determined model. It has since been superseded by a model described in Helliwell et. al. [59]. A different strand of research that also introduces supply determination of output uses a disequilibrium approach. It tries to divide macroeconomic reality into two regimes, one supply-constrained and one demand-constrained, with the economy switching between the two regimes, and with very different behavioural responses coming into play in each. An example is the French model METRIC [75], in which a cutoff point in unused capacity (set at 16.5%) determines whether output is driven by demand or supply factors. The real world does not have built-in switches, however. In the case of Greece the general approach of disequilibrium, dual-regime analysis has not been applied to output, but only to the money market in Dimitriadou and Kouzionis [36]. Since the money market appears to be in a chronic state of excess demand rather than fluctuating between excess demand and supply regimes, the usefulness of the exercise may be doubtful. Chapter 1. Introduction 21 Finally, the approach on which my own research is based, with the appropriate modi-fications made necessary by Greek economic reality and data availability, can be summa-rized by a few key articles by J. F. Helliwell and others. Helliwell et. al. ([58] and [60]) set out the evolving structure of the MACE model of the Canadian economy. Helliwell and Chung [56] explain the theoretical foundations of the variable factor utilization supply block analysis. Helliwell et. al. [59] explain the structure of the supply block of the cur-rent OECD model of the G-7 countries, a structure based on the MACE model. Fisher, Chung and Helliwell [40] re-estimate the production function, output and factor demand equations with a translog functional form, providing direct tests of the assumptions un-derlying the MACE and OECD nested-CES production functions. Finally, Helliwell [54] provides the links of the MACE approach to output determination with Keynesian, New Classical and structure-free models, which are subsumed as special cases of the variable factor utilization model. Econometric tests of the alternative models fail to reject the latter as a null hypothesis. A more detailed exposition of the variable factor utilization approach to supply block modelling will be given in the next chapter. For the moment I would like to say that, apart from the advantages of consistently integrating supply and demand factors within a model including an explicit production function, the separation of the long-term production function, that holds on the average, from the behavioural output equation has helped solve some of the problems that have been plaguing efforts to estimate a supply block for Greece. Leaving the supply block (which is the most important part of my work) aside, the rest of my model is fairly conventional. Imports and exports are explained using reduced form equations along the lines of the synthesis (cf. Alexander [2]) of the elasticities ap-proach (Haberler [51], Metzler [76] and Robinson [90]) and the absorption approach (for example, Machlup [72]). The import function, instead of having the usual explanatory Chapter 1. Introduction 22 variables of national income and the relative prices of imports (as, in the case of Greece, in Prodromidis [85], Prodromidis and Anastassakou [87], and Koutsouvelis and Karade-loglou [66]), attempts to follow the MACE approach of being embedded in an implicit CES aggregate utility function, with domestic output and imports as the two goods. Thus imports become part of the response to changing demand and/or cost conditions, together with output (or capacity utilization) and inventory changes. The export equation is an attempt to combine in a reduced form both supply and demand influences (as in Goldstein and Khan [46]). On the demand side it includes world income and relative export prices as explanatory variables. On the supply side it is enriched, as in MACE, by the inclusion of inventories (desired over actual), thus more firmly connecting it with the supply block, and by attempting to include a profitability of exports variable. The wage and price block share with most of the literature on stagflation the combi-nation of demand-side analysis and supply-side concerns. The former takes the form of an expectations-augmented Phillips curve (Friedman [42] and Phelps [84]) and capacity utilization as a potentially important influence on prices. The latter include unit costs determined by an underlying supply structure. In addition, the price of imports is in-troduced in the mechanism of price formation both at the level of output price and final (absorption and consumption) prices, a crucial inclusion given the openness of the Greek economy. Given Greece's dependence on imported oil, and the apparently severe consequences of the two oil-price shocks, I wanted to include energy as an important determinant of output and prices, and to make my model amenable to addressing energy-related policy questions. This emphasis on energy may not appear as justified for Greece as for heavy users and large producers of energy, like the G-7 countries, and Canada and the U.S. in particular. Energy does not enter costs as strongly, and the indirect effects of the Chapter 1. Introduction 23 oil price shock through world variables (principally world income—through exports and remittances—and inflation) may be more important than the direct effects on domestic output, investment, prices and the like. However, apart from my general aim of deriving results comparable to ones from industrial countries, and determining how the general OECD—G-7 modelling strategy applies to a country like Greece, I wanted to derive empirical answers to this o priori conjecture. Of the methods used, probably the most important step was to include energy as a third factor in the production function. Energy was treated as an input into the production process, because of the fact that energy is not demanded for its own sake, but for the productive services it can provide. The use of energy as an extra factor input has become quite widespread in macro modelling (like the two OECD models described by P. Artus [6] and Helliwell et. al. [59], and many of the models that deal with the stagflationary experience), but in the case of Greece energy demand modelling still stands apart from output, factor demand and price determination, and technical change. To me these latter seem more important issues than the modelling and prediction of demands for individual energy forms by individual sectors of the economy, which is the sole focus of current Greek research on energy issues. Instead, I used a consistently derived factor demand for a single aggregate of all types of energy. This I subsequently split into demands for the three major energy forms (petroleum products, electricity and coal products) by interrelated quantity share equations. This two-level method has been called the "total energy approach," used by McRae [74], the MACE model of Canada (Helliwell et. al. [60]) and others (for example, Berndt, May and Watkins [15]). Any other, more specific references to the relevant literature will be included in the chapters dealing with the specification and estimation of individual equations. Chapter 1. Introduction 24 1.3 Outline Confronted with interesting economic events requiring explanation and policy issues re-quiring analysis, and in the face of existing research that did not fit my purpose, I set out to build a small model that: 1. Elevates the supply side to a central position, while paying due attention to the role of aggregate demand, and 2. Integrates energy into the structure of the macroeconomy. In carrying out the exercise I hoped for two distinct, but related, results: 1. To gain new insights into the structure of the Greek macroeconomy, and 2. To provide a framework for answering questions on medium- and long-term macroe-conomic and energy issues, for evaluating macro and energy policy alternatives, and for deriving medium-term forecasts. A third, also related, motivation was to see how well a specific industrial-country supply block model applies to the Greek economy, and thus do my part towards the eventual integration of Greek national macro models with the OECD's new supply-oriented inter-nationally linked model. In my model, output is endogenized, and explained by an explicit, estimated un-derlying production function; this production function only holds on the average, and the (variable) utilization rate depends on cost and demand conditions. Thus the supply structure of the Greek economy is investigated at the macro level and integrated with aggregate demand. Factor demands are endogenous, consistently derived from the pro-duction structure. Imports and exports are integrated with the supply block. Prices are determined by factor cost (consistently derived from the production structure), prices of Chapter 1. Introduction 25 imports and demand conditions. Energy is introduced into the production structure; its demand is a macro issue and is integrated with energy supply. The task I set myself is clearly large. Fortunately the emphasis that needs to be given to the monetary sector and international finance is minimal, given the almost complete government control over money and banking, the absence of functioning capital and financial markets, and the strong insulation of the Greek economy from external financial influences. Most of the relevant variables are either exogenous or have no bearing on the rest of the model. Thus the total number of behavioural equations estimated is no larger than 18. In Chapter 2 I set out the theoretical background of the supply block, define or derive the specifications of the production function and output equations, and report estimation results on output determination. Chapter 3 reports the results of alternative hypothesis tests on output determination. In Chapter 4 I describe the derived factor demand equations and I present their estimation results. The same is done with the export, import and consumption equations in Chapter 5, and the wage-price block in Chapter 6. Chapter 7 covers many different issues, by describing the energy block, the government finance sector and residential housing investment, and tying the loose ends in the rest of the macroeconomy in an attempt to close the model. Some simulation results are reported in Chapter 8. Chapter 9 reports the conclusion and suggestions for further research. Appendix A deals with data sources and problems, and Appendices B and C list the variables and the model equations and identities. Appendix D reports results based on an alternative (translog) specification of the production function. Chapter 2 Output Determination: Background, Specifications and Results 2.1 Background The purpose of this thesis is try to give supply considerations the attention they deserve, while not neglecting the influence of aggregate demand. This is attempted by using a general approach, developed by J. F. Helliwell and used in the OECD's G-7 medium-term macro model (Helliwell et. al. [59]), that focuses attention on the disequilibrium adjustment processes that are set in motion when there is an imbalance between sup-ply and demand.1 This approach tries to integrate the New Classical concern with the fact that output is essentially supply-determined, in the sense that it reflects explicit choices by producers, together with the more traditional Keynesian concern with mod-elling domestic and foreign demand and the wage-price-employment block. In addition, it retains some elements of disequilibrium models by recognizing that decisions by firms and consumers alike are constrained by quantities as well as prices. Instead of the dis-tinct regimes (demand-side and supply-side determined) between which the economy is allowed to switch, however, demand and supply forces are allowed to operate simultane-ously, with the economy gravitating towards shifting equilibria. Equilibrium is a special case in which all exogenous variables have their anticipated values. But if final demand is determined by income and relative prices, aggregate output is directly decided by producers, and at the same time prices are not assumed to be flexible enough in the short run to lead to simultaneous equilibrium in goods and factor J I am indebted to the above mentioned paper for the arguments in this section. 26 Chapter 2. Output Determination: Background, Specifications and Results 27 markets, then an adjustment process has to be specified. This adjustment process is based on the role of inventories and utilization rates as buffers between aggregate supply and demand. Any differences between actual and normal inventories and utilization rates trigger changes in prices, production and imports. A rich system of dynamic interactions results, endogenizing output, factor demands, inventory stocks and prices, and Unking together all the blocks of the macro model. In order to describe the supply block of my model in terms of this approach, a key distinction has to be made among three different output-based concepts. Apart from actual output, Q, there is expected output, Q*, and normal output, Q„. Q* is a level of future output that is considered profitable and stable enough for producers to use as a basis for longer-term decisions about factor inputs. Given this expected output, producers choose the factor mix that minimizes costs, and decide on employment and investment plans that will bring about the desired factor levels. It is important to note that I use the terms "strategies" and "desired levels": all factors of production are considered quasi-fixed, with implicit costs of adjustment. The quasi-fixity of factors of production brings us to the third concept of output, Qn- Normal output measures the quantity of output that would be produced if existing quantities of employed factors were used at normal or average utilization rates. Because of the quasi-fixity of factors, there is usually a discrepancy between final demand and normal output, except in the special case where the evolution of variables is smooth and perfectly foreseen. Thus, according to this approach, the level of actual output is based on the level of normal output, Qn, which has a technological relationship with inputs of currently employed factors, and the capacity utilization rate, QIQn, which is determined by a behavioural relationship with demand and cost conditions and their interactions with inventory stocks. At the same time, factor demands are partial adjustments to desired Chapter 2. Output Determination: Background, Specifications and Results 28 factor stocks, denned by the output technology and expected output, Q*. The speed of these adjustments may also be influenced by cost and demand conditions. Some of the elements of this approach can be illustrated by the following thought experiment. An unexpected change in demand or cost conditions will elicit various re-sponses from suppliers. If, for example, there is an unexpected increase in domestic final demand, suppliers may 1. increase output by increasing capacity utilization; 2. increase imports to keep up with sales; 3. reduce inventories; 4. update forecasts of expected output, Q", and hence try to increase factor input; 5. raise prices. It is important to model all these responses in an integrated and consistent way, be-cause they can be expected to influence each other. For example, inventories are, in an accounting sense, the residual element in the supply system; however, a discrepancy between actual and desired inventory levels can be expected to influence output (through capacity utilization), prices, imports, and exports. In the same way, abnormal utilization rates will probably influence prices, factor demands, imports and exports. Probably the most important part of the model is output determination; I shall discuss it in the next section. Derived factor demands are closely associated with it, and they are covered in Chapter 4. 2.2 Production Structure and Output Determination At the heart of the model lies a long-run production function describing normal output, Qn-, and both output and factor demand equations are derived from it. Chapter 2. Output Determination: Background, Specifications and Results 29 The long-run production function has three factors of production: capital, labour and energy. Thus the output variable used includes energy inputs into the non-energy sector, and is equal to GDP at factor cost, plus net energy imports, plus net energy taxes. The agricultural sector2 was taken out of the supply block modelling, meaning that agricultural output was subtracted from Q, and inputs into agricultural production were subtracted from the inputs entering the production function and derived demands. The reason is data problems. In particular, labour input into agriculture is very likely upwardly biased in the early part of the sample, since figures for agricultural employment are based on rural population, or at least on all members of agricultural households of the right age. However, the size of rural households has shrunk over the years, and labour input per person officially engaged in agriculture has risen. Thus the fall in reported agricultural employment (from 54% of total employment in 1961 to 28% in 1984) is probably much larger than the fall in actual agricultural labour input, which would lead to biased parameter estimates.3 Another complication is caused by the fact that residential construction investment is estimated separately, because it is a very large component of total investment, is clearly influenced by different factors than other fixed investment, and the inclusion of the residential capital stock produces much worse fitting output and factor demand equations. Consequently the income or output from residential property (actual plus imputed)4 is also subtracted from Q. 2In 1961, agriculture, forestry and fishing represented 22.5% of GDP, and by 1984 their share had fallen to 14%. 3This disaggregation may cause problems, since it ignores the linkages between the agricultural sector and the rest of the economy. In a sense, when agriculture is taken out, inputs from the agricultural sector are similar to imports to the aggregate economy. Assuming no feedback between the two broad sectors may distort results. Given the data and aggregation problems that I faced, the disaggregation had to be done, and a properly modelled agricultural sector that is fully linked with the rest of the economy is a very important topic for further research. 4 As reported in the National Accounts. This measure is computed by an arbitrary rule from perpetual-inventory measures of housing capital. In the absence of a meaningful measure of the value and price of housing stock and housing services, there is little else the National Accounts could do, but the nature of the measure of housing services makes it all the easier to take out of the output measure. Chapter 2. Output Determination: Background, Specifications and Results 30 Normal output (like normal inputs) is of course not directly observable, and to esti-mate a normal output function one has to follow one of two methods: either impose a simple functional form and estimate its parameters from sample averages and a few re-gressions, or start with a set of parameter estimates based on the assumption of constant factor utilization and hence estimate normal output and factor utilization simultaneously by an iterative process. I tried both methods: the first with nested CES and CES-Cobb-Douglas functional forms, the second with a translog production function. In both cases Q„ is connected with actual output by a capacity utilization function; capacity utilization is variable and is described by a behavioural equation. In the translog case, the pitfalls of using a flexible functional form on aggregate and partly suspect data became apparent early on. The translog production function wraps itself around the data in such a way that the parameter estimates are not credible. For example, the resulting own-price and cross elasticities are very large, and they imply capital-energy substitution. This com-bination gives simulation results that are implausible. These doubts about the validity of the estimated production structure cast suspicion on the specification tests conducted using freely estimated translog structural parameters. In addition, the translog-based forms do not perform better in any equation other than output. On these grounds I chose the nested-CES functional form in the structure versus flexibility decision. The translog specification, estimation methods and tests are briefly referred to in Appendix D, along with a summary of the translog-based estimated equations that are different from the main model (as reported in Appendix C). A variable capacity utilization rate is dependent on two conditions holding simulta-neously: factor inputs must be quasi-fixed, and factor demand and cost conditions must be uncertain. Instantaneous and costless adjustments of factor inputs would keep firms on their long-run production functions even in the face of uncertainty, while even with quasi-fixed factor inputs firms could stay on their normal output paths if future cost Chapter 2. Output Determination: Background, Specifications and Results 31 and demand conditions were fully anticipated. Variations in capacity utilization thus represent a response to unexpected changes in demand or cost conditions by a decision to change the utilization rate of currently employed factors of production. The key choice facing the firm5 in this case is whether to change factor utilization or inventory stocks. Thus an unexpected increase in sales (representing a shift in aggre-gate demand) will be met by an increase in capacity utilization that is larger or smaller depending on how close inventory stocks are to their desired levels. Similarly, an unex-pected increase in unit costs relative to output price (that is, a fall in profitability) will influence the utilization rate through decisions of firms to operate at reduced capacity (or to shut down certain operations) because of low current profit rates,6 and will possibly interact with inventory stocks by increasing the marginal cost of holding inventories. Thus, assuming that the interactions of sales, costs (or profits) and inventory stocks with capacity utilization can be expressed multiplicatively, the capacity utilization func-tion is defined for both functional forms as Q/Qn^S^C^I, or lnQ = \nQn + f3slnS + pc\nCg + pIlnIgap (2.2) where S, Cq, and Igap are sales, unit cost, and inventory gap terms. To be more precise: 1. Sales: < s/Qn > where s = C + I + G + Xne (that is, domestic absorption plus non-energy exports) and <> denotes sample averages; 5There is, of course, a problem of aggregating up to the economy level from this analysis. As in other cases, I have to recognize its existence but have no choice but to continue despite it. 6Glick and Wihlborg [45] derive the same result on unexpected cost changes in a theoretical PV-maximizing model of output and inventory responses. (2.1) (2.3) Chapter 2. Output Determination: Background, Specifications and Results 32 2. Costs: C = P k K n e + P t L n e + P e E (2 4) ' PqQ or unit cost of output divided by current output price (C, is clearly an inverse measure of profitability); and 3. Inventories: Igap denotes the inventory gap and is defined as desired over lagged actual inven-tories. Desired inventories can be, and were, defined in two alternative ways: (a) As a constant fraction of capital stock: j ^ Kjnv / Kne 5* Kne /n r\ (b) As a fraction of normal output: r Kinv/Qn > Qn / r , c\ Igam = J7. (2-6) The expected signs for the 5, Cq an Igap variables are (+), (—) and (+) respectively. There are problems, mostly specific to Greece, with all these capacity utilization variables. The sales variable may contain all sales, i.e. may be defined as absorption plus non-energy exports, and thus contain non-energy imports. However, it may also be split into two variables, that is sales of domestically produced and imported goods and services. There were strong a priori grounds for treating imports separately, considering the Greek economy's heavy dependence on imports, and the different expected effects on output of sales of domestically produced output versus imports. Splitting the sales variable had no appreciable effect on the fit of the output equation. The problem with a separate import variable, furthermore, is that its sign is theoretically ambiguous: imports are substitutes Chapter 2. Output Determination: Background, Specifications and Results 33 for domestic supply (an issue dealt with in section 5.1). At the same time imports are positively related to national income through import demand. As for the cost variable, Cq, there are problems with the right choice of the price of capital. One of the main stylized facts about the Greek macroeconomy is the precipitous fall in gross profitability, if profits are measured as a residual and profitability is aggregate profits per unit of capital. Figure 1.1 shows this trend: the return to capital fell from 29% in 1963 to 6% in 1984. Although the extent of the fall can be attributed partly to the growth of the public sector, the fact of a large fall is not removed by concentrating on the private sector only; the extent to which Cq should reflect this fall is problematic. The role of the Cq variable in the OECD G-7 model is to capture the effects of short-term (or unexpected) fluctuations in costs or profitability on capacity utilization, and there are no sizeable long-term trends in the variable. In the case of Greece, however, the secular upward trend in Cq (or downward trend in profitability) in fact dominates cyclical move-ments. It may be argued that the long-term trend in profitability becomes incorporated in expectations, and as such should not be important for output determination. The issue posed is empirical, and a possible way of determining the relative importance of secular trend versus cyclical fluctuations is decomposing Cq into trend and fluctuations around trend. Doing so demonstrates that the secular drop in profitability is as important to output determination (and investment) as cyclical fluctuations. Thus the trend carries important information, which I did not discard by detrending the variable, except as an alternative hypothesis. Another important dilemma (with further implications for Cq) is the correct choice of the price of capital. I considered three alternative prices of capital. The first, pkc, is based on the replacement price of fixed capital, assumed to be the price of absorption pa, multiplied by a constant real supply price of capital, p. p is defined so that on average through the time period payments to the three factors of production exhaust the value Chapter 2. Output Determination: Background, Specifications and Results 34 of output: n = < 1 - (ptLne + PeE + SKnePa)/(QPq) > . 9 < (KnePa)/(Qpq) > ^ ' ) where 8 is the rate of depreciation. Thus the price of capital is: Pkc = + p)pa (2.8) In this way, the replacement price of capital, pa, is multiplied by the average return to capital. The second, pkf, takes into account interest rates to the extent that they enter the financing costs of firms. Given an average debt-equity ratio of 72% (from Tsoris [101]), Pkf is defined as: Pkf = (6 + p' + 0.72r)pa (2.9) where r is the interest rate and p' is defined similarly to p so that the factor payments exhaust output on the average: / = < 1 ~ (ptLne + PeE + (6 + 0.72r)KnePa)/{QPq) >  9 < (KnePa)/(QPq) > { • } Two interest rates were used, a long-term one for industrial investment and a short-term one. I preferred the short-term one in the end, not only because of better results (the long-term one was lower than the inflation rate, implying a negative real rate, in most of the period), but mainly because the industrial investment interest rate is only available to a few corporations on largely arbitrary criteria. In addition to the above, a third capital price that I considered (which I call pkc») tries to take into account the long-term falling trend of the gross profit rate: it uses four-year moving averages of the gross profit rate instead of the constant average return p. On a priori grounds, Pkc, the constant capital price, is the one to use in the production function; presumably, long-term decisions about factors of production should depend on a Chapter 2. Output Determination: Background, Specifications and Results 35 long-term average rate of return, while an interest-sensitive measure of the price of capital should affect the timing and speed of investment and other short-term adjustments. Finally, the inventory gap variable posed the most problems. As the only direct measure of excess supply or demand it is expected on o priori grounds to be important in the output, price and export equations. The problem, however, is denning desired inventory levels. Given any reasonable kickoff values of inventory stocks, inventories have been increasing much faster than output (a stylized fact possibly explained by the growth in complexity of the Greek economy), and even slightly faster than the fixed capital stock. Thus defining the desired inventory stock as a fraction of normal output (eqn. 2.6) causes a marked positive time trend in Igap. Thus in order to use the "output" definition of desired inventories I detrended the desired stock of inventories. 2.3 The Production Function: Nested C E S and CES-Cobb-Douglas 2.3.1 General Characteristics Following the example of the Canada MACE model and the OECD supply model, I started with a highly structured functional form for the theoretical long-run production function. Capital and energy are bundled together in a separable subfunction, and a vintage putty-clay structure for the capital-energy bundle that allows some degree of retrofitting. My reasons for imposing such a functional form were: 1. minimizing the complexity of estimation and saving on degrees of freedom, given my relatively small sample 2. using a vintage model, which I hoped would capture some technological information and provide a well-fitting energy demand curve Chapter 2. Output Determination: Background, Specifications and Results 36 3. imposing some structure on the long-run production function, since a more flexible form might be overly sensitive to unreliable data and specification errors, and might hug actual output too closely 4. making my results directly comparable to the OECD model (Helliwell et. al. [59]), which also uses a nested-CES functional form. The basic assumptions underlying the specific functional form used in this chapter are: 1. weak separability between the capital-energy bundle and labour 2. rigidity in the capital-energy ratio in the sense that installed capital can only grad-ually be remodelled to fit the new optimal capital-energy ratio following a change in relative prices 3. a simple structure of technical change, namely constant Harrod-neutral (i.e. purely labour-augmenting). The first assumption was based on Berndt and Wood's [16] results on the American economy, and the other two performed well in explaining Canadian energy demand. All three assumptions seemed intuitively plausible for the case of Greece. Results on Greek manufacturing (Cavoulacos and Caramanis [27]) indicate that capital and energy are used in close to fixed proportions, which would agree with the first two assumptions. The technical change assumption is further discussed below. The method used throughout to estimate the parameters of the production function again follows the methods of the Canada MACE model and the OECD supply model. It makes extensive use of sample averages, assuming that the production function holds on average over the sample period. Such a method of parameter estimation has been Chapter 2. Output Determination: Background, Specifications and Results 37 criticised as arbitrary, and the assumption that the production function holds on average need not be correct. However, since the production function describes normal rather than actual output, and since normal output is not observable, direct estimation of the parameters always runs the risk of picking up a substantial amount of cyclical rather than long-term behaviour.7 Such estimates would then incorrectly characterize the production structure of the economy. This is especially true in the case of the inner (capital-energy) function, where the vintage effects so complicate the process of free estimation, that some prior structure imposed by the broad properties of the data is necessary. An added reason I used this method of estimation is, of course, comparability. As for the assumption that the production function holds on average, it need not be correct: the slowdown of the early 1980s may be cyclical or permanent, but in the absence of enough observations this question, like the question of whether the productivity slowdown is cyclical or secular, will have to wait for a definite answer. Let me briefly explain this production function and the way its parameters are esti-mated. 2.3.2 The Outer Function In the outer function, the vintage capital-energy bundle, Kev (defined below in equations 2.21 to 2.25), is combined with efficiency units of labour, TrLne, to produce vintage-based normal output, Q „ v . Two key simplifying assumptions are involved: 1. constant returns to scale, and 2. Harrod-neutral (purely labour-augmenting) technical progress, incorporated in the labour efficiency index, n. 7 I believe this is one of the reasons my direct translog parameter estimates were so strange. Chapter 2. Output Determination: Background, Specifications and Results 38 The first assumption is standard in the literature and is used for simplicity. The second, adopted again for simplicity and convention, does not seem unreasonable in the face of steadily increasing real product wages and declining real rates of return to capital. The two assumptions together greatly simplify the characterization of technical progress in the model and the task of estimation. The outer production function can be either CES: Qnv = (jt (irLn)1^ + v K e v ^ r _ l (2.11) or Cobb-Douglas: Q«v = a0(Kev)a(xLne)1-a (2.12) In both cases, parameters are estimated by a combination of regressions and sample averages. The Cobb-Douglas case is much less complicated. First we can use the average share of payments to capital and energy to determine the exponent: Q =< 9h^tEL > (213) We can then determine ir and a0 by imposing the condition that over the sample period Qnv has the same mean and trend growth rate as actual output. Thus the rate of growth of can be derived by regressing ln[(Q/(K°vLn~a)] on a time trend. If ct is the coefficient on the time index, ir grows at a rate of c t/(l — a). From this, by using the normalization of 7 T = 1 for 1970 (the base year of my data set), a series for ir is constructed. Finally, the constant is estimated as: Cto = „ , T v , (2.14) In the CES case, matters are more complicated. First we can derive from the produc-tion function the cost-minimizing ratio of efficiency units of labour to the capital-energy Chapter 2. Output Determination: Background, Specifications and Results 39 bundle:8 where pte is the cost-minimizing price of capital-energy bundle, denned in equation 2.26 below. Because of the assumption that, on average over the sample period, Qnv is equal to Q, and that9 < i r L * / K : > = < T L n e / K e v > (2.16) we can substitute actual for normal values in (2.11) and then rearrange to solve for x: 1=L \ r-1 If we then substitute (2.17) into (2.15) and take sample averages, we can solve for v, given a value of T : • „ = <(PkeM(Q/Lne)^ > _ < (Lne/Kev)r > + < (pke/PtXKeJLne)^ > Then rewrite (2.17) to define output attributable to labour as: r - l r Q r - u K e v f*n - = — m (2.19) We can then estimate the rate of growth of ir, the labour efficiency index, by regressing the logarithm of the RHS of (2.19) on a time index, using OLS. Normalizing so that 7 T = 1 in 1970, we can get a series for w and an estimate for /x, given a value of r. 8Throughout the thesis, I use K* rather than ^ e » , because cost-minimization takes place at the margin and the optimal bundle of capital and energy is not vintage-based. This will become clearer in the next subsection. 9The following equality only holds (asymptotically) if relative price shocks are random, or symmetric on average. During the sample period, however, two upward oil price shocks make this unlikely. Beyond extending the sample period to include downward movements (1986-7), something I am unable to do due to lack of data, there is little that can be done about the problem. Chapter 2. Output Determination: Background, Specifications and Results 40 Finally, it remains to estimate r, the elasticity of substitution in the outer function. An arbitrary value of r can be used as a starting point to calculate initial estimates of p., v and ir. Taking logarithms of (2.15) gives: ln(irL'/K:) = r ]n(p/v) + r \n{pkeir/pt) (2.20) which can be estimated by OLS to give an estimate of r, equal to the coefficient on the price ratio. The new estimate of r is used to calculate new values for p,, v and ir. The process is repeated until the value of r converges. 