Essays on Macroeconomics by Xiaodan Gao B.A., Zhejiang University, 2006 M.A., The University of British Columbia, 2007 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in The Faculty of Graduate Studies (Economics) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) June 2013 c Xiaodan Gao 2013 ⃝ Abstract This dissertation studies two important topics in macroeconomics. The ﬁrst topic is on the corporate cash hoarding. The ﬁrst two chapters analyze the cash-inventory tradeoﬀ from two diﬀerent but complementary perspectives and shed light on the causes of cash hoarding. The second topic is international business cycles. A new feature of capital market is introduced into a standard international business cycle model to account for the disconnect between theory and data. The ﬁrst chapter proposes an explanation for the joint dynamics of cash and inventory — the adoption of the Just-in-Time (JIT) system. I start by demonstrating the importance of JIT in shaping corporate cash. I then develop a dynamic stochastic model to analyze the mechanisms and quantify their impacts. In the model, both cash and inventory can serve as working capital. As ﬁrms switch over from the traditional operating system (Justin-Case, JIC) to JIT, they allocate the resources freed up from inventory to cash, in order to ensure smooth transactions with suppliers. On average, this switchover accounts for 45% and 69% of the observed cash increase and inventory decline respectively. The second chapter provides a complementary explanation for the cashinventory joint dynamics. It models inventory as a reversible store of liquidity and studies the tradeoﬀ between cash and inventory when a ﬁrm manages its liquidity needs. I argue that two key determinants of a ﬁrm’s resource allocation decision are its market power and its exposure to risk. In the model, ﬁrms with lower market power and ﬁrms operating in riskier environments rely more heavily on cash rather than inventory. Model implications are supported by data. The third chapter studies the role of limited asset market participation (LAMP) in explaining international business cycles. We show that when LAMP is introduced into an otherwise standard model of international business cycles, the performance of the model improves signiﬁcantly, especially in matching cross-country correlations. To perform formal evaluation of the models we develop a novel statistical procedure that adapts the statistical framework of Vuong (1989) to DSGE models. Using this methodology, we ii Abstract show that the improvements brought out by LAMP are statistically signiﬁcant. iii Preface The third chapter of this dissertation is a joint work with two members of my committee: Viktoria Hnatkovska and Vadim Marmer. As part of the training and under the supervision of both, I was involved in each stage of the project: collecting and managing data, constructing and numerically solving models, implementing statistical procedures to compare model performance and writing a subsection of the manuscript. iv Table of Contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . v List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi Preface Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . xii Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv 1 Corporate Cash Hoarding: The Role of Just-in-Time tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . 1.2 Empirical Evidence . . . . . . . . . . . . . . . . . . . 1.2.1 Just-in-Time (JIT) Philosophy . . . . . . . . . 1.2.2 Cash and Inventory . . . . . . . . . . . . . . . 1.2.3 Cash and JIT . . . . . . . . . . . . . . . . . . 1.2.3.1 Sample Validation . . . . . . . . . . . 1.2.3.2 The Eﬀect of JIT on Cash . . . . . . 1.2.4 International Evidence . . . . . . . . . . . . . 1.2.5 Discussion . . . . . . . . . . . . . . . . . . . . 1.3 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1 Technology . . . . . . . . . . . . . . . . . . . . 1.3.1.1 Capital . . . . . . . . . . . . . . . . . 1.3.1.2 Inventory . . . . . . . . . . . . . . . 1.3.2 Financing . . . . . . . . . . . . . . . . . . . . . 1.3.3 Firm’s Problem . . . . . . . . . . . . . . . . . 1.3.3.1 Firm’s Problem under JIC . . . . . . Adop. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 6 6 7 10 11 14 16 18 18 19 20 20 21 22 22 v Table of Contents 1.4 1.5 1.6 1.3.3.2 Firm’s Problem under JIT . . . . . . . . . . 1.3.4 Optimal Policy Rules . . . . . . . . . . . . . . . . . . 