2.3.3 The Inner Function The inner function assumes a putty-semi-putty vintage structure,10 such that only a fixed proportion of existing capital can be retrofitted in accordance to the new (marginal) optimal energy-capital ratio. New investment of course does incorporate the latter. Thus the vintage-based capital-energy bundle, Kev, that enters the outer production function, is defined as follows: Kev = {l-8-^)Kev_l+Inew :^ (2.21) where 8 is the rate of depreciation, if; is the retrofitting rate and I„ew is investment malleable in energy use (I„ew = Ine + ^ >-K"ne_i)- m w o r ( i s 5 the current vintage-based capital-energy bundle, Kev, is last year's bundle,11 less the amount lost to depreciation and retrofitting, plus new investment (including retrofitting) times the optimal marginal ratio jj^z. The latter is derived from the fact that in the inner function, capital and energy are combined on the margin as follows: a Ke = (pKnT + 7 ^ ^ ) "~ l (2.22) 1 0The terminology and the theory are from Fuss and McFadden [43]. "The initial value is for 1963 and is assumed to be equal to Ke, the marginal bundle (denned in 2.22) for that year, a decision defended by the fact that energy prices had been relatively stable in that period. Chapter 2. Output Determination: Background, Specifications and Results 41 where cr is the elasticity of substitution between capital and energy. Given the prices of capital and energy, the optimal (cost-minimizing) marginal energy-capital ratio is *LL = (IE*X (2.23) Solving (2.23) for E* and substituting E* and K* for Kne and E respectively in (2.22), gives us or K: 0 + 1 (lPk\ \PPe) < T - \ K* K* 0 + 1 IPk <r - l (2.24) (2.25) Associated with the marginel capital-energy bundle (Ke) is pke, its cost-minimizing price, denned as Pke = {^Pk1-' + 7 V " ) ^ (2.26) Finally, using the same analysis as above, the vintage-based energy requirement is denned as Ev = (1 - 6 - ^E^ + Inew^ (2.27) with | £ denned in (2.23) above. Given any values of a and ip, parameters (3 and 7 can be derived as sample averages. The ratio of 7 to /? is then 1 = (< E/Kne > I < Pk/Pe >*)* (2.28) and j3 can be derived by denning Kev units so that < K„e/' Kev >— 1. Finally, cr and ip are jointly estimated so as to minimize the standard error of the energy demand equation (see below), through a grid-search process. Chapter 2. Output Determination: Background, Specifications and Results 42 2.3.4 Parameter Estimates and Technical Change The estimation of the inner function parameters is common to both functional forms. The capital price used for the estimation of production function parameters12 was PkC, the constant capital price, as opposed to pkj, the interest-dependent one, or pkcs, the one dependent on moving averages of profitability. The fit of all supply-block equations was much better with the constant capital price, as opposed to the other two. The inner function parameters a and ij> (elasticity of substitution between capital and energy, and the retrofitting rate respectively) were determined by a grid-search technique as the ones that minimized the standard error of the energy demand equation. In this equation In 22 was regressed on lnEv, with the coefficient restricted to 1. A political dummy variable, Dcypr, was included for the years 1974 and 1975, to take into account the serious dislocation of the economy during the Cyprus crisis, general mobilization and the fall of the colonels. An added exogenous effect that may be picked up is the introduction of energy rationing in part of that period. The coefficient on the dummy variable (which, as elsewhere, was formed by combining two statistically identical dummies to save degrees of freedom) was statistically significant. Estimation was carried out by 2SLS. Using 2SLS gave the following parameter estimates: a = 0.55,0 = 0:2,0 = 0.692,7 = 0.0401 The fit of the energy demand equation estimate is shown in Figure 2.2. Its summary statistics are given in Table 2.1. In order to explore the possibility that there is energy-using bias in the technical change, I introduced a time trend in the energy demand equation. The new estimated values of parameters a and if; were not very different: a = 0.65, ip = 0.2 1 2 For a priori reasons given in section 2.2. Chapter 2. Output Determination: Background, Specifications and Results 43 T3 4 ) 11 10.3-o < C o E o > » o 1_ 10-9.5 c 9 -o a.5 Lagand • W t f i . -7S —I— ao es Year Figure 2.2: Log Energy Demand, Actual and Vintage-Based Prediction The coefficient on the time trend was statistically insignificant, and the fit of the energy equation is almost identical, as shown by the summary statistics in Table 2.1. A time trend is of course an ad hoc way of capturing energy-using bias, so in the absence of any empirical justification for keeping it I dropped it and will not report it subsequently. In the nested-CES case, the outer function parameters were estimated in the way de-scribed above. The value of the outer function elasticity of substitution, r, is inordinately large (r = 1.376, and significantly different from 1: F-statistic = 13.7), compared with results for the G-7 countries (Helliwell et. al. [59], p. 88), especially since on a priori grounds, given the rigidities in Greek factor markets, one expects it to be smaller rather than larger. The size of this parameter points to the possibility of cyclical effects being picked up in the process—such an estimate might be upwardly biased since changes in Chapter 2. Output Determination: Background, Specifications and Results 44 InE Trended Untrended Constant -0.01513 (3.008) -0.01854 (1.860) ln£„ 1.00 (restricted) 1.00 (restricted) Dcypr -0.11166 (6.693) -0.11850 (7.096) t 0.00019 (0.250) R2 s.e.e. D-W F-test on restriction 0.9983 0.022493 1.606 0.0147 0.9983 0.022488 1.667 0.0888 Table 2.1: Energy Demand Equations, Inner CES Function: Summary Statistics Chapter 2. Output Determination: Background, Specifications and Results 45 factor proportions may be partly due to cyclical changes in demand and costs and not to changes in the relationship between factor proportions and factor price ratios. I tried to disentangle the two effects by regressing \n(irL* / K*) on its lagged value, on ka.(pkew / Pi), and on the three utilization variables In 5, In Cq and Igap. The sum of the first two co-efficients was used as a non-cyclical estimate of r . 1 3 The new estimate of r was used ' to calculate the values of //, v and ir. The process was repeated until the value of r converged. The final estimates of T , v and ir were: where t0 = 63. Before the results of the output equation estimations, let us look at tests of alternative technical progress hypotheses. There has been a marked drop in the measured rate of growth of labour efficiency, as defined by eqn. (2.17), shown in Figure 2.3 normalized and in logarithmic form, along with the log of the fitted constant-growth ir. This drop in labour efficiency is common throughout the OECD countries, and a debate has been going on for many years on its causes. The first hypothesis, called the "catchup" hypothesis, assumes that there has been an international convergence in productivity growth rates: the growth rate of the labour efficiency index would depend on the ratio of the efficiency indices of the US 1 4 and of Greece. The equation used for testing this hypothesis was: 1 3The same process has been followed in the G-7 model after the publication of [59], and the estimates of T also generally approached 1. 1 4The US has had the highest total factor productivity and the lowest trend productivity growth rate. The way the hypothesis is tested, however, allows no room for conjecture about convergence of absolute levels of efficiency; the indices used are not comparable, unless defined in a common currency and constant PPP exchange rates, and the log-difference equation obliterates any differences in absolute efficiency levels. r = 0.937, n = 0.4698, v = 0.3598, ir = exp(-0.26756 + 0.018108 r), (2.29) Chapter 2. Output Determination: Background, Specifications and Results Legend • > C T U * L rear Figure 2.3: Labour Productivity Index and Inn-Chapter 2. Output Determination: Background, Specifications and Results 47 o. 3 O o x -o C o 3 •o O 3 O a 1.4 -i 1.2-1-0 . 8 -Leg«nd • » C T U » L • Q ' l t l -0 . 6 —i— 6 5 7 5 rear Figure 2.4: Labour Productivity Index and ln7r, Catchup Hypothesis The second hypothesis, called the "embodiment" hypothesis, assumes that the growth rate of efficiency depends on the rate of growth of the capital stock, which is a reasonable conjecture given the slowdown in fixed investment (it is shown in Figure 4.8 in Chapter 4). The equation used for testing this hypothesis was: h f e ) - * + A h , f e ) + * ( 2 - 3 0 ) where Kav is a 3-year (current plus two lags) moving average of the capital stock increase. The two hypotheses can be tested jointly in the context of a single equation: Two different tests were conducted. First, the "unadjusted" labour efficiency index from equation (2.17) was used in the definition of the dependent variable. The coefficients, Chapter 2. Output Determination: Background, Specifications and Results 48 with t-statistics in brackets, were: <*! = 0.18431(1.82), ft = 0 .05756(L12) , 7 l = 0.15803(138), 72 = 0.05047(1.00) On the basis of a 1-tailed hypothesis test, there is evidence of "catchup" behaviour. When the "catchup" hypothesis equation (2.30) is estimated by itself, coefficient (a i ) is significant at the 5% level. The joint test of the two hypotheses reduces the "catchup" coefficient to insignificance (at the 10% level). Figure 2.4 plots the predicted Inn- from equation (2.30). On the other hand, there is no evidence of "embodiment" behaviour, whether it is tested singly or jointly with the "catchup" hypothesis. Secondly, the same three equations were run with the labour efficiency index adjusted cyclically by multiplying by the estimate of the utilization rate, Q/Qnv In this exper-iment, changes in capacity utilization account almost completely for deviations of the observed labour efficiency from its trend, and all coefficients become insignificant.15 This is clear also from Figure 2.5, where the actual and adjusted indices are plotted, and the adjusted index is very close to being linear and does not exhibit a decline. These results are puzzling in the context of results from the OECD's G-7 model, where correcting for cyclical factors increases the significance of the estimated catchup coefficients. The cyclical phase of the Greek economy is short, and until more observations become available, the question of whether the observed productivity slowdown is cyclical or secular cannot be properly answered. For the moment my maintained hypothesis is one of constant technical change. A final specification for technical change was also tried, following the example of other researchers of productivity decline (e.g. J. R. Artus [7]). I introduced a quadratic term in the regression to take into account a possible productivity slowdown that was independent of cyclical influences. Such a slowdown is indeed thought to have happened 1 5 « i , the catchup coefficient, has a t-statistic of 0.84. Chapter 2. Output Determination: Background, Specifications and Results Figure 2.5: Labour Productivity Index, Actual and Cyclically Adjusted Chapter 2. Output Determination: Background, Specifications and Results 50 mainly for social and political reasons: since the mid-seventies the number of hours worked has dropped and the number of official and unofficial holidays has increased, and the "work ethic" is thought to have deteriorated. The first two effects are quantifiable, in theory at least, but I lack reliable data on them. The third effect is, of course, unquantifiable. The results of such a specification were, as expected, ambiguous. The coefficient on the quadratic term was very significant, and the resulting output and factor demand equations had a marginally better fit. What's more, the coefficient on the sales variable was smaller (0.798 as opposed to 0.918). This is because allowing the production function's labour efficiency parameter to get closer to the observed labour efficiency index ascribes less importance to cyclical variables, in this case unexpected changes in demand. In addition, the restriction of zero constant and a coefficient on In Qnv of 1 is a lot easier to accept, showing a better specification of normal output. However, there are serious problems involved with using a quadratic time trend to describe technical change: the trend becomes negative by 1978 and efficiency drops at an accelerating rate, something that makes out-of-sample simulations incoherent. In any case, the improvement in the fit of the output and factor demand equations is not statistically discernible through non-nested tests. In conclusion: there is prima facie evidence that there has been a slowdown or even decline in productivity in recent years, but I have not found a satisfactory and statistically significant way of modelling this phenomenon. Thus the assumption of constant Harrod-neutral technical progress stands, and any possible decline in long-term productivity trends will be picked up by cyclical influences. The estimated normal output, Q „ v , is plotted in logarithmic form against actual output in Figure 2.6. Chapter 2. Output Determination: Background, Specifications and Results Figure 2.6: Log Output, Actual and Nonnal-CES Chapter 2. Output Determination: Background, Specifications and Results 52 2.4 Output Equation: Estimation Results The output equation was then estimated. Three alternative specifications are reported. First, the untrended Cq variable, based on the interest-sensitive capital price, pkf, was used in conjunction with the constant technical progress assumption. Then the detrended Cq was used, again in conjunction with constant technical change. Finally, the untrended Cq was used with the quadratic technical change assumption. The fit is uniformly good,16 as shown by the summary statistics in Table 2.2. In addition to the three utilization variables, I also attempted to introduce the political dummy variable, DCypri (for 1974). Given the general mobilization and severe reduction in employment, the economy is expected to have operated above normal levels. The variable was of the right sign in every case, but was significant and was included only in the preferred case of Untrended Co-The restriction that the coefficient of Qnv be one17 is not rejected, except in the case of detrended Cq. Non-normal residuals are only revealed in the same case. The cyclical variables are significant, and the hypothesis of constant factor utilization (i.e. that the production function holds at all times, and that temporary demand and cost conditions have no effect on output), is heavily rejected in all cases, as shown by the F-test results reported in Table 2.2. Further hypothesis tests are described in Chapter 3. One problem with the results is common to all specifications: the coefficient on the sales variable is quite high, implying only a small buffering role for imports and in-ventories, and causing some instability in simulation, since output response to demand 16Time-series output equations can be expected to have high R2,s and other impressive summary statistics, and "good fit" is of course only meaningful in a comparative sense. This is the sense used here, and the comparison is both with results of the same equation for other countries (in [59]), and with results of alternative equations, as reported in the next chapter. 1 7The convention followed in Table 2.2 and everywhere else in this thesis is that restrictions on the coefficients of variables are shown as restrictions on the variables themselves: for example, In Q„ — \ means that the coefficient of In Q„ is constrained to be equal to 1. Chapter 2. Output Determination: Background, Specifications and Results 53 Dependent Variable: In Q Nested CES Cobb-Douglas Untrended Cq Trended Cq Quadratic 1.000 1.000 1.000 1.000 (restr.) (restr.) (restr.) (restr.) In 5 0.92305 0.75990 0.79808 0.91792 (24.94) (10.02) (12.26) (24.07) InC, -0.18396 -0.14625 -0.18508 (7.167) (5.419) (6.972) InC, -0.42245 (4.525) In Igap 0.02271 0.03949 0.01544 0.02281 (0.968) (1.109) (0.668) (0.938) 0.02518 0.02360 (2.70) (2.44) R2 0.9993 0.9984 0.9992 0.9992 s.e.e. 0.009325 0.013673 0.009778 0.094427 D-W 1.1761 0.5712 0.8490 1.0365 X2-test on normality 3.063 7.221 5.414 3.764 of residuals F-test on 2.259 9.756 0.458 2.991 restriction 1 F-test on 293.124 134.57 51.67 285.35 restriction 2 Restriction!: lnQnv-= 1, constant = 0 Restriction 2: In S = In Cq = In Cq = In Igap = 0 Table 2.2: Nested-CES- and CES-Cobb-Douglas-based Output Equations: Summary Statistics Chapter 2. Output Determination: Background, Specifications and Results 54 Lagsnd • 4SULSJ, 0.15-1 O 3 < -o . ioH 1 1 1 : 1 1 6 0 8 3 7 0 7 3 8 0 8 3 Year Figure 2.7: Output Growth rate, Actual and Predicted, CES changes is large. Alternative hypothesis tests, reported in the next chapter, strengthen the suspicion that the CES specification leaves something to be desired. Nevertheless, it will be shown that introducing an explicit production structure and combining it with cyclical demand and cost influences explains output better than the alternatives consid-ered. Furthermore, the CES functional form, while subject to possible misspecification, strikes the best balance between structure and flexibility for the imperfect data set and the "ossified" economy it has to deal with. On the other hand, Chow tests on the stability of the coefficients do not reveal any clear and statistically significant breaks. In the CES-CD case, parameters and output equation results are also given in Ta-ble 2.2. Dcypr^ is significant, and in general the results are not dissimilar, which is not Chapter 2. Output Determination: Background, Specifications and Results 55 surprising given that the outer function elasticity of substitution in the nested-CES case is close to 1. On balance, my preferred specification for the production function and output equa-tion is the nested-CES with constant technical change and untrended Cq. Estimated output is plotted against actual in log difference (or growth rate) form in Figure 2.7 (in the straight log form the fit is too close to show in the graph). Chapter 3 Output Determination: Alternative Hypothesis Tests In this chapter I first examine the usefulness of an exogenous capacity utilization rate for explaining output. Secondly, I compare my model of output determination against a completely demand-determined Keynesian output model. Thirdly, I test my model against completely supply-determined models. These tests effectively compare my work with existing models of the Greek macroeconomy, as described in the literature survey. Finally, I compare my results to a standard Vector AutoRegressive model as proposed by Doan, Litterman &c Sims [37]. To keep all models comparable (especially given the complications of using dummy variables in VAR) I used the CES-based output equation that excludes the political dummy variable Dcypri • 3.1 An Exogenous Utilization Rate First I tried to see if an exogenously defined capacity utilization rate, used for cyclical adjustments of inputs or output, can adequately account for departures of actual output from the production function, without recourse to endogenous cost, demand and inven-tory variables. This capacity utilization rate (derived by the Centre of Planning and Economic Research from manufacturing survey data, and based on physical measures of unused capacity) is used extensively in Greece (Koutsouvelis and Karadeloglou [66], and Avramidis et. al. [8]) in the wage, price and external trade equations, and variables like it have been used in macro models of many other countries. I used it to (multiplicatively) adjust the data series for 56 Chapter 3. Output Determination: Alternative Hypothesis Tests 57 CES prod. fa. (1) K adj. (2) K k E adj. (3) L adj. (4) CES output eq. (5) R2 0.9718 0.9717 0.9757 0.9763 0.9990 s.e.e. 0.06240 0.06247 0.05786 0.05722 0.01060 D-W 0.2569 0.2566 0.2290 0.2740 1.1622 Table 3.3: An Exogenous Utilization Rate as Alternative Hypothesis (Dependent Vari-able: InQ) 1. capital, 2. capital and energy, and 3. labour. The resulting CES production functions, that is, output equations derived under the exclusion of demand and cost effects ( In5 = \nCq = ln/ f f a p = 0 ) are reported in Table 3.3. For the purpose of comparison, they are flanked by the unadjusted CES production function and CES output equation. It is clear from the results that adjusting the capital stock (model 2) does not improve on the unadjusted, constant IFU equation (model 1), while adjusting both capital and energy (model 3) does. Adjusting labour (model 4) also improves the fit.1 All these alternative models are clearly inferior to the output equation based on variable IFU (model 5). 1In another alternative equation I estimated, the dependent variable (output) is adjusted using the same utilization rate. The fit is worse than in model 1 above, and although the results are not directly comparable, the following summary statistics are listed for illustrative purposes: R2 = 0.9703, s.e.e. = 0.06150, D-W = 0.3888. Chapter 3. Output Determination: Alternative Hypothesis Tests 58 3.2 A "Keynesian" Alternative Next I tested my models against a Keynesian, purely demand-determined alternative. In doing so I followed the methodology used by Helliwell [54] with Canadian and US data. The results are reported in Table 3.4. Most Keynesian models, including the ones for Greece, assume output is the sum of separately determined behavioural demand equations: GNP = C + I + G + X + Iinv-M = SALES + Iinv - M (3.32) To change this model into a testable form, I assume that real final sales ( SALES = C +1 + G + X ) are determined elsewhere in the model, thus giving SALES2 the status of an endogenous variable in the 2SLS estimation. Thus we can substitute behavioural equations for /,•„„ and M into the output equation, which leaves output as one possible response to demand conditions (represented by the SALES variable), along with inven-tory investment and imports. The inventory investment equation used depends (as in Koutsouvelis and Karadeloglou) on current demand (SALES) and last period's inventory stocks (iirt-„„_i): Iinv = f (SALES, K^) (3.33) with expected signs for SALES and Kinv-1 (+) and (—) respectively. Real imports mainly depend (again as in Koutsouvelis and Karadeloglou) on final sales and the relative price of imports with respect to domestic output, PREL: M = g(PREL, SALES) (3.34) with expected signs for PREL and SALES (—) and (+) respectively. 2The SALES variable is the same as the s — C + J + G-f-Xne variable defined in section 2.2 and used to define abnormal sales, 5, with the difference that energy exports are not included. Chapter 3. Output Determination: Alternative Hypothesis Tests 59 Substituting these two equations into the output identity (3.32) and imposing a log-linear form (which is only an approximation, given the linear form of the identity), we have, after taking logarithms, a Keynesian output equation to estimate: In Q = a0 + ax In SALES + a2 In Kinv_y + a3 In PREL + e (3.35) (expected signs: cti > 0, «2 < 0,0:3 > 0). The most interesting result of this estimation (model 1) is the large coefficient on the sales variable (1.04). This high coefficient implies no buffering role for inventories, unless imports rise more than proportionally with unexpected sales. Import equation estimates in this thesis, as well as by other researchers (e.g. [66], [85] and [87]) suggest they do not. There is an insignificant coefficient (of the right sign) on lagged inventory stocks. The coefficient on the relative price of imports is similarly of the right sign but insignificant. The ultimate aim is, of course, to embed both the Keynesian model and my own variable utilization models within an encompassing model of sorts, so that nested tests can be conducted. I conducted this exercise with a CES-based model, and the results are shown in Table 3.4. The next step (model 2) is to add the cost variable Cq. This (in contrast with the Helliwell results) has a significant coefficient of the right sign, and clearly improves the fit, bringing the s.e.e. closer to the one of my model. However, as the large F-statistics on the constraint on \nQn suggests, a supply-determined normal output does contain important information. In addition, we get the wrong sign on the coefficients on the inventory and relative import price variables, although they remain insignificant. When sales and inventories are re-specified to make them consistent with the factor utilization model (model 3), the sales coefficient is reduced, and the significance of Cq increases. The relative price of imports variable retains its wrong sign but becomes significant. This strengthens my suspicion that the relative price of imports (which has Chapter 3. Output Determination: Alternative Hypothesis Tests 60 Model (1) Keynesian (2) (3) (4) Factor Util. Constant -0.67744 (1.589) 0.32925 (0.7160) 0.01987 (3.398) 1.00 (restr.) 1.00 (restr.) In SALES 1.0400 (25.00) 0.94142 (20.96) \nKINV-i -0.01571 (0.548) 0.00867 (0.359) \n PREL 0.13396 (0.991) -0.17321 (1.208) -0.16235 (3.541) InS 0.77286 (14.88) 0.91814 (21.34) InC, -0.25476 (3.269) -0.41099 (6.364) -0.19814 (6.782) In Igap 0.04110 (1.795) 0.02470 (0.905) R2 s.e.e. D-W F-test 0.9980 0.015216 1.2981 21.311 0.9987 0.012184 0.8460 19.562 0.9994 0.008547 1.2731 0.1601 0.9990 0.010596 1.1622 4.0814 Equation constraints: (1) lnQ„„ = lnC, = 0.0 (2) lnQ n„ = 0.0 (3) lnQ„„ = 1.0 (4) lnQnv = 1.0,constant = In PREL = 0.0 Table 3.4: From Keynesian Model to Factor Utilization (Dependent Variable: InQ) Chapter 3. Output Determination: Alternative Hypothesis Tests 61 a strong time trend) picks up some of the time trend in the Cq variable. Thus PREL contains some (spurious) information, as shown by the value of the F-test on dropping it. The relative import price variable is then dropped to get us to my variable factor utilization model. This suggests, as in the Canadian results (Helliwell [54]), that the high coefficient on SALES in models (1) and (2) is due to the exclusion of supply-determined normal output Qnv. In an important sense model (3) encompasses Keynesian models (1) and (2), as well as my own (model 4). The very high F-statistics on the restrictions on both (1) and (2) show that the Keynesian models can be rejected relative to the encompassing model. At the same time my model has all the explanatory power of the encompassing model (given the low F-test on its constraints), except for the likely spurious effects of PREL. 3.3 The Lucas Alternative The next alternative hypothesis that I test, again using the methods of Helliwell [54], is a totally supply-determined one based on the Lucas [71] supply equation. (Alogoskoufis [3], a representative example of the supply-side-monetarist macro research on Greece, has essentially a Lucas output equation). In the Lucas supply equation, output departs from its trend value (given in its original formulation by last period's output and a time trend) in response to unexpected price changes. The important issue is, of course, how to derive a measure of expected price. In doing so I have used a method that is standard in the literature (Mishkin [77] and [78], Darrat [32], Helliwell [54]). I derived expected price as the predicted value from a regression on lagged values of key macro variables, plus forecasts of current exogenous variables (like the exchange rate, government spending, world income and prices) based on individual regressions. Then all variables used in the Chapter 3. Output Determination: Alternative Hypothesis Tests 62 estimation of expected price were included in the supply equations' list of instruments. The first model estimated was the original Lucas form, with exogenously determined trend output (model 1). The coefficient on the expected price variable, In (j^), has the expected negative sign but is not significant. The next step was to replace the exogenous variables t and In Q-\ (which are only intended as an approximation) with the consistently derived normal output \nQn (model 2). Next the sales variable In 5, incorporating the effects of unexpected aggregate demand changes, was added (model 3). The following three steps led to the encompassing model (6) that includes the Lucas supply-determined models (1) and (2) as a special case. First, variable In (JJT) was added to take into account unanticipated changes in costs (model 4). C is current unit costs (i.e. C = Cq.pq), and C e is expected unit cost, explicitly derived from the production structure (in the same way desired factor stocks are derived, but by inverting it for normal output rather than expected output): Ce represents what unit costs would be if, given current factor prices, factor proportions were adjusted to minimize costs. Then abnormally low or high profitability, measured by expected costs over expected price, can cause deviations from normal output if it is different from its normal value of 1. Thus the variable In (jpr) was included (model 5). Finally, the inventory gap, In Igap, was included as an explanatory variable (model 6). It is not difficult to notice that variables ln(jp), In (jjz) and In (rpr^j represent a decomposition of the variable In Cq, which is equal to In [j^j. Then my Variable IFU model can be derived from the encompassing model 6 by imposing the restriction that these three variables have the same coefficient (model 7). The results of the tests are reported in Table 3.5. The Lucas model, both with exogenous and endogenous normal output (models 1 and 2), is heavily rejected. Interestingly, using the endogenous measure of normal output in Chapter 3. Output Determination: Alternative Hypothesis Tests 63 (1) Lucas (2) (3) (4) (5) (6) (7) Fact. Ut. Constant 0.3606 (0.40) 0.0037 (0.27) 0.0000 (0.01) -0.0090 (9.18) -0.0081 (7.01) -0.0082 (6.85) t -0.0036 (0.77) InQ-i 0.9786 (12.63) 1.00 (restr.) 1.00 (restr.) 1.00 (restr.) 1.00 (restr.) 1.00 (restr.) 1.00 (restr.) Hpe/Pi) -0.0019 (0.00) -0.3092 (1.29) 0.1680 (2.03) 0.0723 (4.11) 0.0393 (1.40) 0.0412 (1.43) -0.1981 (6.78) ln(C/Ce) -0.8133 (20.78) -0.7597 (14.49) -0.7670 (13.86) -0.1981 (6.78) \n{C°lf) -0.0280 (1.49) -0.0253 (1.27) -0.1981 (6.78) In 5 1.0018 (13.63) 0.1707 (3.99) 0.2404 (3.84) 0.2304 (3.45) 0.9181 (21.34) In Igap -0.0056 (0.52) 0.0247 (0.91) R2 s.e.e. D-W F-test 0.9911 0.03232 1.779 2221.3 0.9713 0.06290 0.522 718.78 0.9972 0.01965 0.283 75.52 0.9999 0.00403 1.235 1.130 0.9999 0.03914 1.183 0.873 0.9999 0.00400 1.281 1.070 0.9990 0.01060 1.1622 21.81 Restrictions: (1) \nQnv = ln(C/Ce) = ln(Ce//f) = lnS = InIgap = 0.0 (2) ln<2„„ = 1.0, t = l n ^ = ln(C/Ce) = ln(Ce/f) = In 5 = l n / g a p = 0.0 (3) In Qnv = 1.0, t = In Q_x = ln(C/Ce) = lnjc 6 /^) = In Igap = 0.0 (4) lnQ„„ = 1.0,* = lnQ_! = \n(Ce/f) = \nlgap = 0.0 (5) lnQ„„ = 1.0, t = lnQ_! = In J 9 a p = 0.0 (6) lnQ„„ = 1.0, t = lnQ_! = 0.0 (7) lnQn t, = 1.0, t = lnQ_x = constant = 0.0,ln(C/Ce) = \n(Ce/f) = \n(pe/pq) Table 3.5: From the Lucas Equation to Factor Utilization (Dependent Variable: In Q) Chapter 3. Output Determination: Alternative Hypothesis Tests 64 the Lucas equation (model 2) results in a worse fit than the original equation with lagged output and a simple time trend (model 1), which points to specification problems with the production function (as does the fall in the D-W statistic). The F-statistic for model (1), however, is much higher than for model (2), showing that Q„ does carry information not contained in the exogenous specification of trend output. The unexpected price change variable, In , is everywhere insignificant or border-line insignificant, and often has the wrong (positive) sign. If unexpected price changes induce changes in output in the Lucas manner, this does not come through in these results, and the problem does not seem to lie with the specification of expected prices, since expected profitability (In (jjf)) also contains expected prices but is everywhere of the right sign. The inclusion of the sales variable, In S (model 3), dramatically improves the fit. Each successive variable inclusion also improves the fit, with the exception of the inventory gap, which performs poorly throughout. The final restriction that reconstructs Cq from its decomposed parts (i.e. that the variables In (p )^, In {^ ) and In {jpr) have the same coefficients) is rejected. The single variable Cq does not seem to contain all the infor-mation of the three separate variables. The actual over expected cost variable (In ( ^ r ) ) seems to pick up spuriously information contained in the sales variable, and seems too strongly correlated with the utilization rate. The latter is disappointing but not sur-prising, since C and Ce are negatively related to actual and normal output respectively. This again points to specification problems with the CES production function and/or the utilization variables. However, despite the possible specification problems, there are strong indications that a completely supply-determined model cannot adequately explain output, and there is no evidence that the Lucas unexpected price effect can be relied upon to explain deviations of output from trend (or normal) levels. Chapter 3. Output Determination: Alternative Hypothesis Tests 65 3.4 Is Money Neutral? The Barro Hypothesis Another "New Classical" variant that I tested as an alternative hypothesis was a reduced-form version of the unanticipated price effect (Barro [12], [13]), where deviations of output from trend depend on unanticipated changes in the money stock. Barro's results on the U.S. economy have come under increasing criticism (e.g. Mishkin [77]), but I still wanted to test this alternative hypothesis in order to confirm (or revise) my suspicion that the money stock has at best a weak and indirect link to the real sector of the Greek economy. An estimate of the anticipated high-powered money stock has to be provided in order to define the unanticipated changes. The same method of using a regression on lagged macro variables was tried as in the Lucas equation tests. The results of the Barro-type equation (Table 3.6) are so disappointing that I did not attempt to conduct proper nested tests against other models. All money growth variables, current and lagged, anticipated or not, have very insignificant coefficients, in almost every case of the wrong sign. The fit of the equation is very poor, with a simple time trend providing a better estimate of output. Introducing lagged transitory income (GNP) as suggested by Darby et. al. [31] does not change the situation: the fit improves marginally, and autocorrelation is reduced, but money stock variables remain insignificant and (in 3 out of 4 cases) of the wrong sign. The distinction between anticipated and unanticipated money growth provides no extra information, as equation 3 shows: lumping both together improves the fit by saving degrees of freedom. Money growth does not seem to have much to do with real output: both monetary variables are insignificant and of the wrong sign. In an important sense money seems to be neutral, even in the short run, and this I believe vindicates my decision not to model the monetary sector of the Greek economy. Chapter 3. Output Determination: Alternative Hypothesis Tests 66 Barro Darby (3) Constant 11.791 (219.2) -2.5761 (1.697) 11.788 (226.7) t 0.07413 (3.917) -0.00053 (0.050) 0.06548 (4.887) \nDMU -0.01057 (0.154) -0.02230 (0.879) In DMU-! 0.00164 (0.023) -0.01957 (0.742) In DMA -0.01397 (0.152) -0.02605 (0.768) In DMA-i -0.09885 (1.268) -0.00793 (0.261) lnFT-i 1.1826 (9.466) In D M -0.02394 (0.483) In DM-i -0.04704 (0.988) R2 s.e.e. D-W 0.8999 0.09936 0.2697 0.9863 0.03670 1.9224 0.9058 0.09637 0.2614 Table 3.6: The Barro and Darby Equations (Dependent Variable: In Q) Chapter 3. Output Determination: Alternative Hypothesis Tests 67 3.5 VAR: The Structure-Free Alternative Finally, another alternative hypothesis I tested was a structure-free Vector AutoRegres-sive (VAR) model, using a second-order autoregressive error structure and instrumental variables. I used the list of variables proposed by Doan, Litterman &: Sims [37], with the exception of non-monetary government liabilities, on which I have no data. I tested a VAR model not because I think a structure-free model is appropriate for my purposes (which are to try to explain the recent economic history of Greece by gaining some in-sights into the structure of its economy, not to use a black box model for predictions), but in order to see whether a structure-free model can predict output better than the one developed in the previous chapter. The results of the estimated VAR model, my own (IFU) and a model that encompasses both are given in Table 3.7. The VAR model has a lower standard error than my models, and it is clear from the nested tests that both the VAR model and my own variable utilization model contain some information that the other lacks. The F-tests on the relevant restrictions show that my utilization variables are a lot harder to reject, however. Chapter 3. Output Determination: Alternative Hypothesis Tests 68 VAR Encomp. IFU Constant 0.69256 (0.854) -0.20502 (0.205) InQ-i 0.72248 (5.554) -0.40653 (1.923) lnp, -1.3102 (9.135) -0.43992 (1.513) lnr -0.00101 (1.034) 0.07000 (2.370) In RNU 0.05842 (2.326) -0.04786 (2.807) In pi 1.2381 (8.530) 0.35895 (1.384) In HPM -0.17254 (6.094) -0.06717 (1.548) l n M n e 0.27896 (2.882) 0.04550 (1.083) \nXne 0.26383 (4.508) 0.01964 (0.472) In TOT 0.97418 (6.520) 0.02673 (0.182) mB -0.00181 (9.308) -0.00006 (0.286) \nYw 29.161 (1.427) 1.1741 (0.167) lnpw -30.915 (1.538) -0.54674 (0.077) InQ™ 1.0433 (4.875) 1.00000 (restr.) In 5 -0.08787 (0.272) 0.91138 (21.34) lnCq -0.46795 (2.844) -0.19814 (6.782) hi Igap 0.06230 (1.815) 0.02470 (0.905) R2 0.9992 0.9998 0.9990 s.e.e. 0.005816 0.004309 - 0.010596 D-W 3.028 2.585 1.162 F-test 23.550 7.4897 Constrain ,s: (l)lnQ„„ = In S = In Cq = In Igap = 0 (2) Unconstrained (3) \nQnv = 1, Constant = In Q-\ = \npq = lnr = In RNU = \npe = = In HPM = In Mne = In Xne = lnTOT = lni? = lnF„, = lnpw = 0 Table 3.7: Unstructured VAR versus Factor Utilization (Dependent Variable: \nQ) Chapter 4 Derived Factor Demands One of the key elements of this model is the attempt to estimate factor demand equations based on consistently derived estimates of optimal factor levels. Such optimal factor levels were derived by inverting the production function to find the cost-minimizing factor proportions for producing some measure of expected output, Q*. 4.1 Derivation and Specification In the CES and CES-Cobb-Douglas cases the production function can be easily inverted for optimal levels of capital stock and employment (energy demand is directly estimated from the vintage structure: the parameters of the inner function are derived by minimiz-ing its standard error). For nested-CES, (4.36) where (4.37) and T (4.38) 69 Chapter 4. Derived Factor Demands 70 In the CES-CD case, optimal capital stock is K* = ot0Q*P^ (<*?<)/(*•)  ^(P'Pk1-* + 7 ° 'pe 1 - < r )^ r ( l - « ) and (4.40) with Kev given by equation 2.21. Results based on the CES-CD formulation are not reported, since they are very similar to the results of the nested CES. It now remains to specify the mechanism through which expected output for factor demands, Q*, is determined. Investment decisions, in particular, have to be forward-looking, with investment designed to fill the gap between the actual capital stock and desired capital stock, capable of producing, at minimum cost, the target output level at the end of the time horizon. Since all three factors of production are assumed to be quasi-fixed, forward-looking target or expected output should also enter their demands. It is arguable whether the same expectations formation process will do for all three, but I did not want to complicate my model by specifying different ones. Ideally, such a process should depend on past output as well as past and present aggregate demand, which is itself endogenous. In the MACE models the parameters of this equation are fixed as the ones that maximize the likelihood of the actual factor demand equations through a grid-search. In the case of Greece, however, I was reluctant, at least initially, to specify a complicated expectations-formation process. More sophisticated adaptive expectations did nothing to improve the fit of factor demand equations. As for consistent or rational expectations, the extra computational burden involved in generating them was out of proportion to the sophistication of the rest of my model. As a first approximation I opted for expectations based on past output, which had to be extrapolated forward —cr (4.39) Chapter 4. Derived Factor Demands 71 using past output growth, given the strong, almost uninterrupted growth rates of the Greek economy for most of the time period. Thus Q* is denned as: = (1 +1.5 ) q - ^ g - a (4.41) where Qav is a three-period moving average (current plus two lags) of output growth rates. Qav is included as the extrapolation factor, and the weight (1.5) was chosen to be consistent with growth rates over the sample. The implicit time horizon is two years, with one-year revisions. I then attempted to include aggregate demand and the utilization rate in the spec-ification of target or expected output for factor demands. The inclusion of aggregate demand, which can only be defined as output minus unintended inventory accumula-tion, naturally makes expected output vulnerable to the suspected data and specification problems that made my inventory gap variable such an inconclusive determinant of out-put. In addition, inventory investment is of course defined as a residual, which makes its behaviour in simulation a lot more unstable than other variables. However, I did use an alternative equation that included aggregate demand, Qa, defined as Qa = Q~ (Iinv - Q(Fgap ~ 1)) (4.42) where i is the coefficient on the inventory gap variable in the output equation and /,„„ is inventory investment. The expression in brackets represents planned inventory invest-ment. Expected output then is, as in MACE: with an implicit time-horizon of two years.1 1 Helliwell and Glorieux [57] show the interaction of forward-looking expectations with a fixed time-horizon defined by the building and installation time of fixed capital. Chapter 4. Derived Factor Demands 72 I also experimented with including the utilization rate, Q/Qnv, in the specification of Q*. However, the results of both experiments were not very satisfactory; derived factor demand equations were much worse (in terms of fit) than the ones derived from simple extrapolative output expectations (equation 4.41). 