1.3.4.1 Cash Holding under JIC . . . . . . . . . . . 1.3.4.2 Cash Holding under JIT . . . . . . . . . . . 1.3.4.3 New Purchase under JIC and JIT Systems . 1.3.4.4 Material Usage . . . . . . . . . . . . . . . . 1.3.4.5 Capital Investment . . . . . . . . . . . . . . Simpliﬁed Model with Analytical Solution . . . . . . . . . . 1.4.1 Steady State under JIC System . . . . . . . . . . . . 1.4.2 Steady State under JIT . . . . . . . . . . . . . . . . . 1.4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . Quantitative Analysis . . . . . . . . . . . . . . . . . . . . . . 1.5.1 Calibration . . . . . . . . . . . . . . . . . . . . . . . . 1.5.2 Quantifying Key Features of JIT . . . . . . . . . . . 1.5.2.1 No Lag in Delivery . . . . . . . . . . . . . . 1.5.2.2 Increased Inventory Adjustment Costs . . . 1.5.2.3 Long-term Commitment Contracts with Suppliers . . . . . . . . . . . . . . . . . . . . . . 1.5.2.4 Discussion . . . . . . . . . . . . . . . . . . . 1.5.3 JIT and the Rise in Corporate Cash . . . . . . . . . . 1.5.3.1 A Weighted-average Approach . . . . . . . . 1.5.3.2 Controlling for Self-selection . . . . . . . . . 1.5.4 Comparison with the Risk-based Explanation . . . . 1.5.5 Impacts of JIT on Firm Values . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Competition and Risk: When the Choice between Inventory Matters . . . . . . . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . 2.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Structure . . . . . . . . . . . . . . . . . . . 2.2.2 Firm’s Problem . . . . . . . . . . . . . . . 2.2.3 Optimal Policy Rules . . . . . . . . . . . . 2.2.3.1 Cash . . . . . . . . . . . . . . . . 2.2.3.2 Inventory . . . . . . . . . . . . . 2.2.4 Choice between Cash and Inventory . . . . 2.2.4.1 Deterministic Environment . . . . 2.2.4.2 Stochastic Environment . . . . . 2.2.5 Testable Hypothesis . . . . . . . . . . . . . 2.3 Estimation and Results . . . . . . . . . . . . . . . Cash and . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 24 24 25 26 27 27 28 28 29 31 32 33 36 36 37 38 38 39 39 41 42 44 44 46 46 51 51 52 53 53 54 55 55 57 61 61 vi Table of Contents 2.3.1 2.4 Methodology . . . . . . . . . . . . . . . . . . . . . 2.3.1.1 The Choice between Cash and Inventory 2.3.1.2 Liquidity Motive for Inventory Holdings 2.3.2 Data and Summary Statistics . . . . . . . . . . . 2.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . 2.3.3.1 The Choice between Cash and Inventory 2.3.3.2 Inventory and Financing Frictions . . . . 2.3.4 Robustness Checks . . . . . . . . . . . . . . . . . 2.3.5 Trends in Cash and Inventory Holdings . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 62 64 65 66 66 71 73 76 78 3 Limited Participation in International Business Cycle Models: A Formal Evaluation . . . . . . . . . . . . . . . . . . . . . 79 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 3.2 Model Economies . . . . . . . . . . . . . . . . . . . . . . . . 84 3.2.1 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 3.2.2 Households . . . . . . . . . . . . . . . . . . . . . . . . 86 3.2.2.1 Financial Autarky, FA . . . . . . . . . . . . 87 3.2.2.2 Bond Economy, BE . . . . . . . . . . . . . . 87 3.2.2.3 Complete Markets, CM . . . . . . . . . . . . 87 3.2.2.4 Equilibrium . . . . . . . . . . . . . . . . . . 88 3.2.3 Limited Asset Market Participation . . . . . . . . . . 88 3.2.4 Investment-speciﬁc Technology (IST) Shocks . . . . . 90 3.2.5 Deﬁnitions . . . . . . . . . . . . . . . . . . . . . . . . 90 3.3 Calibration and Model Solution . . . . . . . . . . . . . . . . 91 3.4 Econometric Methodology . . . . . . . . . . . . . . . . . . . 93 3.4.1 Pairwise Comparison . . . . . . . . . . . . . . . . . . 94 3.4.2 Comparison of Model Classes . . . . . . . . . . . . . . 97 3.5 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . 99 3.5.1 Benchmark Case . . . . . . . . . . . . . . . . . . . . . 99 3.5.2 Comparison Results . . . . . . . . . . . . . . . . . . . 104 3.5.3 Extensions . . . . . . . . . . . . . . . . . . . . . . . . 