4.2 Empirical results Once the desired factor levels were derived, I embedded them into two kinds of factor de-mand equations per factor of production. Under the assumption of quasi-fixed factors, the desired levels form targets that producers try to reach by partially adjusting their factor stocks in each time period. Quasi-fixity of capital, and investment as partial adjustment to the optimal capital stock, are assumptions long used in the literature and empirical research. In fact the explanation of investment as partial adjustment to a desired capital stock goes as far back as the accelerator models (e.g. Hicks [61]). Dale Jorgenson's work ([62]) renewed emphasis on and interest in the specification of the desired capital stock as a solution to an optimization problem, but Jorgenson himself recognizes that early neoclassical work had dealt with it, and cites his indebtedness to Irving Fisher. The existence and importance of adjustment lags is shown very persuasively in Clark [28]. The arguments in favour of quasi-fixity of capital are, of course, less applicable to labour. However, a large and growing literature on explicit and implicit contracts2 argues that labour contracts codify long-term relationships between employers and employees, and implicitly include both insurance against risk and financial intermediation by the firm on behalf of the workers. This can lead not only to (observed) non-market-clearing wages, but also to less fluctuation in employment than warranted by fluctuations in labour demand, given relatively smooth wages. In addition to this, the human capital 2The original papers are by Costas Azariadis [9] and Martin Baily [10], and representative papers discussing inefficient contracts are Calvo and Phelps [23], Hall [52] and Grossman and Hart [48]. Chapter 4. Derived Factor Demands 73 embodied in workers makes large fluctuations in employment in response to demand conditions costly to the firm. Consequently labour (especially in the sense of employment, as opposed to person-hours worked) should also be treated as a quasi-fixed input. In the case of Greece there are added structural and institutional reasons to treat labour as a quasi-fixed input: a chronic shortage of skilled workers (due to some extent to the recent phenomenon of cultural aversion to manual labour), the high cost of training, and, most importantly, serious legal limitations to layoffs and dismissals along with very generous severance payments. The latter appears to be an attempt by governments to impose the costs of unemployment on the employers, while keeping unemployment and welfare benefits very low by Western standards. Energy is, naturally, the least fixed of the three factor inputs, and the assumption of quasi-fixity is least tenable in its case. However, as long as capital is quasi-fixed, and energy is applied to capital that cannot be changed at will to use more or less energy, energy is no longer a fully flexible input. With the CES functional form as specified above in Chapter 2, energy demand incorporates explicit vintage capital effects, and the energy demand equation is estimated while fixing the inner function parameters. With all three inputs the quasi-fixity assumption is supported by the empirical results, and the estimated speeds of adjustment are very slow by Western standards. This is a result that confirms long-standing assumptions and long-accepted stylized facts about the slow response to market signals in Greece and the institutional constraints that cause it. The speed of adjustment should ideally be a solution to an optimization problem on the part of firms, and as such should depend on explicitly defined internal and external costs of adjustment.3 Such an approach lies beyond the scope of this thesis, however, and given the paucity of reliable detailed data at the economy level (or even at a more 3Early theoretical examples of the use of internal adjustment costs in optimal control models of investment are Lucas [70], Gould [47], and Treadway [99]. Chapter 4. Derived Factor Demands 74 disaggregated level) I had to be content with broad proxies for costs of adjustment. The most important such variable was unit costs as a percentage of output price, Cq. In the investment equation, Cq plays the added roles of representing profitability, and being related to the inverse of Tobin's q (to the extent that current profitability is taken as a proxy for the present value of expected future profits). There is a growing literature (Yoshikawa [103], Hayashi [53] and Abel [1]) that shows that Tobin's q, or equivalently Cq, should influence the rate of investment if there are adjustment costs. In the case of Greece, the important role of profitability in investment demand has been confirmed empirically in every macro model mentioned in the literature survey. Given the country's imperfect capital markets, this is usually attributed to the availability of self-financing from retained profits. The high measured debt-equity ratios (around 70%) contradict this argument, however. Cq was only significant in the investment demand equations, a result that should not be surprising, given the OECD results in Helliwell et. al. [59]. As an alternative to C g , I also used a direct measure of gross profitability, defined as (residual) gross profits divided by the nominal capital stock: PROF = QP<-L»eP<-Epe ( 4 4 4 ) Kne Pke The price of capital I used was pkc, the one based on a constant (interest-invariant) real supply price of capital. The reason for this was that if I had used the interest-sensitive price, profitability would fluctuate dramatically with the interest rate. This measure of profitability is, then, unrelated to the interest rate, and the fact that it provides a better fit than Cq suggests that interest rates do not play an important role in investment decisions. The rate of capacity utilization was also included as a variable that would possibly influence the speed of adjustment. This variable was never significant, and results that include it are not reported. Chapter 4. Derived Factor Demands 75 For each factor demand equation two types of model were tested: a conventional partial adjustment model and an error correction model. The latter model is of course a special case of partial adjustment, in log-difference form and with particular restrictions on the adjustment process. As such it can serve as a test of how much of the explanatory power of the partial adjustment model is due to spurious correlation due to the underlying time trends. The investment equation in its partial adjustment form has investment as a fraction of the capital stock as the dependent variable; otherwise it is linear in form. The equation for labour is log-linear; the very strong, almost uninterrupted positive trends in both put a linear form out of the question. The general forms of the estimated investment and labour demand equations were as follows: 1. Investment (a) Partial adjustment: •"•nc V - t t n e / -1 "-ne V n (b) Error correction: + ftln^ + ftln^+e (4.46) 2. Labour Demand (a) Partial adjustment: l n l n e = ft + ftlnlne_1 + ftlnI* + ftmC, + ftln^f-) +e (4.47) Chapter 4. Derived Factor Demands 76 (b) Error correction: in (4.48) The investment and labour demand equation results are reported in Tables 4.8 to 4.11 respectively. The performance of each of the 4 estimated equations is illustrated in Figures 4.8 to 4.11. The most striking feature of all estimated equations is the very slow speeds of adjust-ment of all three factors of production to their desired levels. This is also true of labour, a supposedly variable factor. This of course is only the beginning: slow speeds of adjust-ment to disequilibrium forces are apparent in all my results. Perhaps the characterization of the Greek economy as "ossified'1 is not far off the mark. Before I go on to report the empirical results of the equations described above, let me add that I also tried another approach: equations in which each factor demand depends on the deviation from the optimal level not only of itself, but the other two factors. The theory behind this approach is summarized in Nadiri and Rosen [80] and Mohr [79]. The results, however, were very disappointing (in terms of coefficient significance, overall econometric fit, and expected signs) whether internal costs of adjustment were used or not, and I shall not report them. 4.2.1 Investment In the partial adjustment investment equation results (Table 4.8) all coefficients have the right sign and are significant at the 5% level. Moreover, the most important variable (in the sense of the contribution of my model), (K* — Kne)/Kne, has a significant coefficient. However, it is small (compared, say, to the same coefficient on Canadian investment in MACE), while the lagged investment term has a large coefficient. This is something that Chapter 4. Derived Factor Demands Figure 4.8: Investment Equation (Partial Adjustment) Chapter 4. Derived Factor Demands 78 creates serious stability problems in simulation, especially in the case of shocks that affect investment through a drop in profitability: the effects of such shocks tend to accumulate, and a change in desired capital does not turn the situation around for many years. The gross profitability variable performs much better than Cq. The performance of model 2 is shown in Figure 4.8. Two dummy variables (that proved statistically significant and invariant to model specification) are added to take into account two major political events that undoubtedly produced serious dislocations in investment plans. The first, Dai, represents the political instability that preceded and followed the 1967 coup of the colonels. The second, Dcypr, tries to capture the effects of the 1974 Cyprus crisis, that caused a general mobilization which lasted well into 1975, precipitated the fall of the military regime and started a period of political uncertainty and economic dislocation. Dcoi takes a value of 1 in 1967, and Dcypr takes a value of 1 in 1974 and 1975 (they are both zero elsewhere). Thus Dcypr represents the consolidation (for the purpose of saving degrees of freedom) of two separate dummies for 1974 and 1975 that had statistically identical coefficients. This can be explained if one takes into account the fact that general mobilization lasted an approximately equal number of months in the two years. The error-correction results are reported in Table 4.9 and the fit of the preferred equation (model 2) depicted in Figure 4.9. Of the three error-correction variables, the coefficient on the integral adjustment term (ln(KZi/Kne^)) w a s n e v e r either significant or of the right sign, so it is omitted. The other two variables (ln(K*/Kli) and ln(iT* 1 /I„e_i)) all have the right signs, but are often insignificant. This is disappointing but not surprising, given the log-difference form and the long-run properties of the specification.4 The rest of the variables have significant 4 Helliwell et. al. [59] estimate identical investment equations for the G-7 countries, and their results are very similar: the integral adjustment term is wrongly signed in every case and is dropped, and the size and significance of the other coefficients is very close to mine. Chapter 4. Derived Factor Demands Figure 4.9: Investment Equation (Error Correction) Chapter 4. Derived Factor Demands 80 and correctly signed coefficients. Once again, the dummy variables prove significant and invariant to specification, and the preferred model is shown to be model 2 (with profitability instead of Cq). As an alternative hypothesis I also used a measure of desired capital that is unrelated to any production structure but only to the level of output. I derived an average capital-output ratio by regressing Kne/Q on a time trend (this was to take into account the growing capital intensity of production). By multiplying by Q, I then defined K*, the output-based desired capital stock, and used it in partial adjustment and error-correction equations. The results are reported in Tables 4.8 and 4.9 (models 2 and 4). In all but one case, every variable containing K* is either insignificant or wrongly signed or both, and the equations have a worse fit in every case than the ones using K*, the cost-minimizing desired capital stock. 4.2.2 Labour The results of the labour demand equations are reported in Tables 4.10 and 4.11. In the partial adjustment model (Table 4.10-model 1, Figure 4.10) all coefficients have the right sign and are significant. Variables Cq and IFU were both insignificant and were dropped. The political dummy variable A ^ p r , which takes the value 1 in years 1974 and 1975, tries to take into account the effect on labour supply of the general mobilization in the two years. Again it combines two statistically identical dummies, one for each year, for the same reasons given above. The coefficient on desired employment implies a 10% adjustment to the gap between actual and desired employment per year. Such a speed of adjustment appears very low for a supposedly variable factor of production, and this is confirmed by comparison to G-7 results in the error-correction case. One puzzling result is that the constraint that the coefficients on Lnc_x and L* sum to one is rejected. Chapter 4. Derived Factor Demands 81 CES Q-BASED (1) (2) (3) (4) Constant 0.06102 (2.01) 0.00984 (1.31) 0.07510 (2.28) -0.00109 (0.15) Ir»e_i / Kne_ - i 0.88706 (11.31) 0.73850 (7.34) 0.93393 (11.19) 0.80158 (7.14) (K* - Kni d/Kne 0.03585 (1.63) 0.04533 (2.73) (Kg* — Km .)/Kne 0.01829 (0.58) 0.04301 (1.64) Cg -0.04851 (1.92) -0.06713 (2.74) PROF 0.01896 (2.70) 0.02322 (3.09) Dcol -0.02068 (3.98) -0.01922 (4.05) -0.01991 (3.49) -0.01701 (3.09) Dcypr -0.02817 (7.57) -0.02604 (7.86) -0.02784 (6.82) -0.02418 (6.51) R2 s.e.e. Durbin-fe 0.9659 0.004788 0.58967 0.9723 0.004313 0.93052 0.9632 0.004976 0.71484 0.9660 0.004785 1.1864 Table 4.8: Investment Equation, Partial Adjustment (Dependent Variable: Ine/Kne) Chapter 4. Derived Factor Demands CES Q-BASED (1) (2) (3) (4) Constant -0.16352 (0.54) -0.39647 (1.50) 0.23686 (0.95) -0.25088 (0.95) \n{K*IKU) 0.45510 (0.81) 0.30217 (0.61) -1.5160 (1.47) -1.8831 (1.93) HKIJI^) 0.09183 (0.74) 0.20754 (1.84) \n{K*qJIne_x) -0.01336 (0.13) 0.23528 (1.80) InC, -0.81018 (2.62) -1.6139 (3.17) In PROF 0.21703 (3.46) 0.45194 (2.85) Dcol -0.15622 (2.60) -0.15616 (2.89) -0.19683 (2.77) -0.19350 (2.85) Dcypr -0.24248 (4.24) -0.23295 (4.98) -0.39948 (4.023) -0.38087 (4.56) R2 s.e.e. D-W 0.7464 0.055599 1.2607 0.8036 0.049921 1.4622 0.7012 0.061578 1.8906 0.7262 0.058946 2.2652 Table 4.9: Investment Equation, Error Correction (Dependent Variable: ln(J n e// r Chapter 4. Derived Factor Demands Lagand O 7.3-I 1 , , , 6 5 7 0 7 5 8 0 8 5 Year Figure 4.10: Labour Demand Equation (Partial Adjustment) Chapter 4. Derived Factor Demands 84 In the error-correction model (Table 4.11-model 1, Figure 4.11) again all variables have coefficients that are significant and of the right sign. However, the R2 is very low. The coefficients on the two adjustment variables is substantially lower than the same coefficients in every country in the OECD G-7 model, implying a slower speed of adjustment. Labour demand in Greek macro models is almost always defined as a linear function of output (as in Koutsouvelis & Karadeloglou [66], Garganas [44], Tsoris [100], Avramidis [8] and many more). I decided to treat this approach (the "output approach") as an alter-native hypothesis to the one I used, in which quasi-fixed employment responds partially to changing cost-minimizing factor proportions (the "cost" approach). I first derived an average labour-output ratio by regressing L/Q on a time trend, since labour productivity grew very strongly over most of the period. I then multiplied this rate by the level of out-put to derive an alternative, output-based definition of desired employment, which I call L*. Next I estimated 3 alternative equations using L*. First, a simple log-linear equation with only lnX* and A ^ p r as explanatory variables had an s.e.e. 12 times higher than the cost-based partial-adjustment models, plus very strong evidence of autocorrelation (model 2, Table 4.10). Secondly, a partial adjustment equation with L* instead of L* was estimated (model 3, Table 4.10). Again the summary statistics (s.e.e., t-statistics and Durbin-/i) show a worse fit than the one of the cost approach, although the constraint that coefficients add to 1 does pass the F-test. Finally, I estimated an error-correction equation with L*q instead of L* (model 2, Table 4.11). In this case too, the performance of the output-based equation according to all summary statistics is worse than that of the cost-based equation. Additionally, the coefficient on the first error-correction variable is insignificant. These results point towards the conclusion that labour is a quasi-fixed input. The hypothesis that output by itself, without even adjustment lags, determines employment, Chapter 4. Derived Factor Demands Figure 4.11: Labour Demand Equation (Error Correction) Chapter 4. Derived Factor Demands CES Q Q-PARTIAL Constant 0.36505 (3.52) 2.6582 (1.93) 0.02122 (9.12) In L n e _ i 0.84838 (26.01) 0.94517 (51.0) In I* 0.10600 (4.09) 0.65056 (3.60) 0.05483 (2.96) Dcypr -0.02610 (3.62) -0.06147 (0.74) -0.01725 (1.94) R2 s.e.e. Durbin-/i D-W F-test 0.9966 0.007966 -0.42829 11.118 0.3633 0.10844 0.1645 0.9950 0.096233 0.39997 0.382 Restriction (model 1): lnXne_! - f l n l r * = 1.0 Restriction (model 3): lnX„e_1 + lnL* = 1.0 Table 4.10: Labour Demand, Partial Adjustment (Dependent Variable: lnLne) Chapter 4. Derived Factor Demands 87 CES Q-BASED Constant 0.01848 (6.86) 0.02182 (8.66) 0.13290 (2.91) \n{L*ILne_,) 0.07088 (2-24) HL'q/LqJ 0.02120 (0.40) ln(Z;/A*-i) 0.06426 (2.69) Dcypr -0.02163 (2.45) -0.01725 (1.99) R2 s.e.e. D-W 0.2908 0.009782 1.7123 0.2901 0.009786 1.7338 Table 4.11: Labour Demand, Error Correction (Dependent Variable: l n ( L n e / £ „ e _ i ) ) Chapter 4. Derived Factor Demands 88 is clearly not credible. Even when an output-based desired labour measure is embedded within a lagged adjustment framework, the resulting equations do not outperform the "cost" approach. Output, without any reference to cost-minimizing factor proportions (that are sensitive to factor prices), cannot adequately explain employment. In all fair-ness, the use of such simple specifications for labour demand may have been due to the lack of reliable labour data.5 The fact that employment estimates were derived by the CPER on the basis of output data completed the vicious circle. As I mentioned already, in the missing years I use OECD labour force and employment estimates that were de-rived not from output, but from labour force interpolations and adjusted unemployment data. Thus circularity is avoided and, I believe, employment estimates at least follow the general movements of the actual employment variable. Of course the assumption of an exogenous, steadily growing labour force with employment as a residual reduces the cyclical sensitivity of employment figures. This is not, however, as serious a problem as it seems, since during the period in question all variables grew remarkably steadily and there is no clear evidence of cyclical behaviour. The problem of using possibly counter-cyclical estimates rather than direct obser-vations in part of the sample becomes apparent when I estimate the error-correction equation. Since this tries to explain growth rates rather than absolute levels, the R2 is very low. Furthermore, performance is markedly worse in the earlier years, as is apparent from Figure 4.11. Year 1981 also seems problematic, and the reason for the unexplained jump may be a discontinuity in the data (and a footnote in the OECD's labour statistics indicates there is one). The source for the discontinuity must have continued after 1981, but it is not captured by a significant dummy variable for the years in question. Including an arbitrary dummy variable for 1981 only does improve the fit but leaves the coefficients 5For a detailed discussion of data sources, problems and methods for labour as well as all other variables, see Appendix A. Chapter 4. Derived Factor Demands 89 unaffected. Since the nature of the problem is unknown, I decided to leave it alone. Chapter 5 Imports, Exports and Consumption In this chapter I shall specify the import, export and consumption equations, and report the estimation results of the preferred and alternative equations. 5.1 Imports Imports in my model are defined as real non-energy imports of goods and services. Given the fact that energy imports are directly derived from energy demand and domestic energy supply, they are not included in my measure of imports.1 I experimented with separating goods from services: the service account, especially tourism and the expenses of students abroad, has grown rapidly and for reasons that are not necessarily the same as those governing merchandise imports. The split did not provide any additional relevant information, and since policy analysis at such detailed level is not my objective, I decided to model total imports of goods and services. The price of imports of course includes duty rates. Import determination is closely connected with output determination: imports repre-sent an alternative response to unforeseen changes in total demand (total demand being denned as SALES = C + I + G + X), along with output and inventories. As such, imports belong in the supply block. The decision to report on them here is only due to reasons of space and clarity. The basic assumption underlying my modelling of imports, as used by J. F. Helliwell 1 Energy imports grew in value from approximately 6% of total imports in 1963 to 23% in 1984. 90 Chapter 5. Imports, Exports and Consumption 91 (and explained in Helliwell and Chung [56]), is that non-energy imports and normal output enter a long-run CES aggregate utility function as substitutes for each other. Then, if <f> is the elasticity of substitution between imports and domestic output, normal or permanent imports are given by the relationship Mnep = QniPmnelPa)* (5-49) where the prices involved are actual or, more appropriately, expected. Thus a three-period moving average of relative prices, can be used as a proxy for expected relative prices. Such a lagged variable may also reflect adjustment processes; it is, however, almost impossible to disentangle the expectations from the lagged adjustment effects. Actual imports may differ from normal ones because of a potential buffering role of imports when there are discrepancies between actual and normal operating rates, and inventories. Thus, if the output and import buffering responses were symmetric, then we would have the following import equation, similar to (2.2): In Mne = In Mnep + f3Ms In S + pMC In Cq + @MI In Igap (5.51) or, since M „ e p is not observable, by substituting (5.49) into (5.51), In Mne = lnQn + <f>]n(pmne/pq) + fiMS In S + /3MC In Cq + /3MI In Igap (5.52) Significant estimates of the coefficients 0MS, PMC and 0MI would imply a short-term buffering role for imports. Both current (PREL) and moving-average (PREL) relative prices were used, with PREL resulting in a somewhat better fit but not many other changes. I also attempted to include two dummy variables, D73 and D80, taking the Chapter 5. Imports, Exports and Consumption (1) (2) Constant -1.1382 (33.49) -1.1248 (34.57) lnQ„„ 1.00 (restr.) 1.00 (restr.) In S 0.14968 (0.52) 0.30166 (1.30) InC, -0.58130 (1.48) -0.55551 (1.65) In Igap -0.30133 (2.07) -0.34584 (2.86) In PREL -1.114 (4.09) In PREL -1.0605 (4.67) D73 0.26854 (5.77) 0.22155 (5.10) R2 s.e.e. D-W F-test 0.9934 0.03507 1.8829 0.5273 0.9945 0.03199 1.9940 0.0576 Table 5.12: Imports as Buffers (Dependent Variable: lnM n e) Chapter 5. Imports, Exports and Consumption 93 value of one in years 1973 and 1980 respectively. D7Z represents a speculative surge in imports due to the political situation. D80 in turn captures the effects of strictly temporary quotas and administrative restrictions on imports that were lifted within the year, as soon as the perceived BoP crisis was over. The results are reported in Table 5.12, and are inconclusive. The utilization variables In 5 and In Cq have the right signs but are not significant, while Igap has the wrong sign but is significant. Given the repeated problems with the inventory variable in the supply block, the latter is not surprising. On the other hand, the relative price variable is correctly signed and very significant. Of the two dummies only D73 is significant, a result that is derived repeatedly regardless of specification. Thus D80, despite the a priori reasons for its inclusion is dropped from all reported equations. The restriction that the coefficient on Q„„ be 1 was easily accepted. Thus the attempt to explicitly model the role of imports as a buffer to unexpected demand changes was not really successful. If we stay within the basic framework of Equation (5.49) (desired imports as long-run substitutes to domestic output) the problem of short-run deviations of actual imports from normal imports remains. To some extent these deviations may be captured by using PREL, which allows room for adjustment lags. I also attempted to use the intensity of factor utilisation, Q/Qnv as an additional variable for explaining these short-term deviations. This entailed using both In Q and In Q„„ and testing various restrictions on their coefficients. The F-tests on each restriction are reported in Table 5.13, and summary statistics on alternative equations in Table 5.14. The restriction that is most easily accepted is lnQnv — 1.0, with unrestricted InQ. There is no doubt that both variables carry some information, but such information is probably spurious. Using both In Q and In Q„„, with or without any restrictions, apart from problems of long-run equilibrium, causes the relative price coefficient to become insignificant and of the wrong sign (Table 5.14, equation 1). On the other hand, of all the joint restrictions tested, InQ = 1.0 and lnQ„„ = 0.0 Chapter 5. Imports, Exports and Consumption 94 Restrictions F-Stat. lnQ = 1.0 7.977 lnQ„ = 1.0 0.006 InQ = 0.0 9.239 lnQ„ = 0.0 8.918 lnQ = 1.0,lnQ„„ = 0.0 4.535 InQ = 0.0,lnQn„ = 1.0 22.277 lnQ = 1.0,lnQ„„ = 1.0 78.778 lnQ = 0.0,lnQn„ = 0.0 14.458 Table 5.13: F-Tests on Alternative Restrictions is the most easily accepted, and the resulting equation gives a correctly signed and very significant coefficient on the relative price variable (Table 5.14, eqn. 3). Imposing the restriction lnQ n = 1.0,InQ = 0 (eqn. 2) results in an equation with a markedly worse fit. Thus the preferred equation (model 3) is not related to the production structure directly, since it incorporates neither normal output nor capacity utilization (or its de-terminants). Imports, however, are driven by domestic output rather than absorption or aggregate demand, and the import equation remains within the framework described in the beginning of the section. The restriction that In Q = 1.0, when tested by itself (and in the absence of In Q„v) in this equation is very easily accepted with an F-statistic of 0.1. Another alternative hypothesis against which I test my model of import determination is the conventional one used, among others, by Koutsouvelis and Karadeloglou [66] and Prodromidis and Anastassakou [87]; namely, that imports are a function of total domestic demand (AD = C + I + G + /,•„„) and relative prices. The summary statistics are given again in Table 5.14 under eqn. (4). This equation appears to have a slightly better fit than the output-based equation. The significance of the relative price coefficient is a lot er 5. Imports, Exports and Consumption CES (1) (2) Output (3) AD (4) Constant. -8.0816 (3.23) -1.1712 (55.86) -1.1624 (88.99) -4.5055 (3.00) InQn 1.0207 (3.18) 1.00 (restr.) InQ 0.52130 (3.25) 1.00 (restr.) In AD 1.2241 (10.71) In PREL 0.81142 (1.42) -0.71336 (6.02) -0.77051 (10.43) -0.64568 (2.28) D73 0.12941 (3.56) 0.23599 (3.72) 0.13034 (3.30) 0.09289 (2.07) R2 s.e.e. D-W 0.9945 0.03203 1.3412 0.9795 0.06182 0.5545 0.9920 0.03851 0.8989 0.9923 0.03793 1.0127 Table 5.14: Alternative Import Equations (Dependent Variable: lnAfne) Chapter 5. Imports, Exports and Consumption 96 Figure 5.12: Import Demand Equation, Output-Based Chapter 5. Imports, Exports and Consumption Figure 5.13: Import Demand Equation, Aggregate Demand-Based Chapter 5. Imports, Exports and Consumption 98 lower. The two alternative import equations (output- and aggregate demand-based) are illustrated in Figures 5.12 and 5.13. There appears to be little difference between the two, except the AD-based equation exhibits more fluctuations. 5.2 Exports Exports in this section are defined as real total exports of goods and services, again excluding enegy . Exports of services were not modelled separately for the same reasons as in the case of imports. My modelling of exports tries to combine supply and demand influences in a single equation. Demand enters through world income and the relative price of exports. In this sense my model represents less of a departure from previous attempts to model Greek exports. The two main international trade models that I compare my results to ([66] and [87]) also use world income and the relative price of exports. World income (defined as the IMF index of total world income) has a very large and significant coefficient, indicating sensitivity of the Greek trade balance to cyclical fluctuations in the rest of the world. I also tried instead using the OECD countries' income index, and the IMF's industrial country income index. Both variables were less significant, which is not surprising, considering the fact that Greece does export size-able amounts to non-OECD and non-industrial (Eastern European and Middle Eastern) countries. I tried three variants of the relative price of exports, PRELX = (pxne/Pwx), where p w x is the world price of (non-energy) goods exports, expressed in terms of the domestic currency (i.e., p w x = pwxgpfx, where p w x g is the IMF unit value index of non-energy world exports in US$, and pfx the dollar exchange rate). The first was current PRELX, the second included current PRELX and its value lagged one period, and the third was Chapter 5. Imports, Exports and Consumption 99 a three-period moving average: p-RELx = I (hss. + £=£=1 + (5.53) 3 \Pwx Pwx—i Pwx—2/ Of the three variants, the three period average (PRELX) was more significant and im-proved econometric fit, pointing towards the existence of adjustment lags. On the supply side, two other variables were included. The first, used among others by Koutsouvelis &: Karadeloglou, is the "profitability of exports", defined as the price of exports relative to domestic output price. The justification is that if exporting output is more lucrative than selling it domestically, producers seek foreign buyers. As Cardoso and Dornbusch [25] point out in a paper on Brazilian trade, this variable used by itself is consistent with the assumption that a country is a pure price taker, in which case the relative price of exports should be constant and irrelevant to export determination. I use both "profitability" and relative price, both separately and together, and compare the results. Another supply-side variable that I use to explain exports (and this is a departure from existing results for Greece) is the inventory gap as a measure of domestic excess demand or supply. The inventory gap is defined, as in the supply block, as the ratio of desired to actual inventory stocks, and has a mean of 1. Given the definition of Igap, its expected sign is negative. The justification for using this variable is the assumption that when inventory stocks are higher than desired, domestic producers seek foreign markets for their output, and vice versa. In the case of Greece this tendency is intensified by government policy of export promotion at times of overproduction or slack domestic demand. Some cyclical excess demand measure is often used to explain exports: Cardoso and Dornbusch use a simple deviation from trend, Helliwell [60] use the same measure of the inventory gap that I use, and Koutsouvelis & Karadeloglou [66] use, without success, Chapter 5. Imports, Exports and Consumption 100 legend • ICtUAl I •6 « Q. 6 6 6 8 7 0 7 2 7 4 7 6 7 8 8 0 8 2 8 4 Year Figure 5.14: Export Equation, w. Igap and Rel. Price of Exports the CPER exogenous measure of industrial capacity utilisation. I also attempted to use the endogenous measure of capacity utilization, Q/Q„v, hut it was not statistically significant. In all I report the summary statistics for six equations in Table 5.15. Equation 1 is the encompassing model that includes both relative price and "profitability" variables and the inventory gap, and all the rest are derived by imposing restrictions on it. Equations 1, 2 and 3 include Igap while equations 4, 5 and 6 do not. Given the log-linear specification of the equation, the coefficient on the inventory gap directly represents the elasticity of response: a 1% increase in the inventory gap will reduce exports by 0.57% to 0.94%. The estimated coefficient of the inventory gap Chapter 5. Imports, Exports and Consumption 101 (1) (2) (3) (4) (5) (6) Constant 4.1653 (1.11) 3.0392 (1.80) -2.2420 (2.32) -7.7082 (2.26) -1.9741 (0.74) -3.4923 (3.21) inr„, 1.4286 (1.60) 1.6961 (4.16) 2.9594 (13.42) 4.2586 (5.24) 2.9084 (4.54) 3.2447 (13.07) In PRELX -3.3860 (1.78) -2.8788 (2.54) 2.4666 (1.34) 0.50339 (0.28) ln(Pxne/Pcj) -0.25866 (0.34) 0.63493 (1.08) 1.9521 (2.44) 1.4105 (2.15) In Igap -0.93971 (3.83) -0.87572 (5.77) -0.57004 (3.02) R2 s.e.e. D-W F-test 0.9908 0.05023 2.1169 0.9916 0.04802 2.0769 0.1168 0.9909 0.05786 1.8903 3.1777 0.9756 0.08199 1.3026 14.660 0.9718 0.08807 0.7186 15.289 0.9860 0.07207 1.1938 9.7236 Restrictions: (2) ln(pxne/Pa) = 0.0 (3) In PRELX = 0.0 (4) l n / a a p = 0.0 (5) In Igap = ln(pxne/pq) = 0.0 (6) l n / g o p = \nPRELx = 0.0 Table 5.15: Export Equations (Dependent Variable: lnX„ e) Chapter 5. Imports, Exports and Consumption 102 variable seems quite robust, and constraining it to a lower value, or dropping the variable altogether (as in equations 4, 5 and 6), dramatically worsens the fit and causes the relative price variable to become insignificant and of the wrong sign. The significance of the "profitability of exports" variable, \n(pxne/pq), depends on the specification. When used with the inventory gap, it is definitely not significant and can be dropped, whereas the relative price of exports variable PRELX is significant and cannot be dropped according to the F-test. Once Igap is dropped, however, the situation is reversed, and the "profitability" variable becomes significant, while PRELX gets the wrong sign and becomes insignificant. Thus the questions of whether Greece is a pure price taker or not, and whether the "profitability of exports" variable exerts a stronger influence than the relative price abroad of Greek exports cannot be answered conclusively. In terms of statistical fit, however, the preferred model is equation 2, that includes world income, the three-period moving average of the relative price of exports, and the inventory gap, and its performance is shown in Figure 5.14. By way of illustration, Figure 5.15 shows that dropping the Igap variable (model 6) weakens the explanatory power of the equation, especially in later years. Let us see if the size of the coefficients makes sense in the fight of the stylized facts. First of all, the coefficient on the inventory gap is surprisingly high, and is more than twice as large as the comparable coefficient for Canadian exports in Helliwell et. al. [60]. To see whether such a coefficient is reasonable, I traced the absolute effect of a 1% increase in last period's inventories on exports. Such an increase would reduce the inventory gap by 1%, and cause a 0.87% increase in exports. Since, throughout my sample period, export levels have fluctuated between 50% and 32% of inventory stocks, 43 to 28% of last period's inventory stock increase spills over into increased exports. Such a percentage is high enough to cause suspicion, but not certainty, of spuriousness. The one obvious Chapter 5. Imports, Exports and Consumption 103 Lagand • W « l Year Figure 5.15: Export Equation, without Jgop, w. Profitability Variable problem with such a high coefficient is that it can be a serious source of instability in simulation, since inventory investment is a residual item and can thus be quite unstable. An unstable export equation feeds through the sales variable in the output equation and destabilizes the entire model. However, as mentioned above, constraining the coefficient or dropping the variable results in sizeable worsening of the fit. The price elasticity after the lags have been worked out at the end of three years is quite high (|r;| = 2.89), 3.5 times higher than the estimated price elasticity of imports. Considering the type of commodities Greece exports (agricultural goods, raw materials, and light industrial goods), as opposed to imports (inputs to the production process, cap-ital goods and manufactured goods) the difference in elasticities should not be surprising. Chapter 5. Imports, Exports and Consumption 104 The consequence of this difference in elasticities is that devaluations, domestic inflation and, more generally, anything that affects the terms of trade should affect exports more than imports. The income elasticity of exports (1.7) is higher than the one for imports (which is unity). Consequently, for a given real exchange rate, there should be a continuous improvement in the (non-energy) trade balance, as long as Greek rates of growth are no more than 70% higher than world averages (the latter appears unlikely for the foreseeable future). Despite well-known problems of chronic deficits on the current account, this is in fact borne out by the data: the non-energy balance of trade has been improving, with exports growing faster than imports, while the exchange rate has been staying close to purchasing power parity.2 However, the non-energy trade balance started from such a large deficit position, that the recent problems with the current account can be attributed instead to a collapse of invisible receipts, in the form of shipping and other services, emigrants and seamen's remittances and official unrequited transfers, as well as a growing energy trade deficit (which has improved since the end of the sample due to lower oil prices). Figure 5.16 shows the logarithms of imports, exports and the trade deficit. The latter shows a downward trend in absolute value (with a bulge during the 1981-83 world recession), which, given strong growth in international trade, implies an even faster decline in percentage terms. 