107 3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 Appendices A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 vii Table of Contents A.1 A.2 A.3 A.4 B . B.1 B.2 B.3 B.4 C . C.1 C.2 C.3 Variable Deﬁnitions . . . . . . . . . . . . . . . . . . . . . . Corporate Cash Holdings in Wholesale, Retail and Services Description of the JIT-adopter Sample . . . . . . . . . . . Work-in-process Inventory and JIT manufacturing . . . . . A.4.1 Steady State under JIC-manufacturing . . . . . . . A.4.2 Steady State under JIT-manufacturing . . . . . . . A.4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Proof of Proposition 5 Proof of Proposition 6 Proof of Proposition 7 Variable Deﬁnitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data Sources and Calculations . . . . . . . . . . . . Estimation Details . . . . . . . . . . . . . . . . . . . Derivation of the Asymptotic Variances Formulas in (3.12) . . . . . . . . . . . . . . . . . . . . . . . . . . C.4 Derivation of the Standard Error Formula in (3.16) . . . . . . . 120 120 123 124 126 127 128 . . . . . 130 130 131 132 134 . . . . . . . . . . . . . . . (3.10). . . . . . . . . . 136 136 137 . . . . . . . . . . . . . . . . . . . . 139 140 viii List of Tables 1.1 1.2 1.3 1.4 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . The Regression Results on Corporate Cash Holdings . . . . . Summary Statistics for 169 JIT Adopters . . . . . . . . . . . Eﬀects of JIT Adoption on Corporate Inventory and Cash (Annual) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5 Eﬀects of JIT on Corporate Inventory and Cash (Quarterly) . 1.6 Eﬀects of JIT on Corporate Inventory and Cash (Quarterly) . 1.7 Model Parameterizations . . . . . . . . . . . . . . . . . . . . . 1.8 Implications under the JIT System . . . . . . . . . . . . . . . 1.9 The Role of JIT in Explaining the Rise in Corporate Cash . . 1.10 JIT-based Explanation vs. Risk-based Explanation . . . . . . 1.11 JIT Adoption and Changes in Firm Value . . . . . . . . . . . 8 9 11 2.1 2.2 2.3 2.4 66 67 69 2.5 2.6 2.7 2.8 2.9 3.1 3.2 3.3 3.4 3.5 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . The Choice between Cash and Inventory Holdings . . . . . . OLS Regression of Cash and Inventory: by Mark-up . . . . . Pooled OLS Regression of Cash and Inventory: by Market Share . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Firm Fixed Eﬀect Regression of Cash and Inventory . . . . . Fama-MacBeth Regression of Cash and Inventory . . . . . . . Inventory For Constrained and Unconstrained Firms: by Firm Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Inventory For Constrained and Unconstrained Firms: by SA Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Robustness Checks: The Choice between Cash and Inventory Holdings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Benchmark Parameter Values without Estimation Step Estimated Productivity Process . . . . . . . . . . . . Volatilities: Benchmark Calibration . . . . . . . . . . Correlations with Output: Benchmark Calibration . . Cross-country Correlations: Benchmark Calibration . . . . . . . . . . . . . . . 12 13 15 35 37 40 42 45 70 71 72 73 74 76 . 92 . 93 . 100 . 101 . 102 ix List of Tables 3.6 3.7 3.8 3.9 3.10 Test Results from Benchmark Models Comparisons Volatilities: Robustness . . . . . . . . . . . . . . . Correlations with Output: Robustness . . . . . . . Cross-country Correlations: Robustness . . . . . . Comparison Results of Models with IST Shocks . . . . . . . . . . . . . . . . . . . . . 106 107 108 108 110 A.1 A.2 A.3 A.4 Cash Holdings in Wholesale and Retail . . . . . . . . . . Cash Holdings in Services . . . . . . . . . . . . . . . . . Descriptive Statistics for 169 JIT Adopters . . . . . . . Descriptive Statistics for 169 JIT Adopters (continued ) . . . . . . . . . . . . . 121 123 124 125 . . . . . . . . . C.1 Data Sources and Calculations . . . . . . . . . . . . . . . . . 136 x List of Figures 1.1 1.2 2.1 2.2 2.3 2.4 Average Cash U.S.. . . . . . Average Cash France. . . . . and Inventory Ratio from 1970 to 2011 in the . . . . . . . . . . . . . . . . . . . . . . . . . . . and Inventory Ratios in Japan, Germany and . . . . . . . . . . . . . . . . . . . . . . . . . . . Average Cash and Inventory Ratio from 1975 to 2009 in the U.S.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Eﬀects of Changes in Risk on the Optimal Cash and Inventory Holdings. . . . . . . . . . . . . . . . . . . . . . . . . Comparative Statics in a Stochastic Environment. . . . . . . Financial Constraints and Inventory Holdings. . . . . . . . . . 2 17 47 59 60 62 A.1 Average Cash and Inventory Ratios by Industries. . . . . . . 122 A.2 Relative Material Inventory Holdings after the Adoption of JIT manufacturing. . . . . . . . . . . . . . . . . . . . . . . . . 129 xi Acknowledgements I wish to express my deepest appreciation to all those standing by me during my years at UBC. Without their help and support, this dissertation would not have been possible. My most sincere thanks go to my supervisor Professor Henry Siu, who has always been there in those diﬃcult and trying times when I was working on my job market paper (the ﬁrst chapter). His unfailing enthusiasm and positive energy reboost my passion on research and keep my chin up. His continuous encouragement and support rebuild my self-conﬁdence, which is the most precious gift I received during my PhD. His constructive advice and feedback also greatly improve the quality of my work. In short, Henry is not only a great advisor but also a wonderful friend. The passion and conﬁdence he has passed on will continue to accompany me, which will deﬁnitely be of great help for my future career and of lasting inﬂuence on my life. The second person I owe a tremendous amount of gratitude to is my co-supervisor Professor Viktoria Hnatkovska. She fundamentally inﬂuences my views on macroeconomics, which is much richer, deeper and more interesting than simply deriving Euler equations and implementing computational algorithms. Her open-mindedness and unconditional support provide me with a high degree of freedom to choose my own research topics and help me discover my true research interests. I also appreciate her and Professor Vadim Marmer oﬀering the opportunity to work with them on a project (the third chapter) and receive rigorous training in writing papers. Viktoria is a great supervisor. I remember those days when she asked me to write Euler equations on the whiteboard in her oﬃce and helped me develop intuition behind those mathematical symbols. All the training prepares me better for the job market. I am indebted to her for her guidance, time and patience. I would also like to thank the other members of my committee. Professor Vadim Marmer is an extremely approachable mentor. I have immensely beneﬁted from his enormous help and support and from numerous discussions with him on various topics. I sincerely acknowledge his willingness to supervise me even when I barely knew nothing at the beginning of xii Acknowledgements this six-year long journey and his unlimited patience along the way. Particular thanks also go to Professor Michael Devereux, for squeezing time out of his busy schedule to be a part of my committee and for sharing insightful suggestions on my work. Most importantly, I thank him for believing me and allowing me to pass the dissertation prospectus with a half-ﬁnished paper, without which this dissertation would have been absolutely impossible. Another person I would like to extend my deepest and heartfelt thankyou to is Professor Hiroyuki Kasahara. Although not serving on my dissertation committee, Hiro is always available for questions. His thoughtful and detailed comments help to shape my research ideas and guide the work towards a more interesting and promising direction. Big thanks to my colleagues at UBC as well: especially, Haimin Zhang for providing delicious food when my research crowd outs time spent in cooking and for sharing the joy and dividing the sorrow over the course of the past six years; Andrew Hill for his irresistible sense of humour — often appropriate sarcastic responses and timely self-deprecation, and his kindness in helping me with empirical questions; Lori Timmins for comforting me whenever I was sad and for delivering happy hours with Britney Spears’ music; and David Freeman for his always serious and sincere suggestions whenever needed and for his help during the ﬁrst year of the program. Having awesome colleagues like them makes the oﬃce full of fun and working more enjoyable. I also wish to take this opportunity to thank my family. I thank my parents for their unconditional love and endless support, for trusting in me and all my decisions, and for tolerating my occasional mood swings and stubborn mind. I also greatly appreciate the warm hospitality from my grandparents, aunts, uncles and cousins. They make Vancouver my second home. Lastly, I would like to thank NUS for oﬀering me an exciting job and giving the whole six years a happy ending. The completion of this dissertation is a long journey, ﬁlled with frustration, depression, tears and joy. It witnesses my growing up and getting stronger. It brings with a sense of closure, but more