5.3 Consumption A consumption function is necessary to close the model on the expenditure side (since government expenditure is considered a policy variable). My consumption function at-tempts to explain real per capita consumption. I approached the modelling without 2More on PPP in section 7.4. Chapter 5. Imports, Exports and Consumption Figure 5.16: Log Non-Energy Imports, Exports and Trade Balance Chapter 5. Imports, Exports and Consumption 106 Legend O 3 S - ] Year Figure 5.17: Consumption Equation, Keynesian much preconceived structure. First I estimated a set of linear models, and their results are reported in Tables 5.16 and 5.17. I started with a straightforward Keynesian con-sumption function that explains real per capita consumption, Cpc, in terms of real per capita disposable income, Ydpc (model 1, Table 5.16, and Figure 5.17). Since The General Theory, however, such a straight-forward relation between con-sumption and current disposable income has been challenged by the Kuznets paradox (Kuznets [67]) and all the subsequent attempts to resolve the latter. Among such at-tempts the life-cycle hypothesis3 and the permanent income hypothesis (Friedman [41]) 3The basic ideas behind the life-cycle hypothesis can be attributed to Irving Fisher [39], and were further developed by F. Modigliani, A. Ando and R. E. Brumberg in a series of articles, including the empirical tests in Ando and Modigliani [4]. Other early contributions include Tobin [98] and Nagatani [81]. Chapter 5. Imports, Exports and Consumption 107 are the most important and the most commonly used in empirical work. Real wealth and labour income are the main variables used in life-cycle models. In the case of Greece, however, there are insurmountable problems involved in calculating the actual market value of assets: capital markets are so undeveloped that the only asset whose value is known is money balances. Different proxies for wealth have been tried oc-casionally: Koutsouvelis & Karadeloglou [66] for example use money balances and money balances plus capital stock. Since the capital stock data do not reflect actual market value of assets (they are derived by a straight-forward perpetual inventory method), it is not surprising that both variables have very insignificant (and wrongly-signed) coefficients. I conclude that lack of data on the market value of assets rules out direct modelling of wealth effects. I tried to capture wealth effects indirectly through differential saving rates,4 interest rates and the inflation rate. My next step was to split the disposable income variable into its wage {Ydpcw) and non-wage (idpcntu) components (model 2, Table 5.16). The reason for this was that it is possible that the marginal propensity to save out of each type of income may be different, and splitting the variable might indirectly capture wealth effects. Unfortunately, I have no information on the effective tax rates for the two types of income, so each type was assumed to be taxed at the same rate at first, and then I experimented with different rates, under the condition that they added up to the overall rate, and the wage income tax rate was higher. I tried this because it is well known that Greek wage and salary earners pay most of the income tax (although the statutory tax rates are the same, tax avoidance and evasion are endemic among self-employed people and generally non-wage income earners). Whatever the assumed tax split however, coefficients on the two variables are not significantly different. I tried using the two separate variables in different specifications, but the F-Test on the restriction that their coefficients be equal 4Because non-labour income could be considered a proxy for wealth. Chapter 5. Imports, Exports and Consumption 108 (1) (2) (3) (4) Constant 3.7131 (8.61) 3.6570 (5.28) 3.5829 (6.27) 3.1652 (7.00) Ydpc 0.69650 (55.95) 0.74550 (5.85) 0.42980 (2.96) 0.69033 (11.55) Ydpcnw 0.70528 (8.38) -0.04688 (0.39) 0.36276 (1.84) R2 s.e.e. D-W Durbin-/i F-test 0.9936 0.49247 1.1742 0.9933 0.50751 1.1824 0.0112 0.9927 0.52708 1.3247 3.232 0.9961 0.38833 2.4598 7.743 Restriction on (2): YdpCw = Ydpcnw Restriction on (3) and (4): pc = Tpd = 0.0 Table 5.16: Consumption Equations (Dependent Variable: Cpc) Chapter 5. Imports, Exports and Consumption 109 Figure 5.18: Consumption Equation, Habit-Persistence was always below 0.1. The results reported in the table assume an effective wage tax rate double the non-wage tax rate. A glance at the diagram corresponding to model 1 (Figure 5.17) shows that the estimate has a much higher deviation from trend than the actual dependent variable, thus suggesting intertemporal smoothing of consumption. Such a fact is of course to be expected in the context of permanent income and life-cycle theories of consumption. As a first approximation I tried two alternatives. First I included lagged income (model 3) and then lagged consumption (model 4). Lagged income corresponds to a simple version of the permanent income hypothesis. A significant coefficient on lagged consumption, on the other hand, would imply "habit persistence," which is a consequence of life-cycle Chapter 5. Imports, Exports and Consumption 110 consumption behaviour, although other justifications for it have been advanced.5 The coefficient on lagged income, when included by itself was insignificant and nega-tive. It becomes significant and of the right sign, however, when the inflation and interest rate variables are also added, as explained below. Lagged consumption, on the other hand, is consistently significant and positive with all specifications. Its inclusion by itself (model 4) improves the fit, as shown by Fig-ure 5.18. Clearly there is "habit persistence" in Greek consumer behaviour; consumption fluctuates less than disposable income. The habit persistence involved appears to be strong, given the size of the coefficient on lagged consumption. Such a result makes sense in the context of life-cycle consumption theory. In the case of Greece, however, given the substantial imperfections in capital markets, life-cycle income behaviour has to be understood more in the context of intergenerational transfers than personal retirement planning. Young individuals and couples are often supported by their parents well into their adulthood; the great importance of seniority in most jobs makes this necessary and possible. On the other hand, children are expected to provide for their parents if the state pension is inadequate. Private pension plans are very rare. The consequence of this situation is that the major wealth effects on consumption must be through the value of real balances, and thus the inflation rate is probably important; interest rates should not affect consumption behaviour in any way more complex than through direct effects on the saving rate. Another important asset other than real balances is real estate; since I have no data on its current value, I cannot test its effect on consumption. 5The earliest reference to "habit persistence" I have found, in Brown [20], predates the life-cycle models by over a decade. Brown uses Canadian data, splits disposable income into wage and non-wage income, and uses lagged consumption and lagged income as alternative explanatory variables. Habit-persistence is his maintained hypothesis, as emphasized by his title. Alternatively, lagged consumption could be considered part of an error-correction structure (as Koutsouvelis & Karadeloglou [66] explicitly do, following Davidson, Hendry et. al. [33]). Chapter 5. Imports, Exports and Consumption 111 On the basis of the above, I estimate three more models with the consumer price infla-tion rate, the nominal interest rate, and the two of them together, for each specification (lagged consumption and lagged income). The empirical results are given in Table 5.17. I did not use the real interest rate by itself since I could not define it independently of the observed inflation rate. If the nominal interest rate on private deposits is rpj and consumer price inflation is pc, then the real interest rate is — pc, and adding both variables to the lagged consumption model gives: Cpc = a0 + ctiYdpc + d2Cpc_x + ct3Pc + a4(rpd - Pc) (5.54) or Cpc = a0 + ctiYdpc + a 2 C p c _ 1 + (a3 - <*4)pc + a4rpd (5.55) Similarly, this type of model can be estimated with lagged income instead of lagged consumption: Cpc = 0O + ^Ydpc + 02Ydpc_1 + (03 - /34)pc + 04rpJ (5.56) Thus model 1 estimates equation (5.55). The estimate of (03 — ct4) is not significant, and the F-test on dropping it has a low statistic (model 3). Dropping the nominal interest rate, on the other hand, has a much higher F-statistic, and the fit deteriorates when it is dropped (model 2). Model 3 has a better fit, but it implies a strong degree of money illusion in saving behaviour (households are only interested in the nominal interest rate). All in all model 2 is preferred on theoretical grounds, and a glance at Figure 5.19 where it is plotted shows less serial correlation of errors than model 3. The results of first-difference estimation confirm this preference. Model 4 in turn estimates equation (5.56), and in models 5 and 6 the nominal inter-est rate and inflation variables respectively are dropped. Both variables seem to carry some information (judging by F-tests). However, when the inflation variable is dropped Chapter 5. Imports, Exports and Consumption 112 (1) (2) (3) (4) (5) (6) Constant 1.7805 (3.47) 2.0416 (3.69) 2.1176 (4.65) 3.1942 (6.72) 3.1235 (6.07) 3.6514 (7.12) Ydpc 0.22290 (1.78) 0.34070 (2.68) 0.22085 (1.74) 0.41228 (2.56) 0.52131 (3.35) 0.63343 (4.34) 0.73355 (4.07) 0.54490 (3.04) 0.72227 (3.95) 0.36712 (1.98) 0.22966 (1.31) 0.09436 (0.60) Pc -0.03266 (1.35) -0.06316 (2.82) -0.09142 (2.05) -0.09205 (1.90) rpd -0.12031 (2.42) -0.15598 (3.64) -0.09883 (1.48) -0.10015 (1.23) R2 s.e.e. D-W Durbin-/i F-test 0.9977 0.29883 -0.2138 0.9972 0.32961 1.9428 5.841 0.9976 0.30355 -0.41662 1.827 0.9961 0.38447 1.2505 0.9954 0.41877 0.9444 2.196 0.9942 0.47009 1.1405 4.208 Restriction on (2) and (5): = Restriction on (3) and (6): pc = = 0.0 0.0 Table 5.17: Consumption Equations with Inflation and Interest Rates (Dependent Vari-able: Cpc) Chapters. Imports, Exports and Consumption 113 Figure 5.19: Consumption Equation, Habit-Persistence w. Inflation Variable Chapter 5. Imports, Exports and Consumption 114 Figure 5.20: Consumption Equation, Permanent-Income w. Inflation and Nom. Interest Variables Chapter 5. Imports, Exports and Consumption 115 (model 6) the lagged income coefficient is reduced to insignificance and the fit deterio-rates. Figure 5.20 shows the performance of model 4. The signs (and the frequent low t-ratios) of the coefficients on the inflation and inter-est rate variables in models 1 to 6, Table 5.17, are not surprising; both interest rates and prices have theoretically ambiguous signs in most models of life-cycle consumption, and empirical evidence on other countries is inconclusive.6 A negative coefficient on inflation implies that households increase their savings in periods of price increases, presumably to replenish the value of their real balances and to compensate for increased uncertainty. This rationale would certainly appear plausible to a casual observer of Greek society. In general, the "habit persistence" models have a better fit, although the lagged income specification has as good a theoretical justification in the permanent income hypothesis. Both types of model show evidence of autocorrelation problems. Given the ambiguous role of inflation and interest rates, I decided to use first difference specifications to shed light on these results. Since the main explanatory variables (disposable income, lagged income and lagged consumption) have the same strong time trend as consumption, the probability of serial correlation and multicollinearity is high. Expressing everything in first difference form should eliminate this problem. I differenced and re-estimated models 1 to 6 from Table 5.17. The estimation was performed both with and without a constant. The constant was only almost significant in the cases where nominal interest rate changes were included by themselves, and the results were almost identical in both cases. There does not appear to be an independent trend in consumption. The results of the estimation (without a constant) are given in Table 5.18. Current disposable income, lagged consumption and lagged income proved quite ro-bust as explanatory variables, autocorrelation was reduced, and both types of model 6See, for example, Carlino [26]. Boskin [19] does find a significant interest rate effect, but he is in the minority. Chapter 5. Imports, Exports and Consumption 116 (1) (2) (3) (4) (5) (6) 0.40990 (5.03) 0.40234 (5.21) 0.42958 (5.22) 0.45561 (6.43) 0.46242 (6.82) 0.51562 (7.00) 0.39607 (3.17) 0.40328 (3.34) 0.33026 (2.62) 0.25039 (3.11) 0.25058 (3.23) 0.13441 (1.83) -0.06208 (1.89) -0.04793 (1.95) -0.10459 (2.56) -0.08422 (3-24) 0.06787 (0.68) -0.05221 (0.65) 0.09377 (0.88) -0.06134 (0.63) s.e.e. D-W Durbin-A F-test 0.6832 0.35968 -0.4108 0.7023 0.34866 -0.6677 0.465 0.6491 0.37856 -0.8114 3.608 0.6343 0.38648 1.4133 0.6614 0.37185 1.5469 0.774 0.5563 0.42568 1.6911 6.575 Restriction on (2) and (5): A r ^ Restriction on (3) and (6): Apc = = 0.0 0.0 Table 5.18: First Difference Consumption Equations (Dependent Variable: ACpc) Chapter 5. Imports, Exports and Consumption 117 Legend • « C T U » t Year Figure 5.21: Consumption Equation-lst Diff., Habit-Persistence w. Inflation Variable ("habit persistence" and "permanent income") gave very similar results. The inter-est rate term's significance did not survive the transformation. It is very insignificant whether it is used by itself or with the inflation term. The inflation rate term, on the other hand, is everywhere significant and has the same negative sign as in the undiffer-enced equation forms. The best-fitting of the first-difference models, then, is model 2, that is a model that includes lagged consumption ("habit persistence") and the inflation rate. Its performance is illustrated by Figure 5.21. By way of comparison, model 4 is shown in Figure 5.22, where there is less evidence of negative autocorrelation. The conclusion from this exercise seems to be that inflation does reduce consumption for the reasons summarized above, while neither the nominal or the real interest rate have an Chapter 5. Imports, Exports and Consumption 118 Legend • 4SUJAV. Year Figure 5.22: Consumption Equation-lst Diff., Permanent-Income w. Inflation Variable effect on consumption that is robust enough to explain changes in consumption as op-posed to levels. Such a result is not surprising, given the weak effects of interest rates on households' wealth. Chapter 6 Prices and Wages This chapter contains estimated equations for output price, consumption price, wages, export price and consumer price. The first three, being strongly co-determined, are estimated in a block by three-stage least-squares. Output price enters this block along-side the consumption price because of the fact that the output price is what determines producer behaviour, whereas labour sumers. Descriptions of the individual equations and results are followed by a section on wage rigidity and the "warranted" or full-employment wage. The other two equations are estimated individually by 2SLS, like most of my model. Import prices are considered exogenous: the (nominal) exchange rate is a policy variable, and Greece is not a large enough trader to influence world commodity prices. All equations are in log-difference (or growth rate) form, so the dependent variables are rates of price and wage inflation. I attempted to connect the wage-price block to the supply core of my model by making as much use as possible of the production structure estimated in Chapter 2. In accordance to the stated aim of my thesis, I tried to combine and to fully endogenize both demand-side and supply-side factors that influence wages and prices. Unfortunately, one supply-side factor, namely labour supply, could not be endogenized. Labour supply ought to be explained by a behavioural equation, but the state of labour force data does not permit such an exercise: for 13 out of 22 observations labour supply is simply interpolated between census years and any estimate would be dominated by the assumed 119 Chapter 6. Prices and Wages 120 time trend. 6.1 The Main Wage-Price Block: Wages, Output Price and Consumption Price The output price, the consumption price and wages were treated as a block; simultaneous equation bias is expected to be strong, and my original OLS estimates confirmed this. Since efficient estimation is probably more important in this block, instead of the 2SLS method that I used for most of my estimated equations, I used three-stage least-squares with instrumental variables. 6.1.1 Output Price Output price is determined by three factors: "normal" factor cost, the (exogenous) price of world exports, and some measure of domestic excess demand. The definition of "normal" unit factor cost makes explicit use of the production struc-ture, as estimated in Chapter 2. To be more precise, it is the combination of factor prices determined by the cost function that is dual to the production function. This measure of "normal" unit factor cost gives us what unit costs would have been if all factors were used at cost-minimizing proportions.1 The assumption behind the use of this variable is that producers do not immediately pass on to their customers any cost increases that are solely due to abnormal utilization rates, or, in other words, only changes that affect the basic cost structure (after the factor mix is fully adjusted) result in higher prices imme-diately. This assumption is consistent with the general character of the model (quantity constraints and less-than-full price flexibility). Furthermore, an attempt to use actual ((pkf Kne + Pi L n e + Pe E)IQ) rather than normal unit costs resulted in a much smaller 1Thus it is identical to C e ' ^ e expected cost variable used in the Lucas equation tests in section 3.1.3. I shall continue to use this notation for it. Chapter 6. Prices and Wages 121 to UJ o xi T5 C o o 3 o < c o o c o o 0.30 0.25-0.20 0.15-0.10-_ 0.05-3 a. 3 o 0.00 65 7 0 1 7 5 Year Legend "~ » C T U » L —I— 8 0 Figure 6.23: Output Price Equation l-r + »Tv\7 (6.57) and less significant unit cost coefficient. The CES production function is self-dual, so the derivations are analytically possible, and "normal" cost is ° - [ " ' ( ? ) ' The next important variable that is used to explain output price is pwx, the price of world non-energy exports expressed in domestic currency (energy prices are explicitly included in factor cost). The role of this variable is capturing the influence of the export prices of countries Greece competes with internationally. In this sense, given Greece's character as a small open economy, the variable should also have a large and significant coefficient. On a priori grounds, the coefficients on the unit cost and world price variables Chapter 6. Prices and Wages 122 should add up to 1. The restriction is easily accepted in all cases. Finally, a measure of excess demand and supply should be used to capture the de-mand side. Two such variables (and transformations thereof) lend themselves to this purpose: the inventory gap {Igap), as a direct measure of excess demand or supply (de-fined endogenously in the supply block), and the rate of capacity utilization, defined as QIQnv It is important to make sure that in either case the variable has a mean of one. Then the coefficients on the unit cost and world price variables can be constrained to be equal to 1. Neither excess demand variable proved significant and of the right sign. This was dis-appointing, but there may be good reasons for it. As regards the inventory gap variable, in the light of suspected errors-in-variables of the inventory data,2 and the poor perfor-mance of the inventory gap variable in the output equation, an insignificant coefficient should not be surprising. In any case, there has been in place (until after the end of my sample period) a relatively widespread system of price controls that forces the response of producer prices to market conditions to be slow at best: prices were not allowed to deviate substantially from cost plus fixed markup levels. Whatever the effectiveness of the system (and there are many known cases when it was ignored almost universally in times of shortages of basic commodities), it must have provided an institutional founda-tion to cost-plus-markup pricing behaviour, and reduced the responsiveness of prices to excess supply and demand conditions. In 1974 the actual level of output price substantially higher than can be explained by domestic costs and world inflation. 1974 is an abnormal year, as explained in Chapter 4. I thus added the dummy variable, Dcypri, to take into account the effects 2 Inventory investment data may incorporate some of the statistical discrepancy in the national accounts. Chapter 6. Prices and Wages 123 in u o I •6 9 c o "o 3 o < c" o 0.25-1 0.20 0.15 Ls-gand • ACTUAL ^ 0.10-X. 0.05 M C o o 0.00 —I— 7 0 7 8 Year 8 5 Figure 6.24; Consumption Price Equation of political instability and uncertainty in this year of international crisis, general mobi-lization and political upheaval. This variable proved highly significant.3 Consequently the constant was not dropped. The results of the output price equation are shown below: pq =. -0.00988 + 0.64552 Ce + 0.35448p„,x + 0.05574Z?CMpP1 (2.98) (9.28) (5.10) (3.53) R2 = 0.9638, s.e.e. = 0.014073, D-W = 1.810 Wald x2-test = 3.151 (restriction: Ce + pwx = 1.0) 3The dummy was not significant in the other price and wage equations. Chapter 6. Prices and Wages 124 One remarkable result is the size of the coefficient on the world price variable. It is very large, and shows the degree of openness of the Greek economy. More concretely, it shows that the Greek economy is very sensitive to fluctuations in world prices, and that a devaluation would create inflationary pressures, or exacerbate them if, as is often the case, the devaluation is actually caused by domestic inflation. Both these are confirmed by simulation results reported in Chapter 8. The size of the world price coefficient led me to suspect that there may be some spurious correlation with the exchange rate. pwx is the world price of non-energy exports expressed in drachmas: pwx = pwxg -pfx. If the exchange rate were deliberately adjusted to maintain purchasing power parity, then pwx would be very strongly correlated to the output price, pq, because of exchange rate policy rather than terms-of-trade effects. To test for this possibility I split the pwx variable into its world price and exchange rate components. The two resulting variables had statistically identical and significant coefficients. I return to the subject of PPP in Chapter 7. The performance of the output price equation is shown in Figure 6.23. 6.1.2 Consumption Price The consumption price depends on output price, adjusted for indirect taxes net of sub-sidies (pqi), and the price of imports. Since the agricultural sector was left out of the supply block, I also attempted to include agricultural prices, but they did not prove in the least significant. Expressed in rate of change terms, the a priori constraint is that the coefficients of the explanatory variables sum up to one. The results for the consumption price equation are shown below. Chapter 6. Prices and Wages 125 pc = 0.69016p„ + 0.30984/w (6.35) (2.85) R2 = 0.9641, s.e.e. = 0.013247, D-W = 2.298 Wald x2-test = 4.391 (restrictions: pqi + pmne = 1.0, constant = 0.0) The constraint is easily accepted. The coefficient on the import price term is also sizeable, the implication being that imported inflation affects consumer prices doubly, not only directly but also through the output price. The fit of the equation is shown in Figure 6.24. 6.1.3 Wages The wage equation is an expectations-augmented Phillips curve, that is best interpreted as a quasi-reduced form of both supply and demand factors. The annual rate of change of wages is explained in terms of its lagged value, the rate of change of the consumption price, the log of the ratio of unemployment to its sample average, and two dummy variables, one for 1980 (a year of wage restraint) and one for 1982 (a year of statutory wage increases, following the election of the socialist government). Thus the wage equation describes a disequilibrium labour market with forward-looking wage-setting behaviour. If the coefficients of lagged wage change and the consumption price change add up to one, there is no long-run money illusion. In addition, the significance of the coefficient on the unemployment term tests the hypothesis that labour market tightness influences wage-setting behaviour (the results of other macro research on Greece are largely inconclusive). The estimation results are reported below. Chapter 6. Prices and Wages 126 in 0.25-o I •u c o "5 3 O < 0.20-0.15-O $ o c a Legend o oc o . o s - 70 75 Year 85 Figure 6.25: Wage Equation 0.38767pc+ 0.41662^- 0.01619ln(ru/ < ru >) (6.57) (5.25) (3.08) 0.03523 .D82 (2.83) R2 = 0.9438, s.e.e. = 0.024656 There are two theoretically important variables missing from the wage equation. First, the ratio of unemployment benefits to the average wage; changes in benefits would in-fluence the workers' reservation wage, and hence the wage equation. Unfortunately, the pt = -0.04948+ (5.65) - 0.45989 D80+ (5.85) Chapter 6. Prices and Wages 127 data are very sketchy: all I know is the total amount of benefits paid, and since the number of the unemployed is subject to serious errors (the unemployment figures I used do not refer to the officially registered benefit recipients, but have been adjusted), the average level of benefits is a seriously flawed variable. In any case, the unemployed are known to rely more on their families than on the inadequate (by Western European standards) benefits. I was thus not surprised to get an insignificant coefficient for this variable, which I subsequently dropped. Secondly, labour productivity growth should be part of a Phillips-type wage equa-tion. However, in the CES framework developed here the maintained hypothesis is of constant labour productivity growth (T = 1.811%). Actual productivity growth may of course deviate from the equilibrium growth rate for protracted periods, when cyclical fluctuations in output and capital stock growth are combined with gradual employment adjustment. Given its definition, the actual labour productivity index (equation 2.17) captures a large amount of cyclical output variation. I attempted to introduce the rate of growth of actual labour productivity in the wage equation, but its coefficient was not statistically significant (t-statistic = 0.9). Thus productivity growth is simply subsumed under the constant term of the estimated equation. The restriction that the coefficients on the lagged wage change and the consumption price change add up to one (i.e. the assumption of no money illusion) is heavily rejected (with a Wald x2_statistic of 14.883). However, the constant term of the estimated equa-tion is clearly too large to be explained solely by productivity growth or to distortions due to the dummy variables. For example, the constant implies a 4.9% annual wage increase in the absence of any other influences, while average labour productivity growth as estimated in the CES production function is only 1.8%. It seems that, given the pe-culiar tripartite system of national collective bargaining, negotiated wage increases start from a base rate that is independent of past increases and expectations, and are modified Chapter 6. Prices and Wages 128 according to these plus labour market conditions. The coefficient on the unemployment gap seems robust enough: it was significant in every specification tried, with a size that did not vary much. Thus the hypothesis that the unemployment rate affects wage inflation (i.e. that an expectations-augmented Phillips curve exists) cannot be rejected. In the specification I report the unemployment gap enters log-linearly: the relevant term is hx(ru / < ru >). I used the sample average rather than some estimate of the "natural" rate of unemployment because, given the state of labour data on Greece, the calculation of such a rate is exceedingly difficult and has not been attempted. In an alternative specification the unemployment gap could enter linearly, and would be defined as (ru — < ru >)/100. Such a specification has the advantage of permitting the reading of the responsiveness of wages to unemployment by simply looking at the coefficient. Most research on the Phillips curve points to a highly non-linear response of wages to unemployment, however, and the log-linear specification performs better statistically. The wage responsiveness implied by the two specifications is very similar in any case, which increases my confidence in the robustness of the results. For example, an increase in the unemployment rate by one percentage point (starting from the sample average rate of 3.9%) will reduce the annual rate of wage increase by 0.37%, which compares with 0.39% in the linear specification. This result is unusual for previously published research on the Greek economy: ag-gregate Phillips curve results are almost always ambiguous.4 Apart from the perennial problem of inadequate data I suspect that, given the serious simultaneous equation bias in a wage-price block (of which I got a strong indication when estimating the wage equa-tion by OLS), the disappointing results are to a large part due to the adherence of Greek researchers to OLS estimation methods. 4 For example, every government-sponsored macro model mentioned in section 1.3 either does not attempt to include unemployment or has an insignificant coefficient. Economou [38] estimates wage equations with manufacturing data; sectoral unemployment rates do not have much meaning, however. Chapter 6. Prices and Wages 129 A second general specification of the wage equation that I attempted was one with static extrapolative expectations. The rate of wage increase here depends not on current consumption price inflation but on lagged price inflation. The lagged wage increase was included as well, to capture possible habit-persistence behaviour, or, if its coefficient proved negative, catchup behaviour on the part of workers. The fit of all three equations in the wage-price block was substantially worse, so I abandoned this type of expectation specification. I also attempted to include a capacity utilization variable in order to reflect labour market tightness not captured by the unemployment gap variable, specifically because of the quasi-fixed nature of employment as a factor of production. All institutional evidence points in the direction of quasi-fixity (it is legally difficult to dismiss or lay off personnel, and the longer the person has been employed the more difficult it is). In addition, my estimated labour equation points to a very slow speed of adjustment in response to changing cost-minimizing factor ratios. Thus the effects of capacity utilization are, at least in theory, important. Since my average wage is per person-year rather than per hour (while employment is persons employed rather than labour input), utilization effects should also reflect the role of changing overtime. I used a two-year average of the change in the utilization rate to allow for adjustment lags. The results were, to say the least, mixed. This variable was only significant in one variant of the wage-price block, with a translog production structure combined with static-extrapolative expectations. The fit of all three equations in the block was markedly worse than the ones reported here. I do not consider the result robust enough to report. Thus the elusive demand-side connection (the supply-side, unit cost connection is certainly there) between the supply block and the wage and price block seems to enter solely through the unemployment rate. I suspect that the normal cost variable carries capacity utilization information too, since total cost is derived from normal output and is Chapter 6. Prices and Wages 130 divided by actual output. This is actually borne out by the behaviour of variable C/Ce (actual over normal cost) in the Lucas alternative hypothesis tests (Table 3.5, section 3.1.3), where it is strongly and negatively correlated with capacity utilization. The fit of the wage equation is shown in Figure 6.25. 6.1.4 Wage Rigidity and the "Warranted Wage" International comparisons of "wage rigidity" are a crucial element of the explanation of the phenomenon of stagflation. Supply shocks inevitably lead to higher prices and a reduction of output below what its time path would have been in the absence of the shock. In fact, as Bruno and Sachs [22] and others have stressed, an adverse supply shock (in the form of higher prices for imported raw materials) is equivalent to some forms of technical regress. Downward inflexibility of the real wage would prolong the adjustment process and exacerbate the problems of inflation and unemployment. As described by the estimated wage equations described above, neither real or nominal wages respond instantaneously to the unemployment gap. The measure of real and nominal wage rigidity as defined in the literature5 are based on the following estimated wage equation (using this chapter's notation): pt = ctpa + (1 — ct)pt_t + a/3(ru— < ru >) + ct^x + e (6.58) Then the real wage rigidity index is defined as the coefficient on the absorption price inflation, divided by the coefficient on the unemployment gap, or a/a/?. The index for nominal wage rigidity is defined as real wage rigidity index times the average lag. The latter is defined as a/(l — a). The difference between this equation and my own preferred equation is that the unemployment gap enters linearly, lack of money illusion is imposed, and absorption 5e.g. Grubb, Jackman and Layard [50] and Klau and Mittelstadt [65]. Chapter 6. Prices and Wages 131 price rather than consumption price is used. To make my results comparable, I re-estimated the wage equation under the first two assumptions (substituting pa for pc makes no difference, since the time paths of the two are almost identical). The coefficient on the wage gap thus estimated is a/3 = 0.39305, with a = 0.38229. When we divide the two, we get a real wage rigidity index of 0.973. This compares to Grubb, Jackman and Layard's estimates of an OECD average of 0.92, Japan's low index of 0.13 and the U.K.'s high 2.39. Calculating the nominal wage rigidity idex gives an equally unremarkable result of 1.13 (OECD average: 0.62, highest: U.S.A. with 3.14). The conventional wisdom in the stagflation literature is that the US has high nominal wage rigidity (due to the nature of its labour contracts), which in times of inflation translates into rapid downward changes in real wages. In the UK, however, there is real wage rigidity, which is blamed for more persistent unemployment problems. On the other hand, Japan has the lowest measures of real wage rigidity, and the lowest unemployment rate. Greece seems to have an average degree of real wage rigidity and a slightly above-average nominal wage rigidity.6 These results on wage rigidity are somewhat puzzling, since they do not explain the persistence of stagflation in the Greek economy. However, comparative studies of stagflation and its connection with wage rigidity have yet to establish an incontrovertible link between the two; as described in Helliwell [55], such comparisons are sensitive to specification and have to tackle difficult problems of comparability. Another type of labour market rigidity, captured by the very low coefficient of adjustment in the labour 6Bruno and Sachs [22] argue that the behaviour of wage-setting institutions should affect wage rigidity. Such behaviour is measured by a "corporatism index," developed by Colin Crouch [30]. The index takes higher values the more centralized the union movement and the more coordinated the response of employers. The connection between their measure of nominal wage rigidity and the corporatism index is not statistically significant in cross-country comparisons, but corporatism seems equivalent to flexible real wages in the sense that it negatively correlated with stagnation across countries. Given what we know about the institutions of the Greek labour market, Greece could be classified as more corporatist than average, something that could explain the relatively low wage rigidity indices. Chapter 6. Prices and Wages 132 CL O o C O a w « 3 B) o 1.6 1.4 1.2 1-o.a-0.6 8 ; a » t - r u « 3 7 0 7 3 Yaar S 3 Figure 6.26: Three Measures of the Wage Gap demand equation (10%), is probably an important influence on the slow recovery of the economy. Another important question with a bearing on the persistence of unemployment and inflation is whether actual real wages have been excessively high, that is above the full-employment or "warranted" wage. Bruno and Sachs [22], Grubb, Jackman and Layard [49], Sachs [91], Malinvaud [73], and J. R. Artus [7], among many others, have emphasised the role of the wage gap. Simply put, the warranted wage is the wage that would raise the demand for labour to the level consistent with the non-inflationary (or natural) unemployment rate. The wage gap is, then, the difference between (or ratio of) actual and warranted wage. There are serious problems to sort out in order to compute a meaningful measure of the warranted wage. Chapter 6. Prices and Wages 133 First, there has to be an explicit production structure underlying the calculations: the degree of substitution between factors, which is necessary to determine the size of the wage elasticity of labour demand, and the rate of technical progress, which determines the rate of growth of wages when relative prices are unchanged, are both characteristics of the production structure. The explicit production function developed in Chapter 2 allows me to provides an answer, but as we have seen the specification of technical change is very much an open issue. Secondly, it is necessary to calculate the equiUbrium labour force and the natural rate of unemployment that would hold in the absence of cyclical influences. Both the real wage and the unemployment rate affect labour supply and frictional unemployment, and estimates of a "natural" or non-inflationary unemployment rate are hard put to disentan-gle cyclical from secular influences (as is the case, of course, with productivity change). In the case of Greece one can do little beyond contemplating these problems, because labour data are too shaky for the purpose. I make the simplest possible assumptions, taking labour supply as exogenous and using a sample average as a proxy for the natural rate of unemployment. The third issue is what to hold fixed while defining the full employment level of output. The two main approaches are cost minimization and partial profit maximization. Cost minimization sets the ratio of full-employment marginal physical products of labour and the capital-energy bundle equal to the ratio of the wage rate to the capital-energy price. Partial profit maximization on the other hand takes the existing capital-energy bundle as given and sets the full-employment marginal revenue product of labour equal to the wage rate. Both measures are easy to compute analytically from the CES functional form, but give different answers. The cost-minimization approach has to tackle the thorny issue of what is the appropriate rental price of capital. In the case of Greece the issue is fraught with problems, since profit rates have been declining through time. Different definitions Chapter 6. Prices and Wages 134 of the price of capital (that do or do not incorporate the secular drop in profitability) give different time paths for the warranted wage and consequently for the wage gap, i.e. the ratio of actual over warranted wage. On the other hand, the partial profit-maximization approach is not susceptible to the problems of defining the capital price, but by taking the capital stock as given it does ignore the fact that the desired capital stock could be quite different under the full-employment wage. In Figure 6.26 I show three alternative estimates of the wage gap in ratio form, two of them using cost-minimization and different capital prices, and one using profit-maximization. Clearly the peaks and troughs are in the same places but the time trends and the size of fluctuations vary dramatically. The profit-max approach gives the stablest results, and a quick comparison with the unemployment rate shows a strong correspon-dence. The fact remains, however, that arguments over wage gaps and whose fault unemployment is are fraught with problems, and a completely clear definition of the as-sumptions used is necessary if such arguments are to have any validity. One is left with the suspicion that warranted wage figures, at least in the case of Greece, can be used to support almost any position. 