importantly, a feeling of a brand new start with new hopes and dreams. xiii Dedication To my always encouraging parents; to myself; and to my half-gone youth. xiv Chapter 1 Corporate Cash Hoarding: The Role of Just-in-Time Adoption 1.1 Introduction The build-up of the cash reserve in the U.S. corporate sector has captured considerable attention from academic researchers, policy makers and ﬁnancial practitioners over the past few years, and has become one of the most hotly debated issues during the recent economic recession.1 It raises concerns about resource misallocation from high productivity assets (physical capital) to low productivity assets (cash) in the corporate sector. This paper aims to understand the causes behind the rise in corporate cash holdings. In this paper, I propose an explanation motivated by the simultaneous changes in cash and inventory in the data. As shown in Figure 1.1, for publicly traded U.S. ﬁrms, the average cash-to-asset ratio increased from 9.2% in the 1970s to 23.3% in 2011, and the inventory-to-asset ratio decreased from 24.3% to 10.1%. Despite the striking changes in both, the sum of these two ratios was relatively stable over the past thirty years. The substantial reduction in inventory is most commonly attributed to the widespread adoption of Just-in-Time (JIT) logistics since the early 1980s. Prior to the introduction of JIT, U.S. ﬁrms operated using the Just-in-Case (JIC) system which suggests that inventory be held for every possible eventuality. By contrast, JIT aims to eliminate buﬀer inventory in the production process. In light of the two facts described above: (i) the similar magnitude of the cash and inventory changes, and (ii) a signiﬁcant inventory reduction as a result of JIT adoption, this study investigates the role of JIT in explaining 1 See, for instance, “Companies’ cash piles: Show us the money”, The Economist, July 1, 2010. It states that “if cautious ﬁrms pile up more savings, the prospects for recovery are poor”. See also, “Corporate savings: dead money”, The Economist, November 3, 2012. 1 .1 .15 ratios .2 .25 .3 .35 1.1. Introduction 1970 1980 1990 year average cash ratio average cash and inventory ratio 2000 2010 average inventory ratio Figure 1.1: Average Cash and Inventory Ratio from 1970 to 2011 in the U.S.. The ﬁgure plots the average cash-to-asset ratio, average inventory-to-asset ratio and sum of those two ratios over time. The sample includes all Compustat ﬁrm-year observations from 1970 to 2011 with positive total assets and sales for U.S. manufacturing ﬁrms. Both cash and inventory ratios are winsorized between zero and one. the observed cash increase. I start by providing evidence that JIT implementation plays a role in shaping corporate cash policy. First, I use ﬁrm-level panel analysis to show that a one percentage point drop in inventory ratio is related to a 0.73 percentage point rise in cash, and that the change in inventory holdings is the most important factor associated with the (explained) cash increase. Second, with a sample of JIT adopters and their counterparts, I ﬁnd that a ﬁrm’s cash saving behavior in the post-adoption period is statistically diﬀerent from that in the pre-adoption period. Implementing JIT leads ﬁrms to progressively accumulate cash. On average, ﬁrms increase their cash holdings by over six percentage points after ten years of implementation. Lastly, I show that the same patterns as those in Figure 1 are also found in Japan, UK and France. After demonstrating the importance of JIT in understanding the increase in corporate cash holdings, I develop a model to explore the channels through which JIT implementation impacts cash and inventory. I then quantify their contributions. To keep the model tractable, I focus on input inventory only.2 2 When looking at the components of overall inventory, [25] ﬁnd that the signiﬁcant 2 1.1. Introduction In the model, a ﬁrm purchases material inputs for production. It holds cash to facilitate transactions with suppliers and holds inventory to economize on ﬁxed adjustment costs and to avoid stockouts.3 More speciﬁcally, each period, the ﬁrm makes decisions on inventory adjustment, material input use, ﬁxed capital investment, cash savings and dividend distributions. It faces productivity uncertainty and capital market frictions modelled by costly equity issuance, and is required to pay upon receipt of materials ordered. There are two operating systems with respect to inventory management — JIC and JIT. Under the JIC system, there is a lag between material orders and delivery. Although the ﬁrm adjusts its material input stock before production, new orders are unavailable for current period production. As a result, the ﬁrm adjusts inventory to anticipate future demand and carries inventory forward to avoid a stockout. By contrast, JIT allows the ﬁrm to respond contemporaneously to unexpected events. By adopting JIT, the ﬁrm adjusts inventory holdings with full information about the state of the economy and receives new purchases before production starts. Therefore, under the JIT system, the stockout motive for holding inventory is absent. How does JIT adoption inﬂuence a ﬁrm’s cash policy? Under JIT, the ﬁrm needs to pay for input purchases before current-period cash ﬂow is available. The ﬁrm therefore has only two channels to ﬁnance its input purchases: internal cash balance and costly external ﬁnancing.4 To avoid decline in inventory in U.S. publicly traded ﬁrms over the 1981-2000 period was mainly driven by declines in material inventory and work-in-process inventory. Accordingly, I model input inventory only for tractability. 3 In the model, I assume that both material inventory and work-in-progress inventory are inputs purchased from suppliers and JIT is narrowly deﬁned as JIT-purchasing. In reality, work-in-process inventory are produced within ﬁrm due to production inefﬁciency. However, my story can also be applied to work-in-progress inventory and my results remain unchanged, if I model material inventory and work-in-process inventory separately and consider JIT-manufacturing. This is because, as ﬁrms switch over from JIC-manufacturing (production takes time and ﬁrms have to ﬁrst transform materials into work-in-progress products and then transform those half-ﬁnished products into ﬁnished products) to JIT-manufacturing (ﬁrms can directly transform materials into ﬁnished products), they eliminate work-in-progress inventory but increase their material inventory. I illustrate and prove this statement in Appendix A.4. Therefore, implementing JITpurchasing is a must for implementing JIT-manufacturing, otherwise ﬁrms simply shift resources from work-in-process inventory to material inventory, and the aim of getting rid of inventory is unaccomplished. Once ﬁrms implement JIT-purchasing, all material inventory would be replaced by cash, as shown in this paper. 4 Here, costly external ﬁnancing can be debt ﬁnancing, equity ﬁnancing, line of credit from banks, or short-term credit from suppliers, although in my structural model costly external ﬁnancing takes the form of equity issuance. 3 1.1. Introduction raising expensive external funds, the ﬁrm preserves ﬁnancial ﬂexibility by building up its cash stock ([12]). In other words, to reduce costly inventory holdings (the objective of adopting JIT), the ﬁrm holds cash to fund its day-to-day operations.5 As a result, cash replaces inventory as the main component of a ﬁrm’s working capital. My model delivers a negative cash-inventory correlation of similar magnitude to that found in the data. It predicts that if all ﬁrms in the economy switch from JIC to JIT, the average cash ratio will rise by 10.4 percentage points, while the inventory ratio will decline by 11.6 percentage points. Taking into account the fact that around two-thirds of U.S. manufacturers have adopted JIT by 2008, I ﬁnd that the average cash and inventory ratios change by 6.9 and 7.1 percentage points respectively. That is, 45 percent of the observed cash increase and 69 percent of the inventory reduction are attributable to the JIT adoption. Results are quantitatively very similar after controlling for self-selection bias. Hence, almost half of the corporate cash increase can be rationalized by the adoption of JIT. Previous studies typically view the increase in ﬁrms’ idiosyncratic risk as the main driver behind the observed cash increase (see for example, [11]).6 I use my model to investigate whether this explanation is suﬃciently strong to quantitatively account for the trend. I ﬁnd that ﬁrms would raise their cash ratios by just 0.8 percentage points if risk doubled. This result indicates that the considerable increase in corporate cash is better explained by the transaction motive for cash saving stemming from JIT adoption, rather than the