6.2 Absorption Price I estimated a absorption price inflation equation that is specified in exactly the same way as the consumption price, but is estimated by 2SLS outside the main wage-price block. The absorption price drives the price of capital in the model. The same adding-up restrictions on the coefficients of the variables as with consumption price are again easily accepted. The fit of the equation is very similar to the one of the consumption price equation, which is not surprising. Chapter 6. Prices and Wages L»g«nd • < C T U A l 0.00 H 1 1 r-8 5 7 0 7 5 8 0 Year Figure 6.27: Absorption Price Equation Chapter 6. Prices and Wages 136 pa = 0.78844p„ + 0.21255p mne (6.69) (1.79) R2 = 0.9708, s.e.e. = 0.011717, D-W = 2.132 F-test = 0.930 (restrictions: pqi + pmne = 1-0, constant = 0.0) The performance of the equation is shown in Figure 6.27. 6.3 Export Price Export prices are assumed to be driven by domestic prices, world export prices and the exchange rate. The a priori constraints that I attempt to impose are: 1. that the coefficients on the world price (in foreign currency) and the exchange rate are equal to each other, 2. that each be one minus the coefficient on domestic output, and 3. that there be no constant. The first two are easily accepted, but the last cannot be as easily imposed: the F-statistic is significant at the 95% level (but not the 99% level). Omitting the constant results in the following estimated equation: pxne = 0.54440p, + 0.45560pwxff + 0.45560p/x (1.95) (1.64) (1.64) R2 = 0.8899, s.e.e. = 0.028934, D-W = 0.918 F-test = 3.636 (restrictions: constant = 0, pq + pwxg = l.0,pwxg = pfx) Chapter 6. Prices and Wages 137 Allowing a constant improves the fit in terms of every summary statistic: Pxne = -0.01584 + 0.54011po + 0.45989Ptl,Xff + 0.45989p/:E (2.86) (2.28) (1.94) (1.94) R2 = 0.9200, s.e.e. = 0.024656, D-W = 1.337 F-test = 1.156 (restrictions: pq +pWxg = 1-0, pWXg = Pjx) However, the estimated constant implies an independent negative time trend in Greek export prices. This may mean either that the price series has a trend due to incorrect measurement, or the type of exports has been changing in a systematic way. I suspect that the explanation is a mixture of the two. An examination of exports by SITC group shows a long-term movement away from agricultural products and raw materials (groups 0 and 2) and towards light industrial goods (groups 1, 4, 5 and 6); the unit value indices for the latter have grown less rapidly than output prices. At the same time, the unit value index of Greek exports as reported in the IMF Financial Statistics has a faster time trend than the index reported by the National Statistical Service of Greece (which 1 use, since it is the one used by the National Accounts, and gives better-fitting export equations), so the possibility of incorrect measurement does remain. The difference in time trends is only one-tenth of one percent per year, however, so changes in the structure of exports have to explain the bulk of the time trend. The elasticity of the export price with respect to domestic output price is 0.54, while the elasticity with respect to world price and exchange rate is 0.46, regardless of whether the constant is included or not, so model behaviour in shock-control experiments is not affected. However, the constant has to be suppressed in long-term simulation for stability. The performance of the second equation is shown in Figure 6.28. Chapter 6. Prices and Wages 138 Legand • ACTUAL Year Figure 6.28: Export Price Equation Chapter 7 The Energy Block and the Rest of the Model This chapter, for want of a better arrangement, attempts to close the model by specifying equations and identities covering the energy sector, government finance, and housing investment. Reasons are provided for any failure to model a particular aspect of the economy. 7.1 The Energy Sector Energy is an integral part of my model. It enters the production function directly as a factor of production, and since the supply block permeates the rest of the model, energy enters the structure of the model macroeconomy at every level. Despite the recognized importance of energy at the macro level, the energy sector block that I estimate in this thesis is quite simple. The most striking fact about energy in the Greek economy is the very strong dependence on imported oil. The only domestically available type of coal is lignite. It is quite plentiful and cheap to extract by strip mining, but of a quality not much higher than peat, which makes it unsuitable for almost anything other than electricity generation and process heat for some industrial applications. The total hydraulic potential of the country is relatively low (many mountains but not enough rainfall), and new projects are becoming prohibitively expensive. No natural gas is domestically available: the small amounts produced by a single well do not merit a pipeline or network, and the absence of the latter has prevented importation of cheap Soviet gas from Bulgaria. The amounts of domestically produced oil (again from a single, 139 Chapter 7. The Energy Block and the Rest of the Model 140 high-cost undersea well) can only cover a small fraction of domestic demand. Thus the price of imported crude oil is probably the most important energy variable at the macro level. Domestic supply of energy (for reasons outlined below) is mostly exogenous and its macroeconomic effects is captured through simulations. Thus the energy block I estimate is quite simple. Its supply side is exogenous, energy demand drives energy imports, and through the split among the different energy forms the overall energy price is determined from its components. Of these components, only the price of petroleum products is endogenous, since its domestic value added depends on output price. 7.1.1 Quantities Total energy demand is modelled in Chapter 2 as a factor demand derived from the pro-duction structure. In this section total energy use in the energy-using sector is split into its three major components (petroleum products, electricity and coal plus coal products). I have two reasons for disaggregating aggregate demand like this. The first has to do with the fact that the price paths of oil and electricity (and, less importantly, coal) have been very different. While the world was experiencing rapidly rising crude oil prices, the Public Power Corporation followed a deliberate policy (imposed on it by the government) of supplying electricity at average production cost, without regard to the true marginal cost of the coal that had to be mined to generate it. As a result there was substitution of electricity for oil over time. This substitution is important, especially in its effect on oil imports, the only major component of energy import demand, and ignoring it incorrectly specifies energy imports and the aggregate energy price. An added reason for disaggre-gation is the potential of prediction and policy analysis with regard to specific energy supply issues, a potential that this thesis does not exploit. The split of total energy use is done by interrelated quantity share equations (with Chapter 7. The Energy Block and the Rest of the Model 141 symmetry imposed): SO = a 0 + Poo In pou + POL In pe/c + Poc In pcoai SL = aL + POL hip,*/ + fox, lnpe/c + foe lnp«wj (7.59) SC = a c + /5oc In Poi, + foe In Pe/c + foe In pcoa/ where 50, SL and SC are the quantity shares of oil products, electricity and coal prod-ucts respectively, pdhpeic and pcoai are 3-period weighted averages of current oil products, electricity and coal products prices: jW = f > , P O (7-60) 1=1 with symmetric definitions for the other two energy forms. Weighted averages are used to capture the effects of vintage capital embodying specific energy requirements. The retrofitting parameter, ifr, defined in section 2.3.1 as the rate at which old capital becomes malleable in energy use, is used to define the exponential weights: ' i = 5 M F ' ! = 1-2-3 <7-61> The estimated value of if> from Chapter 2 is 0.2. Since the shares add up to 1, one equation is dropped, and additional restrictions on the coefficients are implied: do + <*L + etc = 1 Poo + Poi + Poc = 0 POL + PLL + PLC = 0 (7.62) POC + POL + PCC = 0 Because of the cross-equation symmetry restrictions, estimation was carried out with iterative Zellner (seemingly unrelated) regression. Parameter estimates and summary statistics are given in Tables 7.19 and 7.20. Chapter 7. The Energy Block and the Rest of the Model 142 do 0.64186 (129.1) 0.32488 (70.36) etc 0.03326 (105.8) Poo -0.065393 (4.619) POL 0.086427 (6.889) Poc -0.021034 (2.782) PLL -0.12621 (8.774) PLC 0.039779 (6.018) Pec -0.018745 (2.845) Table 7.19: Energy Quantity Share Equation Parameters SO SL SE R2 0.6462 0.7750 0.6517 s.e.e. 0.015113 0.017302 0.008643 D-W 0.5177 0.3696 0.5976 Table 7.20: Energy Quantity Share Equations: Summary Statistics Chapter 7. The Energy Block and the Rest of the Model 143 All estimated coefficients are significant. All own-price coefficients have the expected negative sign. Of all the cross-price coefficients only Poc is negative when freely esti-mated, implying that coal and petroleum products are complements. This may be a spurious result. I tested the possibility that the coefficient reflects a general trend away from using coal and its derivatives (like coal gas) at the final use level at a time of rising oil prices by including a time trend in the share equations. The resulting estimate for Poc was still negative but insignificant, the significance and size of all coefficients (apart from the time trend) was reduced, standard errors dropped and a significant trend towards the use of more electricity was revealed. The percentage quantity and value share of coal and products is so low, however, that imposing a coefficient would not be too troublesome. Electricity demand could again be split into coal, hydro and oil generated electricity. Electricity generation and coal extraction are a virtual monopoly of the Public Power Corporation,1 which is government controlled, and clearly not profit-oriented. Thus the behaviour of the PPC cannot be modelled by usual microeconomic criteria, and modelling its behaviour by postulating some implicit objective function is beyond the scope of this thesis. I shall therefore treat lignite output and prices, electricity prices, and the levels of coal-fired and hydro electricity2 generated as exogenous variables. The amount of oil-fired electricity generated could thus emerge as a residual, However, there is a base amount of oil-fired electricity that is always produced: virtually all islands have their own free-standing oil-fired plants, since transmission lines from the mainland are uneconomical. Furthermore, coal-fired generation can always be adjusted within the space of one yearly observation, since the PPC supplies itself with lignite and always has large stocks at hand. The most sensible assumption at this point,3 then, is that the split among types 1In 1984, the PPC produced all the electricity and 98% of all lignite, the only domestically available form of coal. 2 Hydro output in any case depends on hydroelectric capacity and rainfall levels, both exogenous, the former at least in the medium term. 3If I were constructing a model with more energy sector detail, the cost structure and the production Chapter 7. The Energy Block and the Rest of the Model 144 of electricity generation is also exogenous. Since oil used in power generation is the only oil used in the energy-producing sector in significant amounts, the oil requirements of the energy-producing sector as a whole are also exogenous. By adding it to the final demand for petroleum products, plus the energy use in the agricultural sector, we get the total demand for petroleum products. The connection between this total demand and oil imports is complicated by the fact that Greece's refinery capacity was until the early 1970s inadequate for domestic needs, and large quantities of refined petroleum products were imported. Since then Greece has become a net exporter of refined products. In quantity terms (103MT), the relation between crude refined and total demand is Rcr = RL + d*, - Mrpp + A STrpp + misc. (7.63) or, in words, Crude Refined = Refinery Losses + Domestic Demand for Petroleum Products — Net Imports of Refined Petroleum Products -f A Stocks of Refined Products + misc. Of these I treat refinery losses as a constant percentage of crude refined (they have been approximately 10% throughout the sample). I also treat net imports of petroleum products as exogenous. As mentioned above, the amounts have always been decided by the government, and they have been determined by refining capacity and the match of refinery output mix with domestic demand. Finally, stock changes also have to be treated as exogenous. Whether they are a residual representing unexpected changes in demand, or a conscious decision by government to stockpile or run down inventories, stock changes cannot be modelled. Beyond these, there are also miscellaneous petroleum products (including lubricants, tars, semifinished products and feedstocks) that do not form part of the final energy demand system and amount to 5-14% of crude refined, as and pricing decisions of the PPC of course would have to come under more serious scrutiny. Chapter 7. The Energy Block and the Rest of the Model 145 well as statistical discrepancies, both listed as misc. above. Crude oil and gas has been produced in very small quantities (up to 12% of total domestic needs) from two undersea wells since 1981. The demand for imported crude oil and petroleum products emerges when we subtract these quantities, which I also treat as exogenous. Additions to stocks of crude are also treated as exogenous for the same reasons. The identity, again in quantity terms, is Mcr = Rcr + A STcr — (}„ + NE (7.64) or, in words Crude Net Imports = Crude refined + Additions to Stocks — Domestic Output + Non-Energy Use. The quantities of coal imports can safely be treated as exogenous. Coal is imported in small quantities primarily for coking and coal gas production. The latter has recently been phased out and replaced in the one existing gas network (which covers only part of the Athens metropolitan area) with liquid petroleum gas. The PPC also imports small amounts of sub-bituminous coal to mix with lignite in the thermal generating stations in order to increase burning efficiency. Thus the amounts of coal imported are not strongly related to overall energy demand, and virtually all of the coal needs of the country, whether for thermal generation or final use, are covered by domestically produced lignite. Electricity imports are also very small and mostly take the form of exchanges with neighbouring countries, primarily Yugoslavia, at times of seasonal or unexpected short-falls in power output. Again the PPC is solely responsible for such imports, which I shall also consider exogenous. Thus total net energy imports are a Divisia quantity index of net imports of refined products, crude, electricity and coal, of which only crude imports are endogenous. Chapter 7. The Energy Block and the Rest of the Model 146 7.1.2 Prices Until 1986 (after the end of my sample), the Greek government, either directly or through tightly controlled public corporations, controlled almost all energy quantities and prices on the supply side. It imported almost all crude oil to be refined for domestic use, and refined it in two state-owned refineries. Two privately-owned refineries, apart from re-exporting large amounts of their stock of crude, exported most of their refined products, and produced some to order for the state. The government imported and exported whatever specific petroleum products were needed to match the domestic demand mix to refinery output. It controlled the price of all major petroleum products from refinery-gate to wholesale to retail. It did so by setting its own refinery-gate price (since the sale of petroleum products to wholesalers was a state monopoly), taxes and even wholesale and retail margins. The situation is in the process of changing under pressure from the European Community, that ruled that a state monopoly in oil was not acceptable. For the duration of my sample, however, petroleum product prices were set by the government. The price of petroleum products that I use is a Divisia index of the final user prices of all major types of petroleum product. I have the total tax rates collected at all levels for each product, so I can calculate the average price net of taxes, Ponet- The aggregate price of oil is then pan = pmet (1 + ftp), rtp being the average tax rate on petroleum products. This price is in a sense a policy variable, given the state monopoly on refining, and the price controls based on officially set wholesale and retail margins. I did model it, however, because in the face of large crude price changes and domestic cost increases even officially set prices have to respond. The equation is in log-difference (rate of change) form, and is a function of world crude prices expressed in Drs (pcr) and the domestic output price (pq, representing the cost of domestic value-added). The coefficients are constrained to sum to 1 and the constant dropped. These restrictions are easily accepted (F-statistic =1.031). Chapter 7. The Energy Block and the Rest of the Model 147 The relatively poor fit of the equation is partly due to the fact that state refineries did not operate on a profit-maximizing basis, or even necessarily on a cost-plus-markup basis, but probably increased refinery-gate prices when it became inevitable, using stocks as buffers. Probably due to the fluctuating levels of buffer stocks, and the arbitrariness of the government's response to competing interest groups, my attempts to discover a lagged response structure were unsuccessful. Ponet = 0.65126PC, + 0.34874p„ (8.11) (4.34) R? = 0.8522 s.e.e. = 0.08214 D-W = 1.0837 F-test on restrictions (constant=0.0, Per + Pq = 1-0) : 1.031 The price of coal is exogenous, since domestic lignite prices are set by the PPC, and imported sub-bituminous and bituminous coal is again dependent on world prices and the exchange rate, which is a policy variable. Finally, as mentioned above, electricity prices are also exogenous, since they are set by the PPC, and I have not been able to model them in a way that is simple enough to be within the scope of this thesis.4 4The question of course arises of whether the assumption of exogenous non-oil energy prices is de-fensible. If non-oil energy prices actually depend on the price of oil, a simulation of an oil price shock that treats them as exogenous would incorrectly induce a lot of interfuel substitution. Such interfuel substitution is slow, and has a relatively low price elasticity. Electricity and coal prices increased a lot less rapidly than oil prices, with their prices relative to oil dropping by 50 to 60% over the period since 1974. It appears that there was a connection between oil price and electricity price (final coal use is negligible), although, given the non-market nature of the PPC's pricing behaviour, the connection is not quantifiable through an estimated equation. In order to model it properly, I would have to calculate the direct effects of an oil price increase on electricity supply (given the marginal use of oil-fired capacity), as well as the indirect effects through general inflation, plus second-round effects resulting from higher electricity demand. My model so far does not have enough energy sector detail for the purpose, although endogenizing the electricity price would be a necessary step for further research on energy issues. Chapter 7. The Energy Block and the Rest of the Model 148 The overall energy price thus emerges as the share-weighted average price of the three fuels: pe =PoiiSO + pelcSE + PcoalSC (7.65) 7.2 Government Finance Government finance enters my model very simply. Taxes are equal to exogenous tax rates multiplied by the taxable magnitude. The relevant tax rates are for indirect taxes (on GDP), personal income taxes (on personal income), direct corporate taxes (on cor-poration profits), and social security taxes (on the national wage bill). Government expenditure on goods and services and government transfer payments are considered exogenous policy variables. All government finance variables enter national income ac-counting identities in the usual way. Interest rates are controlled and the credit market is in a chronic state of excess demand. Furthermore, I have found no quantifiable short-term relation between the monetary aggregates and either output or the price level. As mentioned in Chapter 1, in the longer run such a link must of course exist, and the extent to which budget deficits are monetized is important. The latter issue has become much more important since the end of my sample, when deficits started growing out of control, and will have to be dealt with in any updates of this model, especially in long-term simulations and forecasting. For the moment, regardless of whether they are monetized or financed through borrowing, government budget deficits only enter the model through their direct effect on aggregate demand. Chapter 7. The Energy Block and the Rest of the Model 149 7.3 Housing Investment The conventional wisdom among Greek economists, economic historians and critics of post-war conservative populist governments5 is that a housing construction boom, ac-companying a rapid urbanization, was responsible to a large extent for driving the engine of economic growth.6 The agenda of the person making this assertion largely determines the context: from the unplanned, uncontrolled growth of the two largest cities, to alleged unbalanced growth, to the conclusion that when the construction boom petered out, the Greek economy stagnated as a result. Implicit in all these points of view about housing construction is the belief that investment in residential construction is fundamentally different from other types of fixed investment. The fact that investment in housing com-prises a very large percentage of total fixed investment, although true (on average, about a third of total fixed investment is in residential buildings), is not enough reason to treat it separately. If investment in housing were determined by the same forces that determine other fixed investment, there would be no point in singling out residential construction, except for the fact that the implicit and explicit services of housing capital is arbitrarily defined in the National Accounts. Thus I first set out to see whether housing investment does behave differently from total fixed investment. I initially tried to include housing investment and housing capital in the supply block, and estimate residential construction as part of the overall investment equation. The resulting output and derived demand equations showed a pronounced drop in the significance of all coefficients and a clear worsening of econometric fit according to every summary statistic. The investment equation suffered most, as expected. Next I attempted to calculate some form of desired housing capital stock (using Conservative governments ruled Greece between 1952 and 1981, with only a two-year interruption in the mid-1960s. 6Residential construction amounted to between 7 and 10% of GDP between 1961 and 1974. If the linkages with the manufacturing sector are also taken into account, this assertion becomes plausible. Chapter 7. The Energy Block and the Rest of the Model 150 disposable income, population and the price of housing relative to the consumption price), and then estimated partial adjustment and error-correction equations similar to the ones in Chapter 4. The results were also very disappointing. Finally I modelled housing investment in a way specific to itself. If Kh is real housing capital, and is desired real housing capital, then the following partial adjustment mechanism can be defined: *' • (19" (766) with 0 < a < 1 being the speed of adjustment. In logarithmic form, this becomes In (Jjr*-} = a l * K ' h ~ « ( 7 - 6 7 ) For small changes between Kh and Kh_x ,7 \Kh_J Kh_, Kh_, K ] Thus the equation becomes .-A- = a\nK-h - alnKn., (7.69) It remains to specify what drives the desired housing capital stock, both in the short and the long run. I must state from the start that the economics of residential construc-tion in Greece are quite different than in Western economies, at least from the supply side. Housing (almost exclusively in the form of strata-title apartment units) is typically built on very short-term credit and is largely financed through pre-sales of the units. Even the acquisition of the land requires no financing since property owners are offered a percentage of the finished units in exchange for the land. Thus interest rates can be expected to play a minor role in the construction costs. With regard to the financing of 7The rate of depreciation of housing capital in Greece is so low (estimated at 1.67%), that Ij^ gross investment, is a reasonable approximation to the change in the capital stock. Chapter 7. The Energy Block and the Rest of the Model 151 the purchases of housing units, the credit market is quite inadequate. Over the decades of credit-rationing, only a few families8 managed to secure housing loans. Thus it is not interest rates, but the amount of available credit for housing loans that is expected to be important. Another important source of financing for the acquisition of housing is capital inflows for the purpose, mostly from emigrants. Such inflows were encouraged (at least to the end of my sample) and enjoyed certain exemptions from exchange controls, so that the Bank of Greece has been publishing their amounts under a specific category. In the absence of well-functioning credit markets most of the financing for the pur-chases of housing must come from savings, and thus disposable income becomes an im-portant determinant of housing investment. Not only current disposable income, but also past levels of disposable income are important. I thus tried various weighted averages of past and present disposable income, but current disposable income by itself performed best. In the long run demographic variables like family size, number of families, urbaniza-tion, net migration etc. should influence the desired housing stock. Unfortunately I have very little reliable information on demographic data, so as a broad proxy I express all variables (other than prices) in per capita terms. Other likely long-term variables would be the net return on housing compared with alternative assets. Since I have no data on the asset price of housing, I tried to capture the effect of net return through the housing investment deflator relative to the consumer price. The results of using price variables in any form were disappointing: the coefficients were never significant and their sign varied with the specification. The expected sign of price variables is ambiguous in any case. For example, rapidly increasing housing costs relative to the cost of living may make hous-ing less affordable and divert more disposable income towards consumer durables. On 8Public sector employees enjoy lifetime secure employment, and their lifetime income can serve as collateral. Consequently their access to housing loans at administered, i.e. below-market rates has been considered a perquisite. Chapter 7. The Energy Block and the Rest of the Model 152 the other hand, in such a situation housing represents a good way of protecting savings against inflation. The equation I finally estimated was the following: V/tffcpe-i = 0.92877+ 0.11304 In Ydpc+ 0.026521 ]n(FH/Pop) (6.03) (1.93) (2.28) + 0.013634 \n(AHL/Pop)- 0.21769 In Hf /^ (4.14) (5.48) OLS, 1964-1984, R2 = 0.9124, s.e.e. = 0.005796, D-W =1.6402 where the pc subscript stands for per capita, Ydpc is per capita real disposable income, Fn are capital inflows for residential investment (in Drs and deflated by the consumer price), and Hr, the stock of outstanding housing loans (deflated by the same deflator). A variable that was tried but omitted (because it was insignificant and of the wrong sign) was the real interest rate on housing loans, which is yet another indication that interest rates were not influencing investment decisions significantly during the sample period. I also had to reject price variables, as mentioned above. The domestic and foreign credit variables were proven significant and of the right sign. There was no need for the same kind of political dummy variables for the 1967 coup and the 1974-75 Cyprus crisis: sharp drops in the capital inflow variable apparently capture the effect of political uncertainty. I must emphasize that I did not discover a correlation between capital inflows (whether general or specifically for investment purposes) and other fixed investment. Chapter 7. The Energy Block and the Rest of the Model 153 Thus the belief that housing investment is driven by different forces than non-residen-tial fixed investment appears justified. In fact housing investment has more in common with absorption than with the supply block, and is included under the "Domestic and Foreign Spending" section of the model in Appendix C. 7.4 Money Markets and International Finance Financial markets in Greece are quite unlike those in industrial countries. Bond mar-kets are virtually nonexistent, and other financial and capital markets range from the rudimentary to the nonexistent. As I have already mentioned in the introduction, the money market and credit markets are almost completely controlled by the government. The two largest commercial banks (with a credit market share of perhaps 70%) are state-owned. Beyond this, the government sets not only the high-powered money supply (so as to control the supply of money, however defined, through bank reserve ratios), but actively interferes with the total amount and allocation of credit. This it does by fre-quently changing a labyrinthine system of special reserve ratios, specific to different uses of credit, and by imposing direct credit quotas, both on total credit and disaggregated by use. Finally it sets all interest rates, on deposits as well as loans, the latter again de-pending on the use they are put to.9 Such heavy-handed control over the economy does not make monetary policy unimportant. Rather, it makes the usual focus of macroe-conomists on total money stock and a few key interest rates inappropriate in the case of Greece. In the absence of market-clearing interest rates and widely available credit to reasonably solvent firms (and individuals), the financing of investment must take un-orthodox routes, the main ones of which were discussed in the introduction. A detailed model of domestic and international credit and its linkages with the real economy would 9Demopoulos [34] provides a good discussion of these institutional complexities. Chapter 7. The Energy Block and the Rest of the Model 154 be a very interesting exercise. Unfortunately there is a serious lack of data that comes from widespread disguising of financial transactions (especially ones with the rest of the world, which are not only regulated but very often illegal). Thus official data exhibit little empirical connection with economic behaviour: I was not able to discover clear transmission mechanisms from either the interest rate or the monetary aggregates to the rest of the economy, or from aggregate capital inflows to investment. Innovative mod-elling techniques as well as more disaggregation are probably necessary (as the housing investment equation shows). Needless to say that given the institutional conditions in the money market, I decided against estimating a money demand function, though it has often been tried before (as in [88] and [36]). As for international finance and the exchange rate, first of all the drachma is not fully convertible. There are tight foreign exchange controls and it is generally impossible to buy foreign securities, and thus interest parity frequently does not hold. The exchange rate was fixed until 1972, and since then has been on an adjustable peg and, more recently, a crawling peg system. Thus it has always been a policy variable. The only hypothesis I test with regard to the exchange rate is whether there has been purchasing power parity. In the 12 years when the exchange rate was not fixed, it diverged from pxne/Pwxg, the ratio of export prices (in Drs) to world export prices (in US$), by between -11.2% and +10.4%. I regressed lnp/x (pfx is the exchange rate in Drs per US$) on ln(pxne/pwxg) and tested whether the coefficient is 1.0 with no constant. The F-test was 12.185 with 2 and 10 d.f., so the PPP hypothesis is rejected,.at least tentatively. PPP is nevertheless imposed for out-of-sample simulations. Beyond the exchange rate, my attempts to model capital flows and unrequited trans-fers were not fruitful. I also treat them as exogenous, and the balance of trade in goods and services will be the balance of payments variable I shall focus on. Until the end of my sample, borrowing from abroad had not been substantial, and consequently neither Chapter 7. The Energy Block and the Rest of the Model 155 had interest payments. The recent balance of payments crises that have led to rapid increases in indebtedness can be dealt with in out-of-sample simulations, and represent another item for further research. The situation in the money market, along with other market imperfections due to institutionalized government intervention, is bound to change under pressure from the European Community. A model that ignores the money market, interest rates and the consequences of the financing of budget deficits will soon prove inadequate. The same is true for capital flows and the exchange rate. Unfortunately it will take some time to gather enough observations on the newly-liberalized domestic money market and inter-national capital flows to model them properly. Such work lies beyond the scope of this thesis. Chapter 8 Simulation Results In this chapter I report the behaviour of my model in simulation. The results represent simulation experiments which followed only limited adaptations of the model structure. An ideal model-building exercise would have included many rounds of feedback from simulation behaviour to model specification and estimation. A macro modeller's work is seldom done, however, and the work in this thesis goes only part of the way in an essentially asymptotic process. The simulation results reported here should be considered as steps in this process rather than as definitive answers to causal relationships in the Greek economy. I first ran in-sample simulations of different variants of the model without the residuals and computed the Root Mean Square (RMS) errors.1 The size and dynamic behaviour of the RMS errors was an important factor in deciding on the final form of several equations, and often the choice dictated by stability in simulation was different from the one originally chosen on the basis of better econometric fit. Next I conducted simulations of three key shocks. Two of these shocks are cases of political instability, and occurred in 1974-75 and 1967. The 1974-75 simulation results have to be interpreted with particular caution, partly because of intractable data prob-lems, partly because of a less than fully systematic specification of the dummy variables. The third shock involves the rapid increase of crude oil prices in the 1970s. Finally I conducted three key tests of the stability of the model. First I did a predictive 1The residuals are reintroduced for in-sample simulations, and are constrained to exponentially decay to zero out of sample. 156 Chapter 8. Simulation Results 157 run, in other words I ran an out-of-sample simulation until 1999 with exogenous variables following smooth growth paths. Then I ran 25-year (in and out of sample) simulations of a fiscal and a devaluation shock. The out-of-sample behaviour of the model is not completely satisfactory. More careful specification of exogenous variable paths could lead to substantial improvement, but the inherent dynamic stability of the model is a question that needs to be addressed in the future. Another possible reason for long-term instability, especially in response to shocks, is the exogenous interest rate: fixed nominal interest rates, or fixed real interest rates different from the Wicksellian "natural rate" have been shown to cause explosive long-run model behaviour. Given the doubts about the long-term stability of the model, all simulation results in this chapter should be interpreted with caution. For relatively short in-sample periods, however, problems with long-run stability need not necessarily invalidate the results. Simulations of other events and alternative government policies represent obvious further applications of my model. 8.1 R M S Runs and Choice of Specification The behaviour of the different variables in simulation varied widely, the ones defined as residuals usually having the largest RMS errors. For example, inventory investment, the unemployment rate and the profitability variable that enters investment are all defined as residual items; their RMS errors were higher and the variables less stable than the rest of the variables. The investment equation also had large RMS errors, something I had expected: given the strong influence of lagged investment and the slow speed of adjustment, errors tend to accumulate over time, and the departure of actual from desired capital stock does not correct the situation very rapidly. In general, using the error-correction forms of the investment and labour demand Chapter 8. Simulation Results 158 /ne Kne E Qnv Q 1974 -4.605 -0.500 0.618 -0.235 0.119 1.200 1975 -11.305 -1.457 0.025 -3.704 -0.646 -8.513 1976 -19.187 -3.098 -1.812 -2.489 -2.448 -8.624 1977 -18.614 -4.547 -2.623 -6.982 -3.514 -4.600 1978 -16.689 -5.706 -2.060 -12.803 -3.612 -6.600 1979 -21.238 -7.279 -1.725 -10.786 -3.883 -6.445 1980 -26.579 -9.201 -0.755 -11.275 -3.821 -9.730 1981 -25.252 -10.657 -3.321 -11.257 -6.014 -1.630 1982 -21.530 -11.552 -2.515 -10.094 -5.959 0.863 1983 -4.515 -11.079 0.610 -10.237 -3.956 2.376 1984 10.685 -9.811 2.672 -12.745 -2.362 2.775 Average 17.928 7.798 1.982 9.400 3.747 5.768 Co Mne C A h 1974 4.658 -2.251 4.516 2.015 1.429 12.131 1975 13.127 -13.872 -9.117 -4.715 -4.348 -6.966 1976 11.476 -11.222 -9.693 -6.079 -5.232 6.051 1977 1.521 -9.299 -1.611 -5.285 -4.908 0.824 1978 1.174 -9.051 -6.880 -5.862 -6.003 -12.042 1979 -0.179 -8.960 -7.281 -5.383 -6.469 -17.051 1980 2.644 -9.051 -9.221 -5.490 -7.491 -24.852 1981 -7.601 -2.814 -0.999 -0.698 -1.874 20.013 1982 -7.136 -4.147 6.048 0.248 -0.940 17.907 1983 -7.554 -2.596 3.070 0.371 -0.398 -7.971 1984 -8.258 -1.224 2.923 -0.555 0.169 -3.628 Average 7.201 7.905 6.332 4.108 4.380 13.765 Table 8.21: RMS Errors (%), selected variables (1) Chapter 8. Simulation Results 159 Pi Pc Pi Pa P* Ce 1974 -5.424 -4.085 0.442 -4.033 -5.447 -0.115 1975 -1.408 0.096 2.475 0.488 -3.364 2.719 1976 -2.871 -0.131 0.038 -0.365 -4.231 0.951 1977 -3.033 0.888 -4.150 -0.862 -5.890 -2.753 1978 -1.652 0.929 -4.820 -1.841 -2.447 -3.141 1979 -2.152 2.147 -6.464 -1.773 0.759 -4.847 1980 -2.301 1.937 -7.494 -1.449 -2.326 -5.639 1981 -3.819 -1.496 -9.889 -3.939 -5.899 -7.135 1982 -5.931 -0.766 -10.952 -2.519 -9.972 -6.874 1983 -3.113 2.845 -9.352 0.157 -6.123 -4.893 1984 -0.613 5.358 -8.843 2.648 -0.391 -3.969 Average 3.321 2.456 6.929 2.234 5.022 4.461 Pe M e Igap BoT 1974 1.847 0.717 0.124 0.707 2.347 1975 5.460 -2.071 5.167 -1.725 12.461 1976 5.092 -3.047 13.643 -2.869 4.007 1977 4.542 -7.407 18.550 -8.026 11.316 1978 4.696 -10.248 18.480 -11.044 9.346 1979 -2.385 -7.477 16.351 -6.525 11.752 1980 -6.882 -8.044 13.035 -7.190 4.576 1981 -0.852 -8.976 10.083 -6.336 0.434 1982 5.760 -7.998 9.396 -6.032 5.662 1983 10.261 -5.713 13.042 -4.947 3.330 1984 17.843 -7.211 12.590 -6.555 11.226 Average 7.462 6.900 12.983 6.298 8.094 Table 8.22: RMS Errors (%), selected variables (2) Chapter 8. Simulation Results 160 equations reduced RMS errors, and so did using the unexpected cost variable, C 0 , instead of the residual gross profitability variable, PROF. Note that on purely econometric grounds the preferred equations had the partial adjustment form and used PROF rather than Cq. Using the political dummy variable Dcypr^ for 1974 in the output equation went a long way towards reducing RMS errors and generally improving the stability of the model in simulation. The opposite is true in the case of the output price equation: dropping the dummy and constant led to lower RMS errors for the entire model. Other important choices I had to make involved the specification of the export and consumption equations. The export equation, as mentioned in section 5.2, performed best econometrically with the inclusion of Igap, the inventory gap variable. Given the definition of inventory investment as a residual, however, and the large size of the coefficient on In Igap, I had anticipated stability problems in simulation. Such problems did not surface very clearly in the RMS runs, because the errors in the Igap variable were not very large. On the other hand, as we shall see later on, simulated shocks do push Igap substantially away from its historical values, with destabilizing effects on exports. On balance, from the point of view of stability in simulation, the export equation that does not include Igap seems preferable. As for consumption, there was no perceptible difference in the behaviour in simulation of the preferred consumption equation (habit-persistence with pc) and its first-difference variant. The RMS errors (%) of selected variables of the preferred model are given in Tables 8.21 and 8.22. The run starts in 1974, the beginning of the period that is most interesting and richest in change. Chapter 8. Simulation Results 161 8.2 Political History as Economic Shock: Cyprus Crisis and Coup 8.2.1 The Cyprus Crisis The events of the years 1974 and 1975, that include the Cyprus crisis, the threat of war and general mobilization that lasted until late 1975, plus the fall of the military regime of the colonels, enter my model through the political dummy variables D c y p r i (1974) and Dcypr (1974 and 1975). Both variables have large and significant coefficients in many equations throughout the model. Investment and employment appear to have felt the influence of the events most strongly, both for clear reasons: investment because of political uncertainty, employment because of general mobilization. Capacity utilization increased in 1974 to make up for the reduction in employment. Inflation also jumped by more than can be explained by domestic and imported costs, but the inclusion of the dummy variable DCypri results in larger RMS errors and I am inclined to drop it. Energy use also dropped by more than can be explained by the production structure equations, and thus Dcypr is also included in the energy demand equation. I use two different dummies of different durations mostly for reasons to do with prior knowledge about the duration of the shock. For example, with labour and conscription, general mobilization is known to have lasted approximately the same number of months in the two years, and the the length of the shock to investor confidence is expected to have been long-lasting. On the other hand, the war footing of the economy did not last beyond 1974, so its effect on capacity utilization is not expected to have continued into 1975. Statistical inference confirms these prior beliefs. Another way political instability very likely influenced economic reality was through a sizeable drop in capital inflows for the purpose of real estate investment, an important variable in the housing investment equation. Whether these channels are the only ones through which the Cyprus crisis affected Chapter 8. Simulation Results 162 the economy is of course debatable. Throughout the model specification and estimation process, I used a combination of prior knowledge of the economy and simple econometric tests to identify the equations and variables that are expected to have been affected and seem statistically to have been affected. The process used was not as systematic as it could have been, but the dummies were directly introduced into every behavioural equation, and retained when statistical significance combined with some prior reason for inclusion. In no case was a dummy rejected on entirely a priori grounds. As for exogenous variables, I avoided adjusting their paths unless there was good reason to believe that changes could be attributed to the political events. To assess the impact of these events, or, in other words, to quantify the effects of political instability, I simulated the model without the above mentioned dummies, and set the real estate capital inflows at trend levels. I then included the dummies again and reset the inflows at their historical levels, and re-ran the model. Thus history is simulated as a shock to the "counterfactual" situation where the 1974-5 events never took place. The results are quite interesting, though somewhat counterintuitive. If Dcypri is in-cluded in the output price equation (Tables 8.23 and 8.24), the first few years of the simulation are as expected. Investment, employment, energy use, the capital stock, nor-mal output, actual output, imports, exports, consumption, housing investment all drop in the first few years, while prices and wages increase. The situation turns around, however, and soon prices start falling and economic activity increasing again. Output, exports and housing investment rise above the counterf actual levels in 4 years, consumption and in-vestment in 5, normal output in 7, and the capital stock in 8. Prices start dropping in 4 years, and real wages never recover (at least in sample). The mechanism through which this starts happening is clearly unemployment and wages. According to my model as it stands, without the Cyprus crisis the unemployment rate would have been even Chapter 8. Simulation Results 163 Ine Kne E Qnv Q 1974 -18.961 -2.538 -1.987 -11.980 -2.074 -2.821 1975 -34.628 -6.612 -4.135 -15.187 -4.848 -8.325 1976 -31.752 -9.696 -4.737 -8.428 -6.357 -7.068 1977 -24.757 -11.352 -4.965 -10.341 -7.137 -3.928 1978 -11.554 -11.372 -4.349 -10.673 -6.791 0.864 1979 9.676 -9.618 -2.806 -9.311 -5.220 7.342 1980 38.829 -6.383 -0.860 -6.711 -2.885 14.532 1981 76.475 -2.188 1.529 -3.105 0.082 20.260 1982 117.119 2.504 4.072 1.239 3.351 22.345 1983 150.923 6.722 6.235 5.261 6.217 21.612 1984 172.155 10.635 7.736 9.061 8.547 19.600 Co Mne Xne C A h 1974 -5.431 -1.306 -3.920 -3.531 -5.514 -22.837 1975 -3.332 -5.816 -2.872 -5.315 -7.886 -15.119 1976 -5.968 -3.870 -1.562 -5.335 -7.223 -7.813 1977 -10.672 -1.866 -0.795 -3.836 -4.916 0.672 1978 -15.538 1.785 0.119 -0.911 -1.105 9.203 1979 -20.223 7.148 1.246 3.432 4.236 21.194 1980 -23.654 13.022 2.093 8.392 10.327 41.374 1981 -25.238 17.594 2.404 12.777 15.245 68.042 1982 -24.296 19.081 2.418 14.522 16.914 50.214 -21.586 18.269 2.302 15.911 17.208 40.515 1984 -18.323 16.462 2.074 15.789 16.235 33.486 Table 8.23: Cyprus Crisis Shock, % Deviations from Baseline, Dummy in pq (1) Chapter 8. Simulation Results 164 Pi Pc Pt Pa Px Ce 1974 6.359 4.347 0.123 5.100 3.386 0.920 1975 4.596 3.150 -2.677 3.691 2.456 -1.660 1976 . 2.457 1.690 -6.388 1.978 1.320 -4.757 1977 1.239 0.853 -9.967 0.998 0.667 -6.506 1978 -0.184 -0.127 -13.576 -0.148 -0.099 -8.533 1979 -1.891 -1.309 -17.226 -1.529 -1.026 -10.946 1980 -3.142 -2.179 -19.290 -2.543 -1.710 -12.698 1981 -3.596 -2.496 -19.953 -2.912 -1.959 -13.332 1982 -3.616 -2.510 -19.755 -2.928 -1.970 -13.359 1983 -3.447 -2.392 -19.125 -2.791 -1.877 -13.124 i984 -3.116 -2.161 -18.195 -2.521 -1.695 -12.661 Pe Me Igap M„ BoT 1974 0.788 -6.022 -2.074 -5.939 3.738 1975 0.518 -10.025 -16.634 -8.493 13.167 1976 0.339 -4.977 -19.352 -4.700 9.553 1977 0.175 -6.959 -22.242 -7.497 5.975 1978 -0.027 -6.830 -24.934 -7.322 -1.005 1979 -0.326 -5.543 -27.072 -4.872 -11.775 1980 -0.627 -4.238 -28.763 -3.806 -24.854 1981 -0.716 -2.108 -29.143 -1.497 -36.731 1982 -0.693 0.970 -29.731 0.730 -37.849 1983 -0.586 3.662 -30.766 3.156 -43.303 1984 -0.478 6.339 -32.059 5.730 -48.537 Table 8.24: Cyprus Crisis Shock, % Deviations from Baseline, Dummy in p q (2) Chapter 8. Simulation Results 165 lower than the already below-average historical levels, thereby accelerating wage infla-tion. However, increased unemployment in the aftermath of 1974-5 pushes wages down, which increases profitability, thus stimulating output, and international competitiveness, which improves the trade balance. Soon output, absorption, investment, employment and normal output pick up again and end up higher than if the Cyprus crisis had not taken place. Even unemployment turns around and ends up below the counterfactual level in 7 years. In the case where Dcypri is not included in the output price equation, the detrimental effects of the Cyprus crisis are even more muted and short-lived (Tables 8.25 and 8.26). Exports never fall, which keeps sales up, and consequently neither does output (except for a very small drop in 1975). Increased profitability (i.e. a lower Cq) from lower wages, as well as a higher utilization rate (due to the war economy) also contribute to strong output. The turnaround in investment comes after 4 years, and even unemployment drops under the counterfactual level after 5 years. These somewhat puzzling results can almost certainly be attributed to a flawed set of labour data. The OECD estimates of the civilian labour force that I use have been interpolated between the census year 1971 and the first independent observation, 1977, apparently without taking into account the 1974-75 mobilization, which clearly reduced the civilian labour force as well as employment. If the 1974-5 events had never happened, employment would have been higher, but the civilian labour force also would have been higher. Thus the direction of the counterfactual change in the unemployment rate is indeterminate. The average size of the standing armed forces in the years 1974 and 1975 is of course classified and can only be guessed at, and since the OECD estimate of employment is essentially a residual it almost certainly overestimates the number of people employed. Of all labour force variables, the number of the unemployed is the least suspect, and it rose in 1974 and 1975. However, even the number of the unemployed Chapter 8. Simulation Results 166 Kne E Qnv Q 1974 -17.737 -2.343 -2.075 -12.473 -2.152 2.503 1975 -30.817 -5.789 -3.575 -15.594 -4.346 -0.935 1976 -24.061 -7.845 -3.209 -8.150 -4.897 2.174 1977 -11.650 -8.214 -2.471 -9.012 -4.618 7.990 1978 8.291 -6.864 -1.090 -8.124 -3.298 14.486 1979 37.190 -3.739 1.206 -5.480 -0.760 20.407 1980 70.782 0.609 3.576 -1.828 2.233 25.967 1981 105.288 5.524 5.959 2.534 5.447 28.340 1982 132.962 10.562 8.075 7.339 8.562 27.355 1983 145.333 14.774 9.611 11.511 11.019 23.375 1984 141.478 18.389 10.314 15.191 12.724 19.142 Cq Mne C A h 1974 -6.087 2.271 0.603 1.524 -1.049 -1.840 1975 -5.841 -1.814 1.718 2.077 -1.615 5.470 1976 -9.122 0.126 2.901 3.821 0.683 14.423 1977 -14.932 4.413 4.105 7.575 5.110 25.823 1978 -19.431 9.303 4.988 11.969 10.253 32.683 1979 -21.618 13.923 5.273 16.579 15.534 38.075 1980 -22.531 18.696 5.145 20.505 20.374 47.714 1981 -21.322 21.010 4.801 22.730 22.671 52.126 1982 -18.153 20.495 4.361 21.967 21.842 30.620 1983 -13.137 17.305 3.870 20.802 19.635 18.038 1984 -8.153 13.923 3.316 18.453 16.764 5.017 Table 8.25: Cyprus Crisis Shock, % Deviations form Baseline, No Dummy in p q (1) Chapter 8. Simulation Results 167 Pi Pc Pt Pa Px Ce 1974 -0.922 -0.642 -2.036 -0.745 -0.499 -1.443 1975 -2.592 -1.809 -5.811 -2.097 -1.409 -4.037 1976 -4.312 -3.018 -9.826 -3.494 -2.353 -6.682 1977 -6.014 -4.220 -13.632 -4.881 -3.294 -9.272 1978 -7.229 -5.083 -16.120 -5.874 -3.972 -11.106 1979 -7.616 -5.359 -16.637 -6.191 -4.188 -11.687 1980 -7.443 -5.235 -15.891 -6.049 -4.091 -11.427 1981 -6.974 -4.902 -14.511 -5.666 -3.829 -10.723 1982 -6.369 -4.472 -12.879 -5.171 -3.492 -9.810 1983 -5.685 -3.988 -11.226 -4.613 -3.112 -8.774 1984 -4.904 -3.436 -9.498 -3.976 -2.679 -7.586 Pe M e Igap BoT 1974 -0.154 -6.252 -2.152 -6.166 -1.324 1975 -0.338 -10.248 -12.962 -8.686 7.061 1976 -0.601 -4.740 -12.446 -4.475 2.578 1977 -0.904 -5.959 -11.805 -6.425 -5.880 1978 -1.093 -5.067 -11.382 -5.439 -18.486 1979 -1.329 -3.126 -10.505 -2.739 -28.017 1980 -1.519 -1.032 -9.323 -0.923 -42.542 1981 -1.407 1.858 -7.731 1.305 -50.214 1982 -1.233 5.406 -6.067 4.024 -44.725 1983 -0.976 7.844 -5.018 6.722 -43.934 1984 -0.759 10.451 -3.812 9.412 -43.758 Table 8.26: Cyprus Crisis Shock, % Deviations form Baseline, No Dummy in pq (2) Chapter 8. Simulation Results 168 is arrived at by multiplying the number of the registered unemployed (which always underestimates total unemployment) by a correction factor. That correction factor was very likely upwardly biased in a period of general mobilization, when many of the non-registered unemployed were called up. In short, it is by no means clear that in the absence of the Cyprus crisis and change of regime the unemployment rate would have reached the improbably low levels that simulations suggest; if anything, the unemployment rate might very well have been higher. Without the unemployment-wage connection, the simulation results mentioned above would have been very different: wages would not have fallen as quickly and historical levels of economic activity would not have overtaken counterfactual ones as quickly, if at all. Unfortunately, with a labour data set as flawed as this, it is not possible to properly model what actually happened and what would have happened to employment and unemployment had the Cyprus crisis never occurred. In short, it seems that the political events of 1974-75 caused a drop in economic activity and the capital stock, a temporary surge in inflation, and a drop in real wages. Whether they were beneficial to the economy in the long run is doubtful, and in the absence of good labour force data the question must remain unanswered. 8.2.2 The Colonels The 1967 coup enters the model solely through the investment equation: investment clearly fell in 1967, and the political uncertainty before and after the coup is the leading candidate for explanation. The military regime is widely believed to have kept real wages low by repressing labour union activity, and to have at least attempted to attract foreign investment through tax and other concessions. There is some indirect evidence supporting the first assertion. Profit rates held their own around 24% between 1967 and 1974, the years of the military regime, while both before and after there has been a strong downward trend (see Figure 1.1), and real wage growth slowed down in a period when unemployment Chapter 8. Simulation Results 169 rates were dropping. There is no equivalent impressionistic evidence in support of the second assertion. I have been unable to model either of these two assertions. The effects of political repression of labour union activity are not confirmed by a significant dummy variable in the wage equation. One should not blame everything on bad labour data, but it may be the reason. Whether investment was stimulated through concessions is not clear either, since I have no data on these concessions (in order to see whether capital inflows were increased as a result), and the connection between capital inflows and total investment is statistically tenuous. The only other clear effect of the 1967 coup is a substantial drop in the level of capital inflows for real estate investment that, as in 1974-75, reduced housing investment. Thus I quantify the effects of the 1967 coup solely through these two connections, again running a "counterfactual" simulation without the Doj variable and housing capital inflows set at trend levels, and then running history as a shock. The results are given in Tables 8.27 and 8.28, and show that the model is relatively stable in simulation. The initial shock mostly affects investment and housing investment, and through the latter the capital stock and normal output. These effects are long-lasting; actual output and other indicators of economic activity do pick up eventually. Again the crucial mechanism seems to be employment-wages-profitability. The initial downturn causes extra unemployment, depresses wages, and increases profitability, which stimulates economic activity. The behaviour of all variables is clearly cyclical, with cycles taking from 6 to 11 years, a length of time not surprising given the slow speeds of adjustment throughout the model. Chapter 8. Simulation Results 170 Ine Kne Lne E Qnv Q 1967 -16.820 -2.307 0.002 -2.523 -0.860 -3.912 1968 -18.137 -4.652 -0.523 -4.929 -2.055 -3.505 1969 -17.973 -6.690 -0.953 -6.984 -3.074 -3.530 1970 -16.238 -8.061 -1.221 -8.374 -3.748 -2.851 1971 -12.492 -8.662 -1.134 -8.963 -3.910 -1.546 1972 -7.781 -8.552 -0.997 -8.869 -3.778 0.251 1973 -1.326 -7.651 -0.619 -8.008 -3.205 2.322 1974 6.691 -6.246 -0.078 -6.791 -2.366 4.711 1975 14.817 -4.691 0.608 -5.367 -1.378 5.745 1976 21.496 -2.746 1.237 -3.442 -0.274 6.090 1977 25.847 -0.639 1.681 -1.263 0.776 5.960 1978 27.413 1.484 1.934 1.043 1.715 5.114 1979 26.260 3.534 2.024 3.192 2.504 3.891 1980 22.450 5.139 1.935 4.817 3.007 2.313 1981 16.202 6.059 1.673 5.885 3.174 0.171 1982 8.395 6.249 1.216 6.307 2.966 -1.899 1983 -0.156 5.797 0.598 6.079 2.430 -4.098 1984 -8.452 4.849 -0.151 5.328 1.637 -5.823 Co Mne Xne C A h 1967 3.506 -3.909 -0.008 -1.117 -2.972 -12.917 1968 1.822 -3.512 0.025 -1.519 -2.864 -1.843 1969 0.682 -3.570 0.093 -1.689 -3.036 -1.469 1970 -0.958 -2.977 0.215 -1.431 -2.621 -0.283 1971 -2.688 -1.823 0.408 -0.792 -1.686 1.454 1972 -4.602 -0.262 0.676 0.288 -0.342 3.013 1973 -6.270 1.517 0.909 1.703 1.326 5.209 1974 -7.822 3.641 1.029 3.129 3.270 15.466 1975 -7.726 4.526 1.011 4.080 4.281 11.564 1976 -6.846 4.888 0.849 4.477 4.864 8.639 1977 -5.497 4.957 0.544 4.532 5.037 5.685 1978 -3.536 4.475 0.141 3.934 4.587 2.383 1979 -1.261 3.733 -0.332 3.009 3.843 -0.349 1980 1.156 2.692 -0.765 1.744 2.664 -4.000 1981 3.878 1.048 -1.105 0.153 0.916 -9.402 1982 6.177 -0.634 -1.356 -1.363 -0.827 -12.199 1983 8.285 -2.562 -1.522 -3.038 -2.756 -14.460 1984 9.574 -4.113 -1.626 -4.525 -4.359 -19.542 Table 8.27: Coup D'Etat Shock, % Deviations from Baseline (1) Chapter 8. Simulation Results 171 Pq Pc Pi Pa Px Ce 1967 0.012 0.008 0.004 0.010 0.007 0.019 1968 -0.038 -0.026 -0.089 -0.031 -0.021 -0.059 1969 -0.144 -0.099 -0.323 -0.116 -0.078 -0.223 1970 -0.331 -0.228 -0.763 -0.267 -0.179 -0.512 1971 -0.626 -0.433 -1.409 -0.506 -0.339 -0.969 1972 -1.034 -0.715 -2.321 -0.835 -0.560 -1.597 1973 -1.385 -0.958 -3.132 -1.119 -0.751 -2.138 1974 -1.566 -1.083 -3.552 -1.265 -0.849 -2.415 1975 -1.538 -1.064 -3.480 -1.243 -0.834 -2.373 1976 -1.295 -0.896 -2.845 -1.046 -0.701 -1.999 1977 -0.833 -0.575 -1.673 -0.672 -0.451 -1.287 1978 -0.217 -0.150 -0.174 -0.175 -0.117 -0.336 1979 0.514 0.355 1.525 0.415 0.277 0.797 1980 1.191 0.820 3.056 0.960 0.641 1.851 1981 1.728 1.189 4.245 1.392 0.930 2.690 1982 2.127 1.463 5.071 1.712 1.143 3.314 1983 2.393 1.646 5.571 1.926 1.286 3.732 1984 2.560 1.760 5.805 2.060 1.375 3.994 Pc M e Igap M „ BoT 1967 0.143 -1.364 -0.860 -1.896 7.218 1968 0.038 -2.765 -0.170 -3.619 6.010 1969 0.028 -3.959 -0.202 -5.627 6.272 1970 0.112 -4.748 -0.350 -7.147 5.677 1971 -0.059 -5.398 -0.968 -7.854 3.967 1972 -0.115 -4.902 -1.636 -6.887 1.335 1973 -0.093 -4.098 -2.460 -3.617 -1.945 1974 -0.225 -3.290 -3.479 -3.243 -4.877 1975 -0.192 -3.378 -4.147 -2.830 -7.957 1976 -0.173 -1.959 -4.242 -1.847 -9.881 1977 -0.119 -0.807 -4.088 -0.874 -10.293 1978 -0.030 0.644 -3.748 0.695 -10.859 1979 0.088 1.803 -3.216 1.570 -8.784 1980 0.231 2.924 -2.612 2.605 -7.959 1981 0.335 3.960 -1.925 2.763 -4.964 1982 0.399 4.510 -1.423 3.364 -1.431 1983 0.397 4.109 -1.084 3.539 2.431 1984 0.386 3.691 -0.797 3.344 5.996 Table 8.28: Coup D'Etat Shock, % Deviations from Baseline (2) Chapter 8. Simulation Results 172 8.3 Oil Price Shocks The next question of economic structure and economic history I dealt with by simula-tion was the impact of the two major crude oil price increases in 1974 and 1979. The counterfactual simulation only differs from history in having a series for crude prices that grows at a steady 4% per annum after 1973, and correspondingly lower prices for petroleum product net imports. History is then run as a shock, and the deviations of selected economic variables from their counterfactual paths reported in Tables 8.29 and 8.30. The results support the view that an adverse supply shock is equivalent to certain types of technical regress. Investment, the capital stock, normal output, actual output, consumption, non-energy imports and exports, all drop permanently. Despite a drop in the quantity of energy imports (energy use declines both because of the direct price effect and reduced economic activity), the increase in the price of crude is sufficient to cause a permanent deterioration in the trade balance. Employment increases slightly in the short run, since labour is a substitute to the capital-energy bundle, but soon drops again as a result of lower output. Prices are also permanently increased, but not by much, since energy is a small part of total costs, and real wages quickly drop in response to increased unemployment, never to recover. It must be pointed out that this simulation does not take into account the effects of the oil price shocks on the rest of the world. Given the structure of the export and price equations in particular, the adverse effects of the oil price shocks on world incomes and the price level of world non-energy exports must have hurt the Greek econ-omy in ways not captured in this simulation. Exports undoubtedly fell in response to falling world incomes (with an estimated elasticity of 1.6), and world inflation must have boosted domestic prices (with estimated elasticities of around 0.3 for both output and Chapter 8. Simulation Results 173 consumption-absorption prices). Thus the adverse effects of the oil price shocks on both the trade balance and prices must be underestimated by these simulation results. An obvious item for further research would be to link this model to a world model, and use counterfactual estimates of world incomes and prices in the absence of the oil price shocks in order to correct this omission. 8.4 Out-of-Sample Behaviour of the Model: Prediction Runs In order to conduct an out-of-sample simulation (or prediction run) of the model, I had to specify smooth paths for the evolution of exogenous variables. As a preliminary assumption I set the rate of growth of world real income at 2% per year, the rate of growth of world export prices at 4% per year, population growth at 0.5%, labour force growth at 0.7%, farm employment growth at -1% per year, agricultural output and energy use growth at trend levels etc. The exchange rate has to be endogenized for long-run out-of-sample simulation. I imposed PPP in the by pegging the exchange rate to the differential between domestic and international inflation rates. Clearly other assumptions could have been imposed (and some were in alternative runs), but the purpose of the exercise at this stage was not actual prediction, but an investigation of the out-of-sample properties of the model. One very clear result of out-of-sample simulations echoes results of the in-sample simulations reported in the previous two sections: the long-term properties of the model depend very strongly on the interaction of labour demand, unemployment, wages and profitability. One factor that contributed to long-term instability was the sizeable pos-itive constant in the error-correction labour demand equation.2 This coefficient implies that in equilibrium (i.e. when actual and desired employment are equal) non-agricultural 2It is +0.0185, compared to coefficients for the G-7 countries of 0 to -0.02. Chapter 8. Simulation Results 174 Ine Kne line E Qnv Q 1974 -3.353 -0.385 0.311 -3.410 -0.279 -2.740 1975 -7.160 -1.027 0.072 -5.835 -0.865 -3.494 1976 -10.757 -2.017 -0.228 -8.624 -1.605 -4.235 1977 -13.698 -3.240 -0.505 -11.391 -2.384 -4.869 1978 -15.312 -4.534 -0.655 -13.785 -3.056 -4.917 1979 -17.523 -6.030 -0.643 -17.222 -3.823 -5.964 1980 -21.876 -7.880 -0.513 -22.240 -4.862 -8.736 1981 -26.080 -9.907 -0.867 -26.378 -6.163 -9.873 1982 -28.516 -11.819 -1.405 -29.472 -7.398 -9.624 1983 -27.397 -13.065 -1.825 -30.504 -8.118 -7.138 1984 -22.846 -13.702 -1.650 -30.528 -8.188 -4.381 Mne C A Ih 1974 3.805 -2.261 -1.280 -1.740 -1.869 -8.775 1975 3.979 -2.535 -1.248 -2.489 -2.587 -6.897 1976 4.137 -2.758 -1.314 -3.097 -3.316 -6.591 1977 4.045 -3.371 -1.363 -3.654 -3.966 -5.987 1978 3.254 -3.441 -1.185 -3.728 -4.135 -4.370 1979 3.979 -4.405 -1.584 -4.474 -5.055 -5.952 1980 7.250 -6.809 -2.485 -6.255 -7.186 -11.516 1981 6.911 -7.512 -2.396 -7.261 -8.330 -13.494 1982 4.617 -7.071 -2.066 -7.119 -8.228 -9.339 1983 0.260 -4.911 -1.468 -5.961 -6.498 -0.708 1984 -3.148 -2.595 -1.106 -4.191 -4.271 12.090 Table 8.29: Oil Price Shocks, % Deviations from Baseline (1) Chapter 8. Simulation Results 175 Pi Pc Pt Pa Pt Ce 1974 2.006 1.391 0.745 1.616 1.079 3.166 1975 1.955 1.355 1.050 1.574 1.051 3.085 1976 2.059 1.427 1.056 1.658 1.107 3.250 1977 2.139 1.482 0.695 1.722 1.149 3.376 1978 1.854 1.286 -0.006 1.493 0.997 2.925 1979 2.492 1.726 -0.545 2.006 1.339 3.938 1980 3.955 2.733 -0.569 3.179 2.117 6.274 1981 3.810 2.633 -0.886 3.062 2.040 6.041 1982 3.271 2.263 -1.464 2.631 1.754 5.180 1983 2.307 1.598 -2.237 1.857 1.239 3.643 1984 1.730 1.199 -2.934 1.393 0.930 2.727 Pe M e Igap M c r BoT 1974 24.337 0.443 -0.279 -2.349 -24.727 1975 22.589 -1.622 0.104 -4.353 -18.239 1976 26.604 -3.865 -0.124 -6.361 -28.734 1977 29.304 -10.009 -0.642 -10.323 -18.900 1978 29.239 -11.583 -1.138 -11.616 -21.548 1979 43.199 -11.494 -1.894 -11.413 -29.073 1980 72.462 -12.161 -2.827 -16.675 -62.089 1981 74.927 -13.319 -3.469 -18.026 -59.098 1982 71.125 -22.069 -4.396 -22.499 -40.368 1983 53.934 -22.892 -5.168 -24.636 -44.539 1984 44.970 -21.443 -5.995 -25.557 -67.426 Table 8.30: Oil Price Shocks, % Deviations from Baseline (2) Chapter 8. Simulation Results 176 employment will grow at an annual rate of 2.13%. During the period of post-war urban-ization such a growth rate may have been possible, but with rural depopulation and women's participation slowing down and a very low birth rate, it is not tenable in the long term. In addition, the implied annual output growth rate is 3.94% (employment growth plus productivity growth), a rate that exceeds the average performance of the economy since 1974. As a result, allowing the constant to be as high in the long term causes actual employment to grow much faster than desired employment. This in turn reduces unemployment and increases wages, and unit costs rise faster than prices, in-creasing Cq and depressing investment. To deal with this long-term inconsistency I allowed the constant in the labour demand equation to gradually decline out of sample until it reaches 0.007, a figure more compatible with the long-run behaviour of the labour force. The long-term stability of the model was greatly improved. Since there is no labour supply equation, and the response of employment to labour market conditions is sluggish, care had to be taken to specify labour force growth rates that are consistent with the behaviour of other exogenous variables. The results on the same set of variables as above are reported in Tables 8.31 and 8.32. The simulation runs until 1999, and the last two years of the sample are also included for comparison purposes. Generally speaking the model proved relatively stable out of sample, with most vari-ables growing at a steady rate, but with troubling signs of explosive behaviour towards the end of the run. These signs are also present in the long-term policy shock simulations in the next section. Two variables that become unstable towards the end of the run are the inventory gap (Igap) and cost-to-price ratio (C0). The former increases out of control, and the export equation that contained it could not be used, a fact that was already clear from the RMS results. The latter is probably due to wages that rise more rapidly Chapter 8. Simulation Results 177 than prices plus productivity. Given the slow speed of adjustment of actual to desired employment, profitability declines (as shown by rising unit costs over output price, Cq), which depresses investment and output in later years. I attributed this to the size of the constant in the wage equation. Constraining the coefficients of current price inflation and lagged wage change in the wage equation to add to one results in a smaller constant and reduces the instability of Cq. In general, the equilibrating forces that dampen fluctuations in most economies are either weak or not represented in my model. The weakness of the equilibrating forces present in the model (most importantly the adjustment of actual to desired factors of production) makes fluctuations last longer. Many equilibrating forces, like interest rate and exchange rate effects and an endogenous labour force, are not present in the model at all, and their inclusion would add a lot more stability. In particular, a fixed nominal interest rate or a fixed real interest rate different from the Wicksellian "natural rate" can cause long-term explosive behaviour, especially in response to fiscal shocks. Interest rates should therefore be endogenous. For reasons already given, however, that will have to wait. The question of whether the model is intrinsically dynamically unstable, or whether the problem lies in misspecified exogenous variable paths, could be answered by lin-earizing the model and deriving its characteristic roots. This is a project that strongly suggests itself for further research. For the moment, given the suspicion of instability, the simulation results reported in this chapter should be interpreted with caution. Chapter 8. Simulation Results 178 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 34280.39 30541.40 30128.84 31432.66 33444.66 35774.60 38184.68 40759.37 43457.23 45959.76 47935.18 49051.55 48998.50 47573.17 44733.73 40668.63 35779.77 Kne 512577.1 523748.4 534150.2 545530.3 558563.0 573507.4 590375.8 609265.3 630228.6 652998.8 676987.7 701297.4 724752.7 746018.4 763765.5 776900.4 784769.4 2429.001 2443.070 2477.765 2524.963 2580.743 2642.638 2702.679 2762.085 2821.283 2880.134 2937.157 2990.808 3039.219 3080.561 3112.849 3134.452 3144.479 E 38234.76 39915.79 40612.36 41290.13 42090.98 43048.48 44163.29 45439.82 46879.23 48459.54 50135.66 51840.23 53485.81 54973.84 56206.50 57103.43 57616.93 Qnv 366555.1 374793.1 385077.3 397096.0 410670.4 425657.4 441260.6 4576442 474879.2 492890.0 511360.7 529872.5 547880.9 564786.4 579952.9 592828.2 603047.9 Q_ 332316.9 342081.3 366368.4 395626.8 426441.3 457542.3 489536.6 521973.4 554597.5 586003.1 615283.7 641239.1 662636.1 6785047 6884341 692847.7 6930942 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 1.171 1.169 1.133 1.121 1.112 1.112 1.107 1.106 1.108 1.121 1.140 1.167 1.202 1.247 1.302 1.365 1.435 108508.6 111118.0 118344.4 127982.2 137931.5 147971.0 158296.7 168762.8 179286.9 189414.6 198852.8 207214.4 214101.2 219200.5 222380.3 223778.3 223830.5 87521.7 100631.4 101547.1 103038.4 105072.3 107626.3 110686.9 114247.6 118307.8 122872.2 127950.0 133554.5 139703.1 146416.0 153717.5 161634.9 170198.7 331698.6 335359.6 367172.8 401034.6 434335.3 467336.7 500327.0 533314.3 566129.7 598111.4 628432.2 656134.3 680215.8 699816.5 714447.4 724192.8 729814.9 490787.1 493575.6 524668.6 567727.1 612667.4 658239.8 704031.6 750343.9 7968248 8422641 8849546 923193.9 955188.4 979440.1 995136.4 100253a0 1003131.0 20507.05 15614.66 15403.73 16070.33 17098.99 18290.20 19522.38 20838.72 22218.03 23497.48 24507.44 25078.20 25051.08 24322.36 22870.66 20792.33 18292.84 Table 8.31: Out-of-Sample Simulation, Levels (1) Chapter 8. Simulation Results 179 Pi Pc Pi Pa Px Ce 1983 7.044 6.102 690.724 6.561 6.385 7.541 1984 8.386 7.201 839.946 7.767 7.325 8.981 1985 9.894 8.447 1020.766 9.131 8.296 10.919 1986 11.025 9.363 1214.248 10.141 8.909 12.533 1987 12.056 10.191 1416.634 11.057 9.424 14.114 1988 12.970 10.918 1625.075 11.864 9.841 15.634 1989 13.764 11.542 1837.595 12.560 10.166 17.075 1990 14.443 12.071 2053.625 13.152 10.413 18.437 1991 15.029 12.522 2274.545 13.660 10.603 19.735 1992 15.554 12.923 2504.121 14.112 10.760 21.003 1993 16.057 13.308 2748.551 14.546 10.914 22.291 1994 16.581 13.712 3016.069 15.000 11.093 23.659 1995 17.163 14.165 3315.823 15.507 11.319 25.162 1996 17.825 14.685 3655.876 16.087 11.606 26.845 1997 18.566 15.270 4040.393 16.739 11.949 28.715 1998 19.354 15.894 4466.790 17.433 12.327 30.732 1999 20.126 16.507 4923.873 18.113 12.699 32.802 Pe M e Igap BoT 1983 9.151 6304.693 1.060 12079.10 -4557.506 1984 10.898 6364.688 1.128 11392.32 -3841.379 1985 12.857 6463.939 1.439 11553.83 -4791.273 1986 14.326 6555.396 1.427 11702.66 -5925.802 1987 15.666 6660.961 1.421 11874.45 -7111.954 1988 16.854 6796.595 1.421 12095.16 -8347.563 1989 17.884 6956.335 1.429 12355.11 -9654.623 1990 18.767 7140.262 1.439 12654.41 -11028.240 1991 19.529 7347.874 1.455 12992.25 -12466.430 1992 20.210 7575.694 1.478 13362.98 -13927.260 1993 20.864 7817.070 1.510 13755.77 -15375.610 1994 21.545 8062.280 1.555 14154.80 -16758.460 1995 22.301 8298.922 1.615 14539.89 -18011.170 1996 23.161 8513.391 1.696 14888.89 -19068.500 1997 24.124 8692.788 1.804 15180.82 -19877.870 1998 25.147 8827.544 1.948 15400.11 -20415.970 1999 26.150 8913.313 2.146 15539.68 -20700.350 Table 8.32: Out-of-Sample Simulation, Levels (2) Chapter 8. Simulation Results 180 8.5 The Long-Term Effects of Fiscal and Devaluation Shocks Having set up the out-of-sample paths of exogenous variables, I used the out-of-sample results listed in the previous section as control values for a fiscal and a devaluation shock, each spanning 25 years both in and out of sample. First I tried a 10% permanent increase in government spending on goods and services in 1970, and did a simulation run until 1994. The results are given in Tables 8.33 and 8.34. After the first 7 years only alternate years are reported to save space. The short-term effects are normal: a surge in spending stimulates output, employ-ment, profitability, investment and imports. Soon rising wages push costs up and prof-itability down, and investment declines, and so do all types of economic activity and employment. Higher unemployment eventually reduces wages and the whole cycle be-gins again. After the end of the sample, however, many variables, notably costs, prices and wages, profitability and investment, start exploding. As a test of the out-of-sample properties of my model, the fiscal shock simulation gives disappointing results. Within the sample, however, there is evidence of damped cyclical behaviour. Next I simulated a 10% one-off devaluation shock in 1974. I again ran a 25-year simulation to 1998, covering in and out of sample years. The results, given in Tables 8.35 and 8.36, were similar to the fiscal shock results, in the sense that cyclical behaviour in sample is followed by explosive behaviour in the last out-of-sample years. The initial effect of the devaluation is negative. Prices go up, while profitability, investment, absorption, output and imports drop. The transmission mechanism is through the effect of the devaluation on energy prices and directly on output, consumer and absorption prices. The negative effects of the devaluation are reversed through the same unemployment-wage-cost-profitability mechanism, and a cyclical behaviour ensues. Out-of-sample, however, after 1990 or so prices quickly get out of hand, leading to very rapid inflation and a Chapter 8. Simulation Results 181 /ne E Qnv Q 1970 2.753 0.360 0.000 0.378 0.130 3.496 1971 4.652 0.917 0.454 0.952 0.621 1.201 1972 4.164 1.329 0.577 1.384 0.848 0.707 1973 2.525 1.488 0.506 1.582 0.861 0.012 1974 -0.212 1.299 0.186 1.437 0.592 -0.924 1975 -2.220 0.986 0.022 1.174 0.382 -1.573 1976 -4.544 0.472 -0.238 0.694 0.038 -2.067 1978 -7.967 -0.923 -0.682 -0.722 -0.742 -2.506 1980 -8.521 -2.380 -0.919 -2.170 -1.409 -1.924 1982 -5.156 -3.026 -0.807 -2.940 -1.583 -0.131 1984 2.057 -2.663 -0.262 -2.758 -1.125 2.208 1986 12.318 -1.307 0.569 -1.536 -0.125 5.226 1988 25,675 1.465 1.758 1.137 1.612 8.655 1990 42.255 6.087 3.305 5.681 4.226 12.518 1992 60.150 12.741 5.170 12.273 7.723 15.563 1994 71.188 20.401 6.898 19.894 11.417 14.832 Cq M n e -"^ ne C A h 1970 -3.296 3.496 0.001 1.077 2.719 4.007 1971 -0.607 1.216 -0.039 0.883 1.032 0.607 1972 0.216 0.780 -0.143 0.546 0.709 -0.123 1973 1.081 0.197 -0.282 0.045 0.197 -1.002 1974 1.910 -0.603 -0.398 -0.545 -0.558 -4.160 1975 2.455 -1.122 -0.479 -1.091 -1.101 -4.040 1976 2.683 -1.526 -0.512 -1.514 -1.568 -3.984 1978 2.2U -1.976 -0.372 -1.958 -2.102 -2.539 1980 0.546 -1.709 0.023 -1.558 -1.832 -0.209 1982 -1.814 -0.395 0.420 -0.249 -0.410 4.861 1984 -3.910 1.516 0.716 1.546 1.545 11.653 1986 -5.658 4.338 0.716 3.343 4.336 22.879 1988 -7.287 7.738 0.716 5.496 7.521 37.491 1990 -8.613 11.569 0.716 7.743 11.174 55.630 1992 -8.908 14.588 0.716 8.824 14.151 75.208 1994 -6.704 13.863 0.716 6.657 13.698 87.284 Table 8.33: A 10% Fiscal Shock, % Deviations from Baseline (1) Chapter 8. Simulation Results 182 Pi Pc Pt Pa Px Ce 1970 -0.001 -0.001 0.000 -0.001 -0.001 -0.002 1971 0.060 0.042 0.131 0.048 0.032 0.094 1972 0.221 0.153 0.490 0.178 0.119 0.346 1973 0.436 0.303 0.996 0.352 0.235 0.685 1974 0.617 0.429 1.437 0.497 0.333 0.970 1975 0.743 0.516 1.753 0.599 0.401 1.168 1976 0.795 0.552 1.862 0.641 0.428 1.250 1978 0.576 0.400 1.243 0.464 0.311 0.905 1980 -0.035 -0.024 -0.228 -0.028 -0.019 -0.055 1982 -0.644 -0.449 -1.620 -0.520 -0.349 -1.009 1984 -1.094 -0.762 -2.546 -0.884 -0.592 -1.711 1986 -2.390 -2.062 -3.549 -2.182 -1.895 -2.999 1988 -2.478 -2.151 -3.634 -2.271 -1.984 -3.087 1990 0.282 0.618 -0.903 0.495 0.790 -0.344 1992 12.397 12.774 11.073 12.636 12.967 11.696 1994 48.557 49.056 46.812 48.873 49.310 47.631 Pe M e Igap M C T BoT 1970 -0.017 0.206 0.130 0.319 -6.217 1971 0.015 0.551 -1.181 0.824 -2.294 1972 0.054 0.732 -1.092 1.050 -1.612 1973 0.068 0.775 -0.914 0.680 -0.497 1974 0.149 0.666 -0.548 0.657 0.724 1975 0.168 0.715 -0.138 0.595 1.917 1976 0.176 0.380 0.069 0.358 2.997 1978 0.126 -0.455 0.141 -0.490 4.496 1980 -0.019 -1.349 -0.110 -1.206 4.344 1982 -0.166 -2.143 -0.468 -1.616 1.868 1984 -0.268 -1.938 -0.627 -1.762 -1.803 1986 -1.574 -0.997 -1.019 -0.909 -6.788 1988 -1.663 0.999 -1.701 0.913 -11.902 1990 1.119 4.242 -2.563 3.894 -17.457 1992 13.335 8.500 -3.551 7.840 -21.949 1994 49.796 11.986 -4.729 11.107 -21.441 Table 8.34: A 10% Fiscal Shock, % Deviations from Baseline (2) Chapter 8. Simulation Results 183 depression through the collapse of consumption spending (consumption reacts negatively to the rate of inflation). Eventually rising unemployment would reverse the cycle, but it appears that the cycles are explosive. Whether the fault lies with improper long-run specification of exogenous variables or with the structure of the model remains to be seen. The comments on instability mentioned in the previous section also apply here. Chapter 8. Simulation Results 184 Ine Kne Lne E Qnv Q 1974 -4.067 -0.453 0.159 -0.353 -0.045 -5.909 1975 -6.738 -1.011 -0.734 -0.857 -0.813 -2.819 1976 -6.931 -1.560 -1.057 -1.450 -1.220 -1.583 1977 -6.027 -1.977 -0.944 -1.883 -1.297 -1.411 1978 -3.692 -2.140 -0.562 -2.107 -1.115 -0.363 1979 -1.876 -2.114 -0.459 -2.160 -1.049 0.778 1980 2.136 -1.693 -0.195 -1.892 -0.749 2.731 1982 13.436 0.348 0.913 -0.134 0.654 5.672 1984 22.776 2.811 2.051 2.201 2.243 6.555 1986 32.203 5.787 3.048 5.105 3.921 9.140 1988 40.113 9.641 4.289 8.908 6.057 9.786 1990 40.292 13.570 5.106 12.813 7.927 8.040 1992 28.208 16.026 5.121 15.279 8.760 2.109 1994 0.631 14.968 3.489 14.270 7.326 -8.982 1996 -34.551 9.804 -0.439 9.178 2.998 -22.142 1998 -62.158 2.682 -6.111 2.123 -3.150 -32.232 c. M„e Xne C A h 1974 3.951 -6.834 2.606 -5.985 -5.547 -28.076 1975 2.277 -3.732 -0.147 -2.660 -2.342 -1.215 1976 0.301 -2.629 0.241 -1.648 -1.532 2.024 1977 0.161 -1.415 -0.080 -1.379 -1.350 0.584 1978 -0.986 -0.511 0.218 -0.586 -0.541 2.249 1979 -2.301 0.518 0.539 0.394 0.352 4.016 1980 -4.249 2.017 1.030 1.864 1.905 8.423 1982 -5.778 4.268 1.206 4.095 4.430 13.198 1984 -5.056 5.038 1.200 5.404 5.532 14.352 1986 -5.979 7.605 1.200 5.767 7.954 23.132 1988 -5.071 8.242 1.200 5.640 8.706 30.499 1990 -2.445 6.521 1.200 3.715 7.298 30.666 1992 3.095 0.672 1.200 -1.555 1.974 19.412 1994 13.304 -10.262 1.200 -10.753 -8.307 -6.273 1996 27.918 -23.237 1.199 -21.417 -20.801 -39.042 1998 40.099 -33.185 1.199 -30.280 -30.908 -64.754 Table 8.35: A 10% Devaluation Shock, % Deviations from Baseline (1) Chapter 8. Simulation Results Pa Pc Pt Pa Px Ce 1974 5.722 7.008 2.660 6.536 7.669 3.363 1975 0.227 0.157 1.530 0.183 0.122 0.356 1976 -0.370 -0.257 0.277 -0.299 -0.200 -0.580 1977 0.123 0.085 -0.768 0.099 0.066 0.193 1978 -0.335 -0.233 -1.770 -0.270 -0.181 -0.525 1979 -0.826 -0.575 -2.706 -0.667 -0.447 -1.292 1980 -1.568 -1.092 -3.533 -1.267 -0.850 -2.448 1982 -1.831 -1.277 -4.218 -1.480 -0.993 -2.858 1984 -1.822 -1.270 -3.921 -1.473 -0.988 -2.844 1986 -0.123 0.438 -2.103 0.233 0.725 -1.162 1988 5.374 5.966 3.291 5.749 6.269 4.277 1990 18.699 19.366 16.360 19.122 19.707 17.464 1992 50.198 51.041 47.246 50.732 51.473 48.635 1994 118.823 120.051 114.534 119.600 120.680 116.546 1996 234.134 236.007 227.606 235.319 236.970 230.659 1998 389.176 391.917 379.646 390.909 393.326 384.089 Pe Me Igap iWcr BoT 1974 6.127 -0.458 -0.045 -0.451 12.454 1975 -0.083 -0.770 -0.287 -0.641 7.678 1976 -0.164 -0.955 0.525 -0.899 6.167 1977 0.031 -1.216 1.005 -1.318 3.325 1978 -0.080 -1.287 1.043 -1.387 1.712 1979 -0.207 -1.228 0.757 -1.072 -0.373 1980 -0.398 -1.152 0.476 -1.030 -3.053 1982 -0.439 -0.057 0.389 -0.043 -7.035 1984 -0.427 1.611 0.600 1.464 -11.205 1986 1.296 3.628 0.620 3.307 -14.256 1988 6.871 6.048 0.430 5.530 -14.348 1990 20.384 8.045 0.518 7.386 -11.441 1992 52.328 7.951 1.163 7.334 -2.950 1994 121.923 3.760 2.673 3.484 12.776 1996 238.857 -4.188 6.009 -3.897 32.231 1998 396.077 -12.992 13.299 -12.118 49.252 Table 8.36: A 10% Devaluation Shock, % Deviations from Baseline (2) Chapter 9 Conclusion In this chapter I present a summary of my findings and a list of topics strongly suggested for further research on the basis of this thesis. 9.1 An Overview of the Model The model that I estimate is based on an approach that has "been used extensively to model the Canadian economy, and has also been applied to model cover the supply side of the OECD's G-7 country model. This approach places heavy emphasis on the supply block. At the heart of the model lies an explicit production function defining "normal output," i.e. output that is produced when factors of production are utilized at normal rates. Since normal output is not observable, it was assumed to hold on average over the 22-year sample. Thus many of the production function parameters were estimated as sample averages, and others by regression. The functional form I decided upon was nested-CES, with an adaptable vintage capital-energy bundle entering through a separa-ble inner subfunction. Other structure imposed on the production function was constant returns to scale, and a constant rate of Harrod-neutral (purely labour-augmenting) techni-cal change. This functional form was chosen as a suitable compromise between flexibility and structure. An attempt was made to estimate a translog production function (jointly with the behavioural capacity utilization equation) and to test the assumptions involved in the CES specification. The resulting estimates of parameter values were not plausible, and the tests and the alternative translog-based model have been relegated to Appendix 186 Chapter 9. Conclusion 187 D. In the process of estimating the production function parameters, I tested alternative hypotheses about technical change. Despite an observed slowdown in labour produc-tivity growth, the null hypothesis of constant technical change could not be rejected, and changes in factor utilization appear to account for all variations in observed labour productivity growth. Next a behavioural factor utilization (or actual output as a deviation from normal output) equation was estimated. The explanatory variables were unexpected sales, unex-pected costs (or inverse profitability) and the gap between desired and actual inventories. This factor utilization equation for output was tested extensively against alternative hy-potheses, including fixed factor utilization, a Keynesian equation, Lucas- and Barro-type new classical reduced form equations and a structure-free VAR equation. The results do not support the rejection of the null (variable factor utilization) hypothesis. Demand functions for labour and investment derived from the production function were also estimated. Desired levels of the factors of production, analytically derived from the production structure, entered partial adjustment and error-correction equations for labour and investment demand. The error-correction variants proved more stable in simulation. The same unexpected costs variable that enters the output equation was an important explanatory variable in the investment equation. Energy demand had already been estimated separately during the fixing of the parameters of the inner CES production function. Non-energy imports were a function of output and the relative price of imports. Two different equations for non-energy exports were estimated, each including a different supply-side variable: one the inventory gap, the other the price of exports relative to the domestic price of output. The former proved unstable in simulation, because of a large coefficient on an inventory gap variable that itself exhibits instability. Chapter 9. Conclusion 188 The (per capita) consumption equation was of the "habit-persistence" form, and included disposable income, lagged consumption and the inflation rate. In the absence of data on wealth, a life-cycle model proved impossible. Investment in residential construction was treated separately from other fixed invest-ment, and is described by a partial-adjustment equation (in per capita terms) depending on disposable income and domestic and foreign financing of housing construction. An energy block separates total energy demand into demands for the three main energy forms through interrelated quantity share equations. The price of petroleum products, the price of energy, and net energy imports are also endogenously determined. The balance of trade emerges as the difference between total (energy and non-energy) imports and exports. Taxes and government spending are modelled in a straightforward way. There is no monetary and international finance block, due to the chronic state of excess demand for money (dealt with by credit-rationing), extensive government intervention, lack of functioning capital markets, strict foreign exchange controls, and the relatively small size of foreign debt (until recently). Once the model was closed, I conducted simulations. RMS errors were relatively large but not explosive. I simulated three key shocks in the recent economic and political history of Greece. Two had to do with the colonels' military regime: the coup that ushered it in and the Cyprus crisis that ended it. The third was the rapid increase in crude oil prices in the 1970s. Out-of-sample simulations proved relatively unstable, especially in the presence of shocks. The results are sensitive to the specification of the paths of exogenous variables, so the endogenization or more careful specification of some of them would improve stability. Chapter 9. Conclusion 189 9.2 A Summary of Results A result permeating almost every estimated equation in my model is the very slow speed of adjustment of variables to their desired or equilibrium values. As a result, demand or supply shocks cause large and long-lasting deviations from trend (in sample) or explosive deviations from trend (out of sample). The estimated production function does not provide a clear answer to the question of whether productivity slowdown is secular or cyclical in nature, whether it is caused by low economic activity or is something happening independently. More data on post-1983 recovery (which is not yet complete) will shed more light on the question. For the moment, cyclical factors can satisfactorily account for the productivity slowdown. By all available evidence the production function does not hold at all times: the intensity of factor utilization does vary in response to unexpected variation in demand and cost variables. The role of unexpected (or unintended) inventory changes is not as clear. Hypothesis tests on output determination reveal no evidence of unexpected price or money stock effects; if fact, no statistical relation between the money stock (expected or unexpected) and output could be detected. Investment and employment equations show that relative factor prices do matter. Partial adjustment to cost-minimizing factor levels performs better than simple output-based factor demand equations. The speeds of adjustment for both capital and labour are substantially slower than the ones estimated by the same equations for industrial countries. Energy demand is estimated in the context of a vintage capital model, and the estimated retrofitting rate is comparable to the one of G-7 countries. Import and export equations show an income elasticity of export demand larger than the one for imports, which implies that the balance of (non-energy) trade improves through time, if real exchange rates stay constant and incomes grow at the same speed. Chapter 9. Conclusion 190 This is in fact borne out by the data: the non-energy balance of trade has been improv-ing, while the real exchange rate has been relatively constant. The recent balance of payments crises were in fact due to a collapse in unrequited transfers, and unrequited transfers had in fact been the balancing item for a chronic trade deficit. At the same time, my estimates show that the Greek economy is highly vulnerable to world recessions, since the latter strongly affect exports. The estimated consumption equation shows intertemporal consumption smoothing, and an adverse effect of inflation on consumption. Wealth effects could not be modelled directly because of a complete lack of data. In the price and wage block the vulnerability of the Greek economy to imported inflation, whether through world inflation or devaluation, becomes clear. Wages are influenced by the unemployment gap, and the degree of responsiveness implies a degree of real and nominal wage rigidity that is unremarkable by international standards: the real wage rigidity index is very close to the OECD average, and nominal wage rigidity is higher than average but not near the top. Thus it becomes hard to blame continuing stagflation on wage rigidity or "excessive" real wages. In any case, explicit measures of the full-employment or "warranted" wage depend critically on assumptions, so the wage gap is not very useful in deciding the extent of real wage unemployment. Simulations of the political events of 1967 and 1974-75 point towards the conclusion that in both cases investment and economic activity dropped, in the second case much more dramatically, and the 1974-75 crisis also fuelled inflation. The long-term effects of the latter are contingent upon the true effects of general mobilization on the labour force, a matter that may remain unresolved. Simulations of the oil-price shocks seem to support the view that an adverse supply shock is akin to certain types of technical regress. Output and investment drop, prices increase (although not to as large an extent as in industrial countries) and real wages Chapter 9. Conclusion 191 drop permanently. These simulation results have to be interpreted with caution, however, since there has not been enough feedback between simulation results and model structure, and the question of inherent model stability remains unresolved. 9.3 Some General Conclusions All in all, the results of this modelling exercise point to an economy where supply-side influences are very important. The level of normal output, as defined by factor inputs (labour, capital and energy) and technology, is shown to be a statistically significant determinant of output, in conjunction to cyclical demand and cost conditions, which also emerge as statistically significant. Thus the general approach that is based on the assumption of a variable rate of factor utilization reveals economic relationships that would be obscured or misspecified under previously used analytical frameworks. The same is the case with factor demands that are based on cost-minimizing factor levels: factor demands would be misspecified, and issues of factor substitution obscured, under the alternative assumption af simple output-based factor demands. The importance of supply factors also permeates the rest of the model, including price and wage equations. In Chapter 1 I argued that the fact of the Greek economy being on the fringe of industrialized nations endows the present modelling experiment with a broader interest. Similar hitherto unavailable insights could presumably be gained by applying the same modelling approach to other countries at a similar or lower level of development. The model presented in this thesis is characterized by an almost complete absence of any monetary and financial influences (whether domestic or international). I have argued that such an absence, in the specific case of Greece during my sample period, is less damaging than may be thought. This does not mean, however, that the monetary Chapter 9. Conclusion 192 and financial sides should be ignored in the future (since they are becoming much more relevant, as well as amenable to modelling), or in modelling exercises on countries at a similar or lower level of development. It is precisely the monetary and international finance sides that have been receiving prominence in past efforts to model such countries, however. Thus the kind of emphasis on supply-side influences that characterizes this thesis could be added to, rather than replace, existing modelling approaches. In any event, the increasing internationalization and integration of financial markets is gradually placing monetary and financial variables beyond the control of national policy making authorities, especially of small and less prosperous countries. Thus more emphasis on real, domestic, supply-side constraints may become desirable, or even inevitable. Despite these arguments in favour of the merits of a supply-oriented modelling ap-proach, there is one serious problem. The relative lack of readily available and reliable data imposed limitations on the scope of the model, and made the modelling task more difficult. Data problems are bound to be worse with countries less developed than Greece. Such problems may make application of the specific modelling approach on such countries prohibitively difficult. 9.4 Topics for Further Research Apart from the general goal of more emphasis on Greek socioeconomic reality, that would require changes in emphasis and modelling strategy, the following points indicate areas of further research that follow from the present work. An omission of this model that will become increasingly unacceptable as time goes by is the lack of proper modelling of monetary and international finance issues. Integra-tion with the European Community is forcing the liberalization of the banking system Chapter 9. Conclusion 193 and international trade and capital flows. Additionally, the recent series of foreign ex-change crises has increased the country's indebtedness to beyond negligible levels, and budget deficits have got out of control, with important consequences for domestic and international finance. Such issues I judged beyond the scope of this thesis, partly for institutional reasons, partly for the lack of data. They represent, however, important areas for further research. An issue of economic history that remains unresolved after my simulations is the long-term impact of the 1974 Cyprus crisis and especially the general mobilization that accompanied it. Given the serious flaws in labour data and the unknown numbers of people called up, the effect of the mobilization on the unemployment rate is uncertain. Another area for further research would be the correction of labour data for the mobi-lization years based on an educated guess of the numbers of young men involved. In the absence of truly independent observations of labour force and employment variables the undertaking would still be doubtful. A more detailed energy sector that endogenizes the domestic supply and prices of coal and electricity is an important prerequisite for proper analysis of issues of energy policy. Since the agency that decides on both is a state corporation under strict government control, more information about its motives, strategy and costs is needed than I have access to. Despite the broad macroeconomic impact of general energy issues covered by my model (for example, the macroeconomic effects of alternative petroleum tax policies can be easily simulated), more detail is necessary for the assessment of specific issues of energy policy, especially when it comes to the domestic supply of energy. My model can be used for simulations of more macro policy issues, like incomes policies and the effect of accession to the European Community (and the consequent trade liberalization and inflows of Community investment funds etc.). The data I have on customs duties and export subsidies almost certainly underrepresent the true extent Chapter 9. Conclusion 194 of trade protection, however, so the effects of accession can only be imperfectly captured at the qualitative rather than the quantitative level. Last, but by no means least, more work is necessary to make the simulations more believable and forecasting possible. In particular, the simulation behaviour of the model has only been used to help with the choice of specification to a limited extent. In addition, a question that needs to be addressed analytically (through the computation of characteristic roots) is the inherent dynamic stability of the model. Appendix A Data Sources, Problems and Methods All the data I use are annual. Quarterly data have only recently become available, and even then only for selected national account variables. Most are from 1963 to 1984, although national accounts and other data start earlier. However, reliable energy data do not exist before 1963, and the reliability of the national accounts themselves diminishes the further back we go. Thus the final data set is 22 observations (1963-1984). All organizations that I list as data sources are Greek (unless they are well-known to be international, or unless otherwise stated), and their names are given in official English translation. The output variable Q, that is modelled in the supply block, is not just value added GDP, but refers to output in the non-energy sector, including the user cost of energy input. Thus it is equal to GDP at factor cost plus net energy imports and net energy taxes. GDP is from the National Accounts (NA). Since I model residential construction separately for reasons given in section 7.3,1 subtracted income from housing (actual and imputed) from GDP. Given the problems I had with the inclusion of the agricultural sector in the supply block, I also subtracted agricultural output from GDP, and inputs into the agricultural sector from the series for capital, labour and energy inputs. Data on the quantities of energy imports come from the National Energy Council, the Ministry of Energy (National Petroleum Agency), the National Statistical Service of Greece (NSSG), the Public Power Corporation (PPC) and the International Energy Agency of the OECD. I used physical quantities (almost always in thousands of metric 195 Appendix A. Data Sources, Problems and Methods 196 tons, or lOPMT) for disaggregated figures for crude oil, petroleum products and coal, and Kilowatt-hours (kwh) for electricity. Prices of imported energy are a more difficult case. The price of crude until 1973 can be derived by dividing the value of crude imports by their quantity in 103MT, as they are given by the NSSG (which is the source of NA external trade data). After 1973, however, the quantities given by the National Petroleum Agency (NPA), the primary data source for flows of crude and petroleum products, do not agree with those of the NSSG. I have to face the problem of this discrepancy when I combine my energy demand equations with the Balance of Payments. There is no question that the NPA figures on energy flows are the more reliable of the two, and the discrepancy probably results from accounting practices. Thus I used a different method to calculate the price of crude: I combine OPEC data on the price of marker crude and average freight costs [83] with the 1963-1973 NSSG prices to construct a price series. The import price of petroleum products I derived as percentages of marker crude (arabian light) from prices ex Ras Tanura, the major Gulf port, quoted in the Petroleum Economist (various issues). I have no data on the import price of electricity, and my enquiries with the PPC have not been answered. Thus I use the annual production plus transmission (pre-distribution) cost of domestically produced electricity. Considering the relatively small amount of electricity imports involved, this should not distort things inordinately. Coal and coke import prices come from the NSSG and the Centre of Planning and Economic Research (CPER). Net energy taxes could not be just lifted from Ministry of Finance revenue statistics, because they are mixed with revenues from other sources or split across different taxes along administrative lines. Net petroleum taxes I calculated by applying the tax and duty rates on each petroleum product (crude was imported duty-free), as provided by the Appendix A. Data Sources, Problems and Methods 197 Ministry of Finance, and the quantities used. Net lignite, coal, coal gas and electricity taxes are negligible and could have been calculated by applying standard indirect tax rates (turnover tax and stamp tax) to different end uses of these energy forms. I do not attempt to do so for the moment. Total prices and quantities of energy used in the non-energy sector presented a prob-lem. Canadian studies (such as McRae [74], and the MACE models as described in [58] and [60]) can draw upon EMR estimates of the efficiency of each energy form in alterna-tive uses. I have no such estimates, however, so I followed the example of OECD studies (P. Artus [6]) and I aggregated using Divisia quantity indices. The aggregation was done 1. for all petroleum products, 2. for lignite, coal, coke and coal gas, and 3. for aggregates (1) and (2) and electricity. Of course only final use of energy (i.e. use outside the energy sector) was included. Thus a single quantity and price index for energy input was arrived at. Quantity data came from the same sources as for energy imports, and user prices from the Ministry of Trade for petroleum products, the PPC for electricity and lignite, and the NSSG and CPER for coal, coke and coal gas. Capital data have been prepared and published by the CPER in conjunction with NA (Skountzos and Matthaios [96]). They are disaggregated by sector, type and agent (i.e. public or private) and come in constant and current prices. They have been calculated via the perpetual inventory method from investment figures and by assuming depreciation rates for different types of fixed capital. I used their kickoff values and their average depreciation rates along with the applicable gross investment at constant prices to specify a capital stock equation. I calculate my own prices for capital, as outlined in Chapter 2. Appendix A. Data. Sources, Problems and Methods 198 The capital stock data that enter my production function do not include capital in the energy and agricultural sectors, nor the stock of residential capital stock. Labour data are a major problem. Before 1977 independent observations for labour force, unemployment, employment, agricultural employment etc. only exist for the years 1961 and 1971 (census years). Since 1977 there has been an annual labour force survey, and the data contained therein can be regarded as direct observations (despite a clear and admitted discontinuity in 1981). I (and other researchers) find it puzzling that the Greek government did not gather, analyze and publish data from the Social Security system. All directly employed workers in the private sector have been covered automatically since the founding of the Social Insurance Administration in the late 30s, and public employees, farmers and self-employed people have had coverage under their own state-sanctioned and often subsidized funds. However, the system is not yet computerized, and it would take a lifetime to disentangle labour data from it. The fact that there are still no reliable figures for the government's own employees (civil servants, contract workers and public enterprise employees) does not bode well for the prospects of such an undertaking. I thus had to fill in the missing years with OECD estimates.1 These estimates start with estimates of the labour force. These were interpolated between census years and the direct observations starting 1977, while taking into account the evolution of the three main sectors of the economy, net migration flows (negative during the time in question), information provided in government "Long Term Development Plans", and information provided by the monthly manufacturing establishment surveys. Then the number of the unemployed was estimated from the number of the registered unemployed, by applying a correction ratio derived by comparing the number of the unemployed as defined and reported by the census, and the number of the registered unemployed in the census years. It is a well-known fact that the number of the registered unemployed seriously *I am grateful to Mr. Christakos-Poulakis of the OECD for the following explanation. Appendix A. Data. Sources, Problems and Methods 199 underreports actual unemployment. Thus employment figures emerge as a residual, with the extra sources of information mentioned above modifying the estimates throughout. Clearly such a method of calculating employment as a residual is only an approxi-mation, and is very likely biased: as long as the supply of labour responds to the wage rate and employment conditions, these estimates will be countercyclical, and not too useful for the estimation of an accurate labour demand equation. This is especially true when labour demand is estimated in log-difference form, and my results show that the rates of change of employment are explained a lot less satisfactorily in the years between 1963 and 1976. In addition the likely countercyclical bias in the early observations may bias the estimated speed of adjustment downwards. These problems are dealt with in Chapter 4. Another serious problem that probably arises from the unreliability of the employment series is a translog production function that has unrealistic elasticities of substitution and own-price elasticities: its flexible functional form wraps itself around any data imperfections, and any direct inferences or hypothesis tests on the production structure derived from it cannot be trusted. Finally, a serious flaw of the labour data comes through in the simulation of the political events of 1974-75. Neither labour force nor employment figures reflect the fact of general mobilization, which leaves the issue of what the unemployment rate was and what it would have been in the absence of these events very much unsettled. Deriving a series on the average wage was not simple, since there is a large number of self-employed people, and there are no data on income from self-employment (it has always been lumped with other types of "income from property and entrepreneurship", including profits, dividends and rent). I first derived the number of wage and salary earners by dividing the total national wage and salary bill2 (from the NA) by an index 2It is typical of the frustrations involved with econometric work on the Greek economy that this national wage and salary bill is computed from Social Security data, which have never been used to compute the number of workers in direct employment! Appendix A. Data Sources, Problems and Methods 200 of manufacturing wages. This index is normalized and given a time trend so that the estimates for 1961,1971 and 1977-84 fit the observed number of wage and salary earners. If wse is this constructed series for wage and salary earners, and ws is the actual number in the said years, the rates of growth of wse and ws are extremely closely correlated. The next step is to credit the self-employed outside agriculture (who emerge as a residual from the method described above) with an implicit wage equal to the normal-ized industrial wage index described above, minus non-wage labour costs (mostly Social Security), that are part of the national wage bill reported by the NA. The weighted av-erage of the estimated employment and self-employment wages is the average wage that I use in my model. The implicit wages of farmers are not estimated; not only are their numbers wholly unreliable, but their average net income fell well short of the national average wage until very recently. Since agriculture is not part of my supply block and its output is taken as exogenous, labour input into the agricultural sector is also subtracted from the series I use in the supply block. The small part of the labour force employed in the energy sector is also subtracted. Unit value and quantity data for imports and exports are from the NSSG for imports and exports of goods. Imports and exports of services (tourism, shipping etc.), the exchange rate and capital flows are from the Bank of Greece (BoG). The IMF's financial statistics are the source of foreign magnitudes (world incomes, world export price indices), with the OECD used for purposes of comparison. Customs duties and subsidies are from the Ministry of Finance. To get non-energy imports and exports for goods I subtracted figures for SITC cat-egory 3 from the sum of categories 0 to 9. These NSSG figures for energy imports and exports, as mentioned above, do not agree with the figures I use, which are from primary energy flow sources. Thus my figures for total imports and exports will not agree with the NSSG ones (which are used by the National Accounts) when we come to the current Appendix A. Data Sources, Problems and Methods 201 account. Furthermore, BoP figures provided by the BoG do not agree with the NSSG and NA ones (the latter are on a physical movement through customs basis, the former on a payments basis), with the difference representing unrecorded changes in short-term com-mercial credit. The two differences are subsumed under the same residual discrepancy item. Data for the consumption and price equations and for the government sector came from the National Accounts. Data for the energy block came from the same sources as mentioned above for the supply block. Any data used in this thesis are available on request to interested researchers. Appendix B List of Variables and Parameters All real variables are in million 1970 Drs, all nominal quantities in million current Drs, and all prices are 1970-based (1970=1) unless otherwise stated. A dot Q above a variable denotes rate of growth (ln(x/x_i)). The subscript ^ denotes per capita variables, and the subscript _ t-,i = 1,2 etc. denotes the variable lagged 1, 2 etc. periods. Variable Equation No. Variables Description A BoT C Ce Co conv D, col cypr cypri C.2.3 Real Absorption C.6.1 Trade balance, in 1000's current US$ C.2.6 Real personal consumption C.3.1 Domestic unit cost index C.1.5 Production cost relative to output price Exogenous Factor for converting oil demand from Divisia index to physical units Exogenous Political dummy variable capturing the effects of the colonels' coup in 1967 Exogenous Political dummy variable capturing the effects of the Cyprus crisis and general mobilization in 1974-5 Exogenous Political dummy variable capturing the effects of the Cyprus crisis in 1974 d, oil C.5.8 Total domestic demand for petroleum products, 1000 MT 202 Appendix B. List of Variables and Parameters 203 D73 D80 D82 E Ev Fh G Gtr HL lagr igap Ine K* K: Exogenous Dummy variable capturing the effects of the speculative surge of imports in 1973 Exogenous Dummy variable capturing the effects of wage restraint in 1980 Exogenous Dummy variable capturing the effects of statutory wage increases in 1982 C.1.5 Energy quantity Divisia index, 1970 base C. 1.3 Vintage based energy requirement Exogenous Capital inflows for residential investment, in Drs, deflated by consumer price Exogenous Real government expenditure on goods & services Exogenous Government transfers to households (nominal) Exogenous Stock of government housing loans outstanding, deflated with consumer price Exogenous Real investment in agriculture Exogenous Real investment in energy sector Exogenous Real government investment C.1.7 Inventory gap (actual lagged over trended desired inven-tories) Exogenous Real residential housing investment C.2.7 Real inventory investment C.1.20 Real investment in the energy using sector, excluding agriculture and housing C.l. l Real re-investment with energy use malleable in the cur-rent year C. l . l l Desired real capital stock in the energy-using, non-agricultural sector C.1.12 Optimal bundle of capital and energy services Appendix B. List of Variables and Parameters 204 Kev C.1.2 Real vintage-based measure of capital and energy Kh C.2.8 Real housing capital stock Kinv C.1.21 Real stock of inventories, excluding farm and energy Kne C.1.16 Real fixed capital stock in the energy-using, non-agri-cultural sector, excluding housing NIA Exogenous Nominal net income from abroad Le Exogenous Total employment in the energy sector, 1000's Lne C.1.17 & 18 Total employment in the energy-using, non-agricultural sector, 1000's Lagr Exogenous Total employment in the agricultural sector, 1000's L* C.1.12 Desired level of employment in the energy-using, non-agricultural sector LF Exogenous Total civilian labour force, 1000's Mcoai Exogenous Real net imports of coal M„ C.5.10 Net imports of crude oil, 1000 MT Me C.5.11 Real net energy imports, Divisia index Meic Exogenous Real net electricity imports Mne C.2.1 Real imports of goods and services (excluding energy) Mrpp Exogenous Net imports of refined petroleum products, 1000 MT petragr Exogenous Oil use in agriculture, 1000 MT petre Exogenous Oil use in the energy-producing sector, 1000 MT Pop Exogenous Total population, 1000's PROF C.1.13 Gross profitability (gross profits as a residual over capital stock) Appendix B. List of Variables and Parameters 205 pa C.3.5 Implicit price of absorption pc C.3.3 Implicit consumption price index Pcoai Exogenous User price of coal P c r Exogenous Price of crude oil in Drs/MT pe C.5.6 Price of energy to final users petc Exogenous User price of electricity pjx Exogenous Price of foreign exchange, Drs per US$ pkc C.1.8 (1) Price of capital services, interest-invariant pkf C.1.8 (2) Price of capital services, interest-dependent pke C. l . l l Price of the capital-energy bundle pt C.3.4 Average gross yearly wage in the non-agricultural sector, 1000 Drs pme Exogenous Price of net energy imports, in Drs p m n e Exogenous Price of imported goods and services (excluding energy), in Drs pme Exogenous Price of net energy imports, in Drs Poil C.5.1 User price of oil, Divisia index Ponet C.5.1 User price of oil net of taxes pq C.3.2 Implicit price for gross domestic output (excluding en-ergy and agricultural sectors), at factor cost pqi C.3.3 Output price including indirect taxes pwx Exogenous Price index of world exports of goods, excluding energy, in Drs p w x g Exogenous Price index of world exports of goods, excluding energy, in US$ Pxne C.3.6 Price of exports of goods and services, excluding energy Appendix B. List of Variables and Parameters 206 Q C.1.9 Qagr Exogenous QH Exogenous Q* C.1.10 Qnv C.1.4 Qcr Exogenous RELP C.2.1 RELPX C.2.2 RL Exogenous Rcr C.5.10 ru C.1.19 Exogenous Ttcorp Exogenous rttt Exogenous rti Exogenous S C.1.6 s C.1.6 SC C.5.5 Scarp Exogenous SL C.5.4 SO C.5.3 Gross domestic output (excluding energy and agricul-tural sectors) Real output of agricultural sector Imputed real value of housing services Desired level of profitable future output for factor de-mands Real normal vintage-based output, CES Crude oil output, 1000 MT Three-year average, relative price of imports Three-year average, relative price of exports Crude oil refinery losses, 1000 MT Crude oil refined, 1000 MT Unemployment rate, percent Income tax rate Corporate tax rate Social security tax rate Indirect tax rate, excluding energy taxes Sales gap variable Real absorption plus exports Quantity share of coal and coal products Saving of corporations (nominal) Quantity share of electricity Quantity share of petroleum products Appendix B. List of Variables and Parameters 207 Exogenous Stocks of crude oil, 1000 MT STrpp Exogenous Stocks of refined petroleum products, 1000 MT T corp C.4.3 Corporate taxes (nominal) Tdp C.4.2 Total direct personal taxes (nominal) Te C.5.7 Taxes net of subsidies on energy (nominal) Ti C.4.1 Total indirect taxes net of subsidies (nominal) T C.4.2 Income taxes (nominal) T -*• SB C.4.2 Social security taxes (nominal) C.2.2 Real exports of goods and services (excluding energy) Y C.4.3 Income of corporations (nominal) Ydi» C.2.5 Personal disposable income (nominal) Ydpc C.2.4 Real per capita personal disposable income Yhha Exogenous Transfers to households from abroad Y *pera I.B.15 Personal pre-tax income (nominal) Y •* res Exogenous Residual error in National Accounts (real) P C.1.8 (1) Real average net return to capital, interest-invariant P' C.1.8 (2) Real average net return to capital, interest-sensitive Estimated Parameters Parameter Value Description 0 7 0.692 0.0401 Distribution parameter in inner CES function Distribution parameter in inner CES function Appendix B. List of Variables and Parameters 208 6 3.914 Depreciation rate (percent) for the capital stock in the energy-using sector, excluding housing 6h 1.667 Depreciation rate (percent) for the housing capital stock /i 0.4698 Distribution parameter in outer CES function v 0.3598 Distribution parameter in outer CES function ?r Labour productivity index for Harrod-neutral technical progress in CES production function. Estimated annual growth rate = 1.81% 0.55 Elasticity of substitution between energy and capital in the inner CES function 0.937 Elasticity of substitution between the energy-capital bun-dle and labour in the outer CES function 0.2 Annual rate at which energy-capital proportions become malleable in Kev Appendix C Estimated Equations and Identities C . l Supply Block (C.l.l) Vintage-Based Energy Requirement Ev = ( l - S - ip)Ev_x + Jnewfv, where new = J n e + tpK, and £1 = (lE^Y (C.l.2) Vintage Bundle of Capital and Energy Kev = ( l - S - V')i^e«_1 + /new j £ (C.l.3) Energy Demand InE = lnE„-0.11166 J D c y p r -0.15128 (6.69) (3.01) 2SLS, 1964-1984 R2 = 0.9979, s.e.e. = 0.02280, D-W = 1.623 F-test on constraint (lnEv - 1.0): 0.0147 (C.l.4) Vintage-Based Normal Output 209 Appendix C. Estimated Equations and Identities (C.1.5) Unit Cost Relative to Output Price ft Pkf ffnc+pt Lne+ye E U « ~ P«Q. (C.1.6) Sales Gap Variable S = < f ^ , where s = C + I + G + Xne- Qagr - Qh (C.1.7) Inventory Gap Variable T <K\nv IQnv~> Qnu (C.1.8) Capital Prices 1- Pke = (< 6 > +p)pa, where 0 _ <1~(P< Ln+ve E+<S>Kne va)/(Q Pg)^ r <(KP°)I(QP<,)> 2- Pkf = (< 6 > +p' + 0.72r)pa, where a l — <l-{PeLne+veE+«S>+0.72r)Knepa)/(QD9)^ f « * P » ) / ( Q p , ) > (C.1.9) Capacity Utilisation HQ/Qnv) = - 0.18396 lnC0+ 0.92035 In 5+ 0.02271 ln/ f l a p+ 0.02518 A (7.17) (24.94) (0.97) (2.70) 2SLS, 1964-1984 R2 = 0.9993, s.e.e. = 0.009325, D-W = 1.1761 F-test on all coefficients being zero = 228.28 (C.1.10) Desired Output for Factor Demands g - = ( i + i.5g -._ 1)fl=i±2=i Appendix C. Estimated Equations and Identities (C.l.ll) Desired Capital Stock "Mar1 Q* where pke = (/3<rpkcl~<T + I'Pe1^)^ (C.1.12) Desired Labour Input L ^ l ^ - f ^ y * , where e^ = (^ + ^ [ ^ ] T _ 1 ) ~ Q * (C:1.13) Investment-Partial Adjustment Ine/Kne = 0.00984 + 0.73850 (J„ e _ l /^„ e _ 1 )+ 0.04533 (tf* - tfne)/tf„, (1.38) (7.34) (2.77) + O.O1896P.R0F- 01922^- 0.02604£>cypr (2.70) (4.84) (7.90) 2SLS, 1965-1984 R2 = 0.9723, s.e.e. = 0.004313, D-h = 1.462 (PROF = qft-r."«F*-Eg«) (C.l.14) Investment-Error Correction \n{IneIIne_l) = - 0.16352+ 0.45510 l n ^ / i T ^ H 0.09183 ln(tf! 1/7„ (0.54) (0.81) (0.74) - 0.81018 \nC-q- 0.156221^- 0.24248Dcypr (2.62) (2.60) (4.24) 2SLS, 1965-1984 R2 = 0.7486, s.e.e. = 0.055599, D-W = 1.2607 (C.1.15) Total investment I = Ine + h + Ih + hgr + Ig Appendix C. Estimated Equations and Identities 212 (C.l. 16) Non-Energy Capital Stock Kne = Kne_1{l-6) + Ine (C.l. 17) Labour Demand-Loglinear lnLne= 0.36505+ 0.84838 In + 0.10600 lnl*+ 0.02610Dcw>r (3.52) (26.01) (4.09) (3.62) 2SLS 1965-1984 R2 = 0.9966, s.e.e. = 0.007966, D-h = -0.4283 F-test on constraint (InLne^ +lni* = 1.0) = 11.118, not imposed (C.l. 18) Labour Demand-Error Correction HLne/Lne.i) = 0.01848 + 0.13290 ln(I*/II1)+ 0.07088 \n{L*_JLne_J- 0.02163D, (6.86) (2.91) (2.24) (2.45) 2SLS, 1965-1984 R2 = 0.2908, s.e.e. = 0.009782, D-W = 1.7123 (C.l.19) Unemployment rate ru = 100(IF - Lne - L e - Lag) (C.l.20) Inventory Investment Iinv = Q + Qagr + Qh + (H - Tie)/Pq - A - Xne + Mne + NIA/Pq - Yre3 (C. 1.21) Inventory Stocks Kinv = Kinv—i + ^inv Appendix C. Estimated Equations and Identities C.2 Domestic and Foreign Spending (C.2.1) Non-Energy Imports l n M „ e = InQ- 0.77051 I n P R E L + 0.13034D73-1.1712 (10.43) (3.30) 2SLS, 1964-1984 R2 - 0.9920, s.e.e. = 0.03851, D-W = 0.899 F-Test on constraint ( I n Q = 1.0): 4.535) ( p ^ I = ELo(Er)t_,.) (C.2.2) Non-Energy Exports l n X n e = 3.2447 In Yw+ 1.4105 \n(Pxne/pq) - 3.4923 (13.07) (2.15) (3.21) 2SLS, 1964-1984 R2 = 0.9860 s.e.e. = 0.07207, D-W = 1.1938 (C.2.3) Absorption A=C+I+G (C.2-4) Personal Income Yper, = GDP + NIA - 8K-1 -Ti- Toorp - S^p + Gtr + Yhha (C.2.5) Disposable Income Yjia — YperB ; Tjp Appendix C. Estimated Equations and Identities 214 (C.2.6) Consumption Cpc= 2.0416+ 0.3407Ydit/(pcPop)+ 0.5449Cpc_l- 6.316pc (3.69) (2.68) (3.04) (2.82) 2SLS, 1964-1984 R2 = 0.9972, s.e.e. = 0.3296, T>-h = 1.2183 (C.2.7) Residential Housing Investment hpc/Khpc^ = 0.92877+ 0.11304 In Ydpc+ 0.026521 w(Fh/Pop) (6.03) (1.93) (2.28) +0.013634 ln(AHL/Pop) -0.21769 In Khpe_x (4.14) (5.48) OLS, 1964-1984, R2 = 0.9124, s.e.e. = 0.005796, D-W =1.6402 (C.2.8) Housing Capital Kh = Kh_l (l-8h) + h C.3 Prices and Wages (C.3.1) Normal Unit Cost of Output (C.3.2) Output Price p,= 0.64552Ce+ 0.35448jw+ 0.05574Dcypri - 0.00988 (9.28) (5.10) (3.53) (2.98) 3SLS, 1965-1984, R2 = 0.9638, s.e.e. = 0.014073, D-W = 1.8101 Wald x2-test on constraint (C e + pwx = 1-0): 3.151 Appendix C. Estimated Equations and Identities 215 (C.3.3) Consumption Price pc = 0.69016p„ + 0.30984pmnc (6.35) (2.85) 3SLS, 1965-1984 it"2 = 0.9641, s.e.e. = 0.013247, D-W = 2.298 Wald x2_test on constraints = 4.391 (Constraints: p,,- + pmne = 1-0, constant = 0.0) (Pqi = Pi (1 + rti)) (C.3.4) Wages Pt= 0.04948+ 0.41662?^+ 0.38767pc (5.65) (5.25) (6.57) - 0.01619 ln(ru/ < ru >)- 0.06817.D80+ 0.03523D82 (3.08) (5.85) (2.83) 3SLS, 1965-1984 R2 = 0.9438, s.e.e. = 0.011488 (C.3.5) Absorption Price pa = 0.78844p,+ 0.21155p mne (6.69) (1.79) 2SLS, 1965-1984 R2 = 0.9708, s.e.e. = 0.011717, D-W = 2.132 F-test on constraints = 0.930 (Constraints: pgt- + pmne = 1-0, constant = 0.0) (C.3.6) Export Price pxne = 0.54011pg+ 0.45989p„,I9+ 0.45989p/x-0.01578 (2.28) (1.94) (1.94) (2.86) 2SLS, 1965-1984 R2 = 0.9200, s.e.e. = 0.024656, D-W = 1.337 F-test on constraints = 1.156 Appendix C. Estimated Equations and Identities 216 (Constraints: pq + pwxg = 1.0,pwxg = Pfx) C.4 Government Finance (C.4-1) Indirect Taxes Ti = r t i Q P q (C.4-2) Direct Personal Taxes Tdp = Tst + Tinc., where T M = rtaa (Lne + Le)pt, and Tinc = rtinc Y^, (C.4-3) Corporate Taxes Tcorp — Ttcorp ^corp •> where Y^p = Qpq- (Lnept + Epe) C.5 Energy (C.5.1) Oil Price Poil = JWt (1 + rtp), where JWf = 0.65126 pcr+ 0.34874 p, (8.11) (4.34) 2SLS , 1965-1984 R2 = 0.8522, s.e.e. = 0.08214, D-W = 1.0837 F-test on constraints (constant=0.0, pc r + pq = 1.0): 1.031 Appendix C. Estimated Equations and Identities (C.5.2) Weighted Fuel Prices Pelc = E L l w* Pelc Pcoai = Ei=i wi Pcoai, where (C.5.5J Ot7 SWe 50= 0.64186- 0.03326 In 0.086427 In p e / c - 0.021034 In p^, (129.1) (4.62) (6.89) (2.78) 3SLS, 1965-1984 fl2 = 0.6462, s.e.e. = 0.015113, D-W = 0.5177 (C.5.4) Electricity Share SL= 0.32488+ 0.086427, In 0.12621 lnpe/c+ 0.039779 Inp^, (70.36) (6.89) (8.77) (6.02) 3SLS, 1965-1984 R2 = 0.7750, s.e.e. = 0.017302, D-W = 0.3696 (C.5.5) Coal Share SC = 0.03326- 0.021034,InPO./+ 0.039779 Inp e l c- 0.018745 Inpcoai (105.8) (2.78) (6.02) (2.84) 3SLS, 1965-1984 R2 = 0.6517, s.e.e. = 0.008643, D-W = 0.5976 (C.5.6) Energy Price Pe = Poil SO + pelc SE + Pcoai SC Appendix C. Estimated Equations and Identities (C.5.7) Energy Taxes Te = SO ' E • Ponet ftp (C.5.8) Total Domestic Demand for Petroleum Products doii = SO • E • conv + petragr + petre (C.5.9) Crude Refined Rcr = RL + don — Mrpp + A STrpp + misc. (C.5.10) Net Crude fmports Mcr = Rcr + A STcr - Qcr+ Non-Energy Use (C.5.11) Net Energy fmports M e = Mc*r + Mr*pp + Me/c + M^, where M^,. and M'pp have been changed from physical units to units such that their price 1.0 in 1970. C.6 Balance of Trade (C.6.1) Balance of Trade BoT = (Xne • p x n e — Mne • Pmne ~ Me • Pme)/Pfx Appendix D Translog-Based Results D . l The Production Function I chose to estimate the primal production function, and not its dual cost function, as is more commonly done in the literature (including several works on Greek manufacturing-Cavoulacos and Caramanis [27], Lianos [68] etc.-or the Greek economy-Apostolakis [5]) since the cost function does not allow deviations of output (due to variable factor uti-lization) from the underlying production structure. The first step in the translog specification estimation is to define the long-run produc-tion function determining normal output and to test for constant returns to scale.1 The constant returns to scale assumption would of course be useful for reducing the complex-ity of the analysis and for increasing the degrees of freedom by the imposition of many restrictions. The function, a three-factor translog production function with symmetry imposed and technology represented by a time trend T, was:2 lnQ„ = aQ + aKlnK+ aL\nL + aElnE + aTT+^'yKK(\nK)2 +7*rLm KIn 1 + I n Kin E + ^yLL(laL)2 + "fLEln Lin E (D.70) +\IEE{\TL E)2 + \ m T 2 + 6KTln K + SLT\n L + 6ETln E I then imposed constant returns to scale, for the same reasons that apply to the CES XI am indebted to Fisher, Chung and Helliwell ([40]) for the methodology used in this section. 2K and L in the following equations stand for Kne and £«e respectively, the inputs into the energy-using, non-agricultural sector, so that the notation can be simplified. 219 Appendix D. Translog-Based Results 220 form. I decided against testing for CRS because no testing procedure could give credible results, in light of such a flexible functional form applied to somewhat suspect data. For precisely the same reason, the translog equation could benefit from the imposition of some prior structure, and the resulting increase in ease of computation is substantial. The restrictions involved are the following: OK + OCL + CUE = 1 IKK + 1KL + IKE = 0 iKL-TlLL-TlLE = 0 (D-71) IKE + 1LE + 1EE = 0 SK + SL + SE = 0 Once CRS were imposed, the production function was logarithmically differentiated. Under the assumption of long-run cost-minimization, the following normal factor cost share equations hold: SKN = CtK -r^KK^K + 1KL\nL + ~fKEmE + 8 K T SLN = OCL + IKL InK + JLL In X + JLE \nE + 8LT (D.72) SEN = OCE + IKE InK + fLE In I + JEE InE + SET The production function is then jointly estimated with two of the three factor share equations (the third is dropped to avoid singularity), using iterative 3-Stage Least Squares with instrumental variables. The reason for joint estimation was problems of multi-collinearity in the single, unrestricted production function. Adding the share equations restrictions to the CRS restrictions can be expected to reduce multicollinearity, and it increases the degrees of freedom. The estimated parameters were used as starting values for the joint estimation of normal and actual output and factor cost shares. Actual output was defined as in equation Appendix D. Tia.nslog-Ba.sed Results 221 (2.2): In Q = In Qn + 0S In 5 + 0C In Cq + /?/ In Igap (D.73) The actual factor shares associated with actual output are of necessity different from the ones in equation (D.72), since output deviates from the production function and factors are quasi-fixed. They could have been defined similarly to the output equations as functions of normal shares and the same demand, cost and inventory gap variables, with different coefficients for each equation (since factor demands may not be influenced by these variables to the same extent). However, such a specification is ad hoc and has no real theoretical justification. The purpose of including the share equations is to reduce mmticollinearity and save on degrees of freedom, rather than to estimate actual share equations that are consistently derived from an optimization problem and accurately predict factor demands. Thus I used the normal share equations as approximations to the unspecified actual ones. The price paid in terms of lost explanatory power from not including the utilization variables (sales, cost and inventory) is actually negligible, while the iterative process of parameter estimation outlined below is considerably shortened. Each round of estimated production function coefficients yielded new normal output and normal shares estimates, which were used as the starting points for the next round of estimates. (We must keep in mind that two of the three explanatory variables in the actual output equation are denned in terms of normal output). The process was repeated until the parameter estimates converged. In this way the long-term technology, given by the production function, and the temporary response to demand and cost factors, were estimated simultaneously, yielding a behavioural output equation. I first estimated the constant factor utilization equations (equations D.70 and D.72) by iterative 3SLS with instrumental variables. Parameter values and summary statistics Appendix D. Translog-Based Results 222 are given in Tables D.37 and D.38 under CONST. IFU. I then used the parameter values from this estimation as starting values for the variable factor utilization model, described by equation (D.73). Each set of production function parameter estimates was used to define Qn in the next round of estimation. This iterative process was used until estimates of production function parameters converged. There were several choices involved in the definitions of the utilization variables 5, Cq and Igap. The data seem clearly to prefer a definition of Cq based on pkf, the interest-sensitive capital price, a definition more theoretically plausible and closer to the actual (inverse) gross profitability. Whatever the choice of capital price, however, the strong upward trend in Cq caused serious problems in estimation: parameter estimates con-verged slowly and the quadratic time trend tended to increase (in absolute value) out of control, with obvious consequences for stability in simulation. Detrending Cq solved these problems, and the detrended variable Cq is the one reported in the results. The parameter estimates are reported in Table D.37 and summary statistics of the output equation in Table D.38. The fit of the output equation is good, and in fact better than any of the CES variants. This was expected, given the ability of the translog function to wrap itself around the observations. One indication of trouble at this stage is the very high value of the test for normality of the residuals for both specifications. The political dummy variable DCypri was not significant and was not included. This result can very likely be attributed to the flexibility of the functional form. The next step in this estimation exercise was to calculate Allen-Uzawa elasticities of substitution and own-price elasticities for the three factors of production. These were derived from the Berndt-Christensen G-matrix3 of the production function. If G is this 3See Berndt and Christensen [14] for the definition. It is defined in terms of of the production function parameters and the factor shares. Appendix D. Tiajislog-Based Results CONST. IFU VAR. IFU ctq 3.977962 (6.38) 4.139646 (4.31) OCK -0.283896 (2.61) -0.298933 (4.06) CtL 0.970257 (1.56) 0.963882 (5.99) <*E 0.326500 (3.52) 0.335050 (7.05) IKK 0.145005 (5.43) 0.149506 (5.84) 1KL -0.087974 (3.04) -0.089287 (5.54) IKE -0.057922 (5.10) -0.060218 (5.04) ILL 0.032292 (0.35) 0.032641 (2.08) 1LE 0.054826 (3.02) 0.056647 (3.69) 1EE 0.003092 (0.22) 0.003572 (0.25) 6K -0.003298 (3.73) -0.003426 (3.76) SL -0.001836 (0.70) -0.001847 (3.68) 6E 0.005133 (6.04) 0.005263 (5.97) aj 0.005977 (2.91) 0.032442 (4-15) 7TT -0.003800 (11.0) -0.001710 (10.2) Table D.37: Translog Production Function: Coefficient Estimates Appendix D. Translog-Based Results 224 CONST. IFU VAR. IFU 1.00 (restr.) 1.00 (restr.) In 5 0.62523 (12.61) InC, -0.57930 (8.578) In Igap 0.01205 (0.761) R2 s.e.e. D-W X2-test on normality of residuals F-test 0.9945 0.024778 1.1687 12.19 0.9996 0.006943 1.3235 16.22 119.210 Restriction: In S = In Cq = In Igap = 0 Table D.38: Translog-Based Output Equations: Summary Statistics Appendix D. Translog-Based Results 225 matrix, the Allen-Uzawa elasticity of substitution between i and j is: G; CT.V = (D.74) is the i, j cofactor in G. Own price " " |G| where |G| is the determinant of G and JGjj elasticity of demand for i is: r/,- = S,o*,;, with i,j = K, L, E in all of the above. The existence of positive, inordinately large, or volatile own-price elasticities of sub-stitution was used as an extra criterion (in addition to the usual econometric criteria of R2, standard errors and t-ratios) in deciding which alternative specification to use in the cases of alternative variable definitions. For example, the choice of capital price or inventory gap specification affected elasticities as well as econometric performance. The Allen-Uzawa elasticities of substitution and own-price elasticities for selected years, as reported in Table D.39, deserve some comment. They are typically large in absolute value, which monotonically decreases with time in most cases. Given what we know about the general price-insensitivity in most markets in the Greek economy, this result is puzzling and counterintuitive. For example, the own-price elasticity of the demand for capital has values ranging from -12.8 to -3.7, depending on model and observation. The Allen-Uzawa elasticity of substitution between capital and energy is also very large, and, interestingly, positive (showing capital and energy to be substitutes, rather than complements as is assumed in the CES functional form), whereas <TLE is negative (thus labour and energy are shown to be complements).4 The result of this is that energy price increases cause an increase in desired capital large enough to cause an investment boom and explosive growth. This gives very counterintuitive and unstable 4 As pointed out by Blackorby and Russell [18], complementarity and substitutibility are here denned in the Berndt and Wood [16] sense of the response to energy prices of the demand for capital relative to labour: higher energy prices cause a drop in the demand for capital relative to labour if <JKE < 0 and <TLE > 0. Since the Allen-Uzawa elasticities of substitution have the same sign as the corresponding cross-price elasticities, this definition of complementarity has a straight-forward interpretation. Appendix D. Tra.nslog-Ba.sed Results 226 Year CLE VE CONST. IFU 1963 8.215 28.427 -5.544 -8.220 -1.407 -3.025 1966 8.035 26.845 -5.126 -8.031 -1.374 -2.842 1969 6.645 20.956 -3.943 -6.443 -1.203 -2.309 1972 5.926 17.735 -3.269 -5.703 -1.103 -2.096 1975 4.461 12.080 -2.192 -4.144 -0.906 -1.942 1978 4.770 12.242 -2.075 -4.402 -0.965 -1.607 1981 4.262 9.935 -1.578 -3.910 -0.870 -1.719 1984 4.037 8.718 -1.297 -3.702 -0.871 -1.377 VAR. IFU 1963 12.642 45.582 -9.476 -12.799 -2.084 -4.469 1966 12.253 42.637 -8.699 -12.412 -2.009 -4.139 1969 9.108 29.873 -6.030 -8.948 -1.585 -3.031 1972 7.723 24.006 -4.767 -7.531 -1.383 -2.617 1975 5.262 14.741 -2.905 -4.932 -1.038 -2.218 1978 5.726 13.940 -2.543 -4.954 -1.080 -1.823 1981 4.940 11.849 -1.884 -4.182 -0.956 -1.540 1984 4.606 10.210 -1.700 -4.264 -0.964 -1.505 Table D.39: TL Production Function: Allen-Uzawa and Own-Price Elasticities, Selected Years Appendix D. Translog-Based Results 227 results in simulation, both in and out of sample. These results, like all other results of the translog specification, as mentioned already, have to be interpreted with caution. In addition to the vulnerability of such a flexible functional form to unreliable data, these results depend crucially on the specification of, especially, the price of capital. Under the other two prices of capital, pkf and pkc», price elasticities are positive. The preferred specification of the price of capital leaves a lot to be desired, but it is the best of the alternatives tested. Another test undertaken in conjunction with calculating elasticities was a convexity test. The determinants of the bordered Hessian had the correct signs, so the estimated production function is convex in both cases. The improvement in the output equation that results from assuming variable fac-tor utilization and introducing demand and cost variables is evident from the summary statistics. The large values of the F-test statistics on the constant capacity utilization constraints, reported in Table D.38, confirm this. Thus the variable capacity utilization model is superior to the completely supply-determined, constant factor utilization alter-native. The same hypothesis tests as reported in Chapter 3 were performed, and the results were very similar to the CES-based ones. A troubling aspect of the estimation results is the very large x2_statistic on the normality of residuals test. Combined with elasticities that change dramatically with time, and Chow tests that imply unstable coefficients, this places serious doubt on the suitability of the translog specification for my data set or, for that matter, for the Greek economy. On the basis of the estimated translog production function I tested for neutral tech-nical change and different types of separability. Both neutral technical change and all separability combinations are rejected strongly. Technical change appears to have been energy-using, and the KE-L separability assumption underlying the CES functional form Appendix D. Translog-Based Results 228 F-statistics (w. 2 k 21 D.F) Restrictions: 7,„ = 7Jn = 0 for ij — n separability KL-E KE-L LE-K 13.182 15.041 16.021 Table D.40: Additive Separability Tests, TL VAR. IFU model is rejected. There are two alternative conditions for KL-E separability. It is shown in Blackorby, Primont and Russell [17] that separability will be additive (in the sense of a translog aggregator function in K and L which is additively separable from a translog function in E)if IKE =1LE = 0 (D.75) On the other hand, alternative, non-linear conditions for separability (equivalent to the weak separability of two log-linear aggregator functions, one in K and L and one in E) are given by: <*K _ IKK _ IKE _ 1KL _ $K_ <*L 1KL 1LE ILL $L Other combinations of separability are of course given by symmetric conditions. All cases of separability were rejected. In particular, the linear separability conditions were tested on the preferred model by a straight F-test on the restrictions (reported in Table D.40). Since the normal F-test (and the Wald x2-s*a.tistic) are unreliable for non-linear restrictions, the non-linear conditions (D.76) were tested by a likelihood ratio test on the unrestricted and restricted models, which were both estimated by nonlinear esti-mation (the x2-statistics, reported in Table D.41, are twice the difference between the log-likelihood functions of the unrestricted and restricted models). KE-L separability, Appendix D. Translog-Based Results 229 X2-statistics (w. 3 D.F), likelihood-ratio test Restrictions: ^ = ^ = ^ = = p- for ij — n separability KL-E KE-L LE-K 191.0564 13.2298 19.1904 Table D.41: Weak Separability Tests, TL VAR. IFU model on which the nested CES and CES-Cobb-Douglas functional forms depend, was rejected in both the linear and non-linear cases, although not as heavily as the other two. Given the unrealistic elasticities, non-normal residuals in the output equation, and unstable and counterintuitive simulation results associated with the translog form, how-ever, it is uncertain how much faith can be placed in these results. Desired factor levels have to be derived, to be used in the estimation of derived factor demands and the calculation of normal unit costs. This requires that the cost-minimizing factor inputs for producing expected output, Q*, and normal output, Q „ , be found. The translog production function, unlike the CES, is not self-dual, and this problem has no analytical solution, so I used a numerical algorithm (Quasi-Newton) to derive these desired factor levels. The equations used are two out of three cost-minimizing conditions ^/f = pi/Pj,iJ = K,L,E (D.77) and the production function. Unfortunately, around the saddle points these three equa-tions seem to be quite flat, and the results are sensitive to the equation dropped, the precise form of the equations, plus numerical accuracy, step length and number of itera-tions specified. Another difference between translog- and CES-based factor demands is that partial adjustment and error-correction equations have to be estimated for energy demand (the CES energy demand equation is estimated iteratively while fixing the parameters of the Appendix D. Translog-Based Results 230 inner production function). The speed of adjustment in the translog-based equations, which have no vintage structure, presumably picks up the vintage effects incorporated in the CES production structure. This should explain why even energy, the least fixed factor of production, still has a slow speed of adjustment. D.2 Translog-Based Alternative Estimated Equations In this section only the translog-based estimated equations that are different from the ones in the main model are reported, along with their summary statistics. The numbering follows that of Appendix C, with a D instead of a C in the equation number. Equations mentioned in the previous section are not listed. (D.1.9) Capacity Utilisation ln(Q/Qn) = - 0.57930 lnC,+ 0.62523 In 5+ 0.01205 In Igap (8.58) (12.61) (0.76) 2SLS, 1964-1984 R? = 0.9996, s.e.e. = 0.006943, D-W = 1.3235 F-test on all coefficients being zero = 119.21 (D.1.13) Investment-Partial Adjustment Ine/Kne = 0.15587 + 0.77783 ( / „ e _ l / t f n e _ l ) + 0.01896 {K* - Kne)/Kne (4.41) (9.10) (2.80) - 0 .12862C, - 016841?^- 0 .023491)^ (2.70) (3.33) (8.45) 2SLS, 1965-1984 R2 = 0.9712, s.e.e. = 0.004401, T>-h = 0.09036 Appendix D. Translog-Based Results 231 (D.1.14) Investment-Error Correction hx(Ine/Ine.l)= 0.03848+ 0.09059 ]n(K*/Kll)+ 0.01747 l n ^ / J ^ . J (0.30) (1.61) (0.26) + 0.20195 In PROF- 0.17851i9co/- 0.21995 (4.01) (3.42) (5.47) 2SLS, 1965-1984 R2 = 0.8042, s.e.e. = 0.049849, D-W = 1.6031 (D.1.17) Labour Demand-Loglinear lnLne = 0.28401+ 0.9286 In !;„._!+" 0.06231 lnl*+ 0.020771^ (2.36) (29.72) (2.61) (2.48) 2SLS 1965-1984 R2 = 0.9951, s.e.e. = 0.009549, D-h = -0.31524 F-test on constraint (In £„,,_! +lnX* = 1.0) = 4.838, not imposed (D.1.18) Labour Demand-Error Correction ln(WA.e-i) = 0.01923 + 0.07295 \n(L*/L*_1)+ 0.06726^(^1^1^)- 0.022511)^ (6.35) (1.72) (2.20) (2.21J 2SLS, 1965-1984 R2 = 0.0993, s.e.e. = 0.011023, D-W = 1.7911 (D.1.3) Energy Demand-Partial Adjustment lnE = 0.51586+ 0.80424 InE- X + 0.14889 InE*~ 0.078311^ (3.57) (29.15) (4.62) (2.50) 2SLS, 1964-1984 R2 = 0.9968, s.e.e. = 0.02568, D-h = 1.606 F-test on constraint (lnEv = 1.0): 11.223, not imposed (D.l.Sa) Energy Demand-Error Correction ]n(E/E-i) = 0.03915 + 0.182866 ln(E*/E*_1)+ 0.22964 ]n(E*/E.1)- 0.22964Dcypri (2.55) (4.15) (5.43) (1.67) 2SLS, 1964-1984 R2 = 0.7502, s.e.e. = 0.03284, D-W = 1.531 Appendix D. Translog-Based Results (D.3.2) Output Price pq = 0.76935Ce+ 0.23065JW- 0.01229 (10.78) (3.23) (3.99) 3SLS, 1965-1984, R2 = 0.9669, s.e.e. = 0.013447, D-W = 1.9310 Wald x2-test on constraint (Ce + pwx = 1.0): 0.221 (D.3.3) Consumption Price pc = 0.71104p„-+ 0.28896pmne (6.76) (2.75) 3SLS, 1965-1984 R2 = 0.9641, s.e.e. = 0.013248, D-W = 2.255 Wald x2-test on constraints = 4.893 (Constraints: pqi + pmne = 1-0, constant = 0.0) (Pgi = Pq (1 + rti)) (D.3.4) Wages pt= 0.04963+ 0.41198^+ 0.39545pc (5.69) (5.19) (6.64) - 0.01413 \n(ru/ < ru >)- 0.07104D80+ 0.036791>82 (2.68) (6.06) (2.97) 3SLS, 1965-1984 R2 = 0.9421, s.e.e. = 0.011664 Bibliography [1] Abel, A.B. (1984). "Optimal Investment Under Uncertainty," American Economic Review 73, 228-233. [2] Alexander, S.S. (1959). "Effects of a Devaluation: A Simplified Synthesis of Elas-ticities and Absorption Approaches," American Economic Review 49, 23-42. [3] Alogoskoufis, G.S. (1985). "Macroeconomic Policy and Aggregate Fluctuations in a Semi-Industrialised Economy," European Economic Review 29, 35-61. [4] Ando, A. and F. Modigliani (1963). "The 'Life-Cycle' Hypothesis of Saving: Ag-gregate Implications and Tests," American Economic Review 53(1), 55-84. [5] Apostolakis, B. (1983). "Money Balances as Factor Inputs: an Empirical Evidence Based on the Translog Specification," European Economic Review 23, 149-160. [6] Artus, P. (1983). "Capital, Energy and Labour Substitution: the Supply Block in OECD Medium-Term Models." OECD/ESD Working Paper No.2. [7] Artus, J.R. (1984). "The Disequilibrium Real Wage Rate Hypothesis," IMF Staff Papers 31(2), 249-302. [8] Avramidis, 0. et. al. (1984). "A Macroeconomic Model of the Greek Economy," CPER mimeo (provisional) [in Greek]. [9] Azariadis, C. (1975). "Implicit Contracts and Underemployment Equilibria," Jour-nal of Political Economy 83(6), 1183-1202. 233 Bibliography 234 [10] Baily, M.N. (1974). "Wages and Employment Under Uncertain Demand," Review of Economic Studies 41, 37-50. [11] Ball, R.J. (1973). "The Economic Models of Project LINK," in R.J. Ball (ed.) International Linkage of National Economic Models, Amsterdam: North Holland. [12] Barro, R.J. (1977). "Unanticipated Money Growth and Unemployment in the United States," American Economic .Review67(1), 101-115. [13] Barro, R.J. (1978). "Unanticipated Money, Output, and the Price Level in the United States," Journal of Political Economy 86(4), 549-580. [14] Berndt, E.R. and L.R. Christensen (1973). "The Translog Function and the Sub-stitution of Equipment, Structures and Labor in U.S. Manufacturing 1929-68," Journal of Econometrics 1, 81-114. [15] Berndt, E.R., G. May and G.C. Watkins (1980). "An Econometric Model of Alberta Electricity Demand," in W.C. Ziemba, S.L. Schwartz & E. Koenigsberg (eds.), En-ergy Policy Modelling: U.S. and Canadian Experiences (vol. I), Boston: Martinus Nijhoff. [16] Berndt, E.R. and D.O. Wood (1975). "Technology, Prices and the Demand for Energy," Review of Econ. and Stats. 57, 259-268. [17] Blackorby, C , D. Primont and R.R. Russell (1977). "On Testing Separability Re-strictions with Flexible Functional Forms," Journal of Econometrics 57, 259-268. [18] Blackorby, C. and R.R. Russell (1988). "Will the Real Elasticity of Substitution Please Stand Up? (A comparison of the Allen/Uzawa and Morishima Elasticities)," U.B.C. Discussion Paper No. 88-05. Bibliography 235 [19] Boskin, M. (1978). "Taxation, Saving and the Rate of Interest," Journal of Political Economy 86(2, part 2), S3-S27. [20] Brown, T.M. (1952). "Habit Persistence and Lags in Consumer Behaviour," Econo-metrica 20, 355-371. [21] Bruno, M. (1984). "Aggregate Supply and Demand Factors in OECD Unemploy-ment: an Update," Economica 53(210) Supplement, 35-52. [22] Bruno, M. and J.D. Sachs (1985). Economics of Worldwide Inflation, Cambridge, Mass.: Harvard Univ. Press. [23] Calvo, G. and E.S. Phelps (1977). "Employment-Contingent Wage Contracts," in K. Brunner and A. Metzler (eds.), Stabilization of the Domestic and International Economy, Carnegie-Rochester Conference Series in Public Policy, 160-168. [24] Caramanis, M.C. (1981). "Elasticity of Substitution Between Capital and Disag-gregated Labor: a Cost Function Application with Greek Data on the Level of the Firm," Greek Economic Review 3, 187-196. [25] Cardoso, E.A. and R. Dornbusch (1980). "Three Papers on Brazilian Trade and Payments," NBER Working Paper No. 541. [26] Carlino, G.A. (1982). "Interest Rate Effects and Intertemporal Consumption," Journal of Monetary Economics 9, 223-234 [27] Cavoulacos, P. and M. Caramanis (1982). "Energy and Other Factor Input Demand in Greek Manufacturing, 1963-1975," MIT Energy Lab. Working Paper 82-062WP. [28] Clark, P.K. (1979). "Investment in the 1970's: Theory, Performance, and Predic-tion," Brookings Papers in Economic Activity 1979:1, 73-113. Bibliography 236 [29] Cooley, F. and S.F. LeRoy (1985). "Atheoretical Macroeconometrics: A Critique," Journal of Monetary Economics 16, 283-308. [30] Crouch, C. (1983). "The Conditions for Trade Union Wage Restraint," in L. Lindberg and C. Maier (eds.) The Politics of Worldwide Stagflation, Washington: Brookings Institution. [31] Darby, M.R. et. al. (1983). The International Transmission of Worldwide Stagfla-tion, Chicago, 111.: Univ. of Chicago Press. [32] Darrat, A.F. (1985). "Unanticipated Inflation and Real Output: The Canadian Evidence," Canadian Journal of Economics 18, 146-155. [33] Davidson, J.E.H., D.F. Hendry, A. Srba and S. Yeo (1978). "Econometric Modelling of the Aggregate Time-Series Relationship Between Consumers' Expenditure and Income in the United Kingdom," Economic Journal 88(352), 661-692. [34] Demopoulos, D.G. (1981). Monetary Policy in the Open Economy of Greece, Athens: CPER. [35] Diavolitsis, T. and A. Balfoussia (1986). "Energy Demand 1973-85," CPER mimeo (provisional) [in Greek]. [36] Dimitriadou, C. and Kouzionis (1985). "An Econometric Model of the Greek Econ-omy with Emphasis on the Credit Market Under Disequilibrium Conditions (1959-1982)," CPER mimeo (provisional) [in Greek]. [37] Doan, T., R. Litterman and CA. Sims (1984). "Forecasting and Conditional Pro-jection Using Realistic Prior Distributions," Econometric Review 3, 1-100. [38] Economou, G.E. (1981) Prices and Incomes Policy in Greece: the International Experience and its Relevance to Greece, Athens: CPER [in Greek]. Bibliography 237 [39] Fisher, I. (1930). The Theory of Interest as Determined by Impatience to Spend Income and Opportunities to Invest It, New York: McMillan. [40] Fisher, T.C.G., A. Chung and J.F. Helliwell (1986). "A Translog Model of Ag-gregate Production and Variable Factor Utilization," U.B.C. Discussion Paper No. 86-41. [41] Friedman, M. (1957). A Theory of the Consumption Function, Princeton, NJ: Princeton Univ. Press. [42] Friedman, M. (1968). "The Role of Monetary Policy," American Economic Review 58, 1-17. [43] Fuss, M. and D. McFadden (1978). "Flexibility Versus Efficiency in Ex Ante Plant Design," in M. Fuss and D. McFadden (eds.), Production Economics: A Dual Approach to Theory and Applications, Amsterdam: North Holland. [44] Garganas, N.C. (1978). "A Macroeconomic Model for the Greek Economy," Project LINK meeting, Athens, October 1978, mimeo (provisional). [45] Glick, R. and C. Wihlborg (1985). "Price and Output Adjustment, Inventory Flex-ibility and Cost and Demand Disturbances," Canadian Journal of Economics 18, 566-573. [46] Goldstein, M. and K. Khan (1978). "The Supply and Demand for Exports: A Simultaneous Approach," Review of Economic and Statistics 60(2), 275-286. [47] Gould, J.P. (1968). "Adjustment Costs in the Theory of Investment," Review of Economic Studies 35, 47-55. [48] Grossman, S. and 0. Hart (1981). "Implicit Contracts, Moral Hazard and Unem-ployment," American Economic Review71(3), 301-307. Bibliography 238 [49] Grubb, D., R. Jackman and R. Layard (1982) "Causes of the Current Stagflation," Review of Economic Studies 49, 707-730. [50] Grubb, D., R. Jackman and R. Layard (1983) "Wage Rigidity and Unemployment in OECD Countries," European Economic Review2\, 11-39. [51] Haberler, G. (1949). "The Market for Foreign Exchange and the Stability of the Balance of Payments," Kyklos 3, 93-218. [52] Hall, R.E. (1980). "Employment Fluctuations and Price Rigidity," Brookings Pa-pers on Economic Activity 1980:1, 91-124. [53] Hayashi, F. (1982). "Tobin's Marginal and Average q: A Neoclassical Interpreta-tion," Econometrica 50, 213-224. [54] Helliwell, J.F. (1986). "Supply-Side Macro-Economics," Canadian Journal of Eco-nomics 19(4), 597-625. [55] Helliwell, J.F. (1988). "Comparative Macroeconomics of Stagflation," Journal of Economic Literature 26(1), 1-28. [56] Helliwell, J.F. and A. Chung (1986). "Aggregate Output with Variable Rates of Utilization of Employed Factors," Journal of Econometrics 33, 285-310. [57] Helliwell, J.F. and G. Glorieux (1970). "Forward-Looking Investment Behaviour," Review of Economic Studies 37(4), 499-516. [58] Helliwell, J.F. et. al. (1985). "Energy and the National Economy: an Overview of the MACE Model," in A.D. Scott (ed.) Progress in Natural Resource Economics, Oxford: Oxford Univ. Press. [59] Helliwell, J.F., P. Sturm, P. Jarrett and G. Salou (1986). "The Supply Side in the OECD's Macroeconomic Model," OECD Economic Studies 6, Spring 1986, 75-131. Bibliography 239 [60] Helliwell, J.F. et. al. (1987). "Supply-Oriented Macroeconomics: the MACE Model of Canada," Economic Modelling 4(3), 318-340. [61] Hicks, J.R. (1950). A Contribution to the Theory of the Trade Cycle, Oxford: Clarendon Press. [62] Jorgenson, D.W. (1963). "Capital Theory and Investment Behaviour," American Economic Review 53 (Papers & Proceedings), 247-259. [63] Kintis, A.A. (1973). The Demand for Labour in Greek Manufacturing, Athens: CPER. [64] Kintis, A.A. (1977). "Capital-Labour Substitution in a Developing Country: The Case of Greece—Comments and Some New Results," European Economic Review 9, 379-382. [65] Klau, F. and A. Mittelstadt (1986). "Labour Market Flexibility," OECD Economic Studies 6, Spring 1986, 7-45. [66] Koutsouvelis, P. and P. Karadeloglou (1986). "Proposals for the Construction of a Macroeconomic Model for the Greek Economy," CPER, mimeo (provisional) [in Greek]. [67] Kuznets, S. (1946). National Income: A Summary of Findings, New York: National Bureau of Economic Research. [68] Lianos, T.P. (1975). "Capital-Labour Substitution in a Developing Country: The Case of Greece," European Economic Review 6, 129-141. [69] Lianos, T.P. (1976). "Factor Augmentation in Greek Manufacturing," European Economic Review 8, 15-31. Bibliography 240 [70] Lucas, R.E. (1967). "Adjustment Costs and the Theory of Supply," Journal of Political Economy 75(4), 321-324. [71] Lucas, R.E. (1973). "Some International Evidence on Output-Inflation Tradeoffs," American Economic Review 63, 326-34 [72] Machlup, F. (1955). "Relative Prices and Aggregate Spending in the Analysis of Devaluation," American Economic Review 45(3), 255-78. [73] Malinvaud, E (1982). "Wages and Unemployment," Economic Journal 92(365), 1-12. [74] McRae, R.N. (1979). "Primary Energy Demand in Canada," Energy Economics, October 1979, 203-210. [75] METRIC (1981). Une Modelisation de I'Economie Franqaise, Paris: INSEE. [76] Metzler, L.A. (1949). "The Theory of International Trade," in H. Ellis (ed.), A Survey of Contemporary Economics, Philadelphia, 210-214. [77] Mishkin, F.S. (1982). "Does Anticipated Monetary Policy Matter? An Econometric Investigation," Journal of Political Economy 90(1), 22-51. [78] Mishkin, F.S. (1982). "Does Anticipated Demand Policy Matter? Further Econo-metric Results," American Economic Review72(4), 788-802. [79] Mohr, M.F. (1980). "The Long-Term Structure of Production, Factor Demand and Factor Productivity in U.S. Manufacturing Industries," in J.W. Kendrick and B.N. Vaccara (eds.) New Developments in Productivity Measurement and Analysis, NBER Studies in Income and Wealth, Vol. 44. [80] Nadiri, M.I. and S. Rosen (1973). A Disequilibrium Model of Demand for Factors of Production, New York: National Bureau of Econometric Research. Bibliography 241 [81] Nagatani, K. (1972). "Life Cycle Saving: Theory and Fact," American Economic Review 63, 344-353. [82] Nicolaou, K. (1978). Inter-size Efficiency Differentials in Greek Manufacturing, Athens: CPER. [83] OPEC (1982). Petroleum Product Prices and their Components in Selected Coun-tries, Geneva. [84] Phelps, E. (1972). Inflation Policy and Unemployment Theory, New York: Norton. [85] Prodromidis, K.P. (1976). Foreign Trade of Greece, Athens: CPER. [86] Prodromidis, K.P. (1982). "Macroeconomic Model, 1959-81," CPER, mimeo (pro-visional) [in Greek]. [87] Prodromidis, K.P. and J.N. Anastassakou (1983). The Determinants of Greece's Commodity Trade, Athens: CPER. [88] Prodromidis, K.P. and C. Dimitriadou-Kotsikou (1980). The Demand for Money in Greece: Economic Analysis with Quarterly Observations, Athens: CPER [in Greek]. [89] Rigas, K.A. (1985). "Energy Demand in Greece-Applications of Econometric Mod-els," paper presented at the Research &; Development in Energy Technology sym-posium, Ministry of Research & Technology, Athens [in Greek]. [90] Robinson, J. (1950). "The Foreign Exchanges," in H. Ellis and L.A. Metzler, (eds.), Readings in the Theory of International Trade, Homewood, HI.: Glencoe, 83-103. [91] Sachs, J.D. (1983). "Real Wages and Unemployment in the OECD Countries," Brookings Papers on Economic Activity 1983:1, 255-304 Bibliography 242 [92] Samouilidis, I.E. (1982). Analysis of the Energy Needs of the Greek Economy, Athens: CPER [in Greek]. [93] Sargent, T.J. (1976). "A Classical Macroeconometric Model for the United States," Journal of Political Economy 22, 135-167. [94] Sims, C A . (1980). "Macro-economics and Reality," Econometrica 48, 1-48. [95] Sims, C A . (1982). "Policy Analysis with Econometric Models," Brookings Papers on Economic Activity 1982:1,107-152. [96] Skountzos, T. and G. Matthaios (1984). "Net Fixed Capital Stock, Depreciation of Fixed Capital Stock, Gross Fixed Capital Formation, 1950-1983," CPER, mimeo. [97] Spanos, A. and T. Skordis (1984). "A Dynamic Macro Model of the Greek Econ-omy," CPER, mimeo (provisional). [98] Tobin, J. (1967). "Life Cycle Saving and Balanced Growth," in W. Fellner et. al. (eds.), Ten Economic Studies in the Tradition of frving Fisher, New York. [99] Treadway, A.B. (1969). "On Rational Entrepreneurial Behaviour and the Demand for Investment," Review of Economic Studies 36, 227-239. [100] Tsoris, N. (1976). Econometric Studies of Greece, Athens. [101] Tsoris, N. (1984). The Financing of Greek Manufacturing, Athens: CPER. [102] Vlachou, A.S. and I.E. Samouilidis (1986). "Interfuel Substitution: Results from Several Sectors of the Greek Economy," Energy Economics, January 1986, 39-45. [103] Yoshikawa, H. (1980). "On the 'q' Theory of Investment," American Economic Review 70(4), 739-743. 


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items