precautionary motive linked with the increased idiosyncratic risk. I also use my structural model to explore whether the adoption of JIT increases ﬁrm value, and to examine how ﬁrm characteristics generate heterogeneity in the beneﬁts from JIT adoption. These analyses respond to the mixed empirical evidence on the association between JIT adoption and ﬁrm performance.7 My results suggest that all ﬁrms beneﬁt from implementing JIT, and smaller ﬁrms gain more relative to the larger. On average, implementing JIT increases ﬁrm value by 47.4% in the long run. My work ﬁts into three broad streams of literature. First, it contributes 5 According to [78], taking into account service costs, storage costs and risk costs, inventory carrying costs are 19%-43% of total inventory. 6 The rise in the average ﬁrm-level volatility is well documented in the literature. See for example, [22], [31], and [56]. 7 [10] compare a sample of adopter and non-adopters over the period 1985 to 1989 and conclude that on average there is no signiﬁcant positive eﬀects of JIT on short-term return on assets. [54] and [60], on the contrary, ﬁnd that ﬁrms adopting JIT outperform their matched non-adopters, in terms of earnings per share and proﬁt margin. 4 1.1. Introduction to the cash literature by helping understand the reasons behind the signiﬁcant rise in corporate cash over the past thirty years.8 It explores the role of JIT and ﬁnds that it can explain over half of the trend. [11] also highlight the change in inventory as an important factor in understanding cash hoarding. However, their study does not explore the drivers behind the negative correlation between cash and inventory. This paper proposes a channel through which cash and inventory behave in a way that is consistent with the empirical evidence. Second, this paper complements recent cash studies by modelling cash as a source of working capital. There are a number of structural cash models focusing on the non-operational use of cash. In those studies, cash is modelled as a precautionary hedge against future uncertainty. In my model, cash serves two motives: non-operational use (precautionary savings) and operational use (working capital). When operating under the JIT system, ﬁrms hold cash not only to ﬁnance future capital investment, but also to purchase production inputs and facilitate operations. This is consistent with the survey evidence that a large portion of corporate cash savings is held for operational purposes (see [65]). Accordingly, it is of great importance to model operational cash. Lastly, this paper adds to the JIT literature by relating it with ﬁrms’ ﬁnancial policies. To the best of my knowledge, no previous work on JIT links it with cash management. This is despite abundant evidence that JIT is an eﬃcient approach to reduce inventory and therefore lower costs by freeing funds tied up in buﬀer stocks. How do ﬁrms allocate those released funds from inventory? My model suggests that ﬁrms choose to augment their cash stocks to maintain smooth operations. In addition, my work provides a structural framework to quantitatively evaluate the impact of JIT on ﬁrm performance. The advantages of this exercise relative to previous reduced-form empirical studies are twofold: (i) it can isolate other unobserved factors which possibly aﬀect ﬁrm performance along with JIT, and (ii) it can help understand whether ﬁrms with heterogenous characteristics beneﬁt diﬀerently from JIT implementation, particularly for ﬁrms of diﬀerent sizes. The remainder of the paper is organized as follows. Section 2 provides evidence that JIT implementation plays a role in shaping corporate cash policy. In Section 3, I develop a dynamic stochastic model in which a ﬁrm manages its cash, inventory and capital. Section 4 derives analytical solutions 8 There is a number of papers examining the cash hoarding behaviour of U.S. ﬁrms. An incomplete list includes [11], [71], [82], [5] and [19]. 5 1.2. Empirical Evidence of a simpliﬁed model to highlight the intuition behind the inventory-cash substitution. Section 5 describes the calibration of model parameters and presents simulation results to evaluate the role of JIT in explaining corporate cash hoarding and in improving ﬁrm performance. Section 6 concludes and discusses policy implications. 1.2 Empirical Evidence In this section, I use ﬁrm-level data to present empirical evidence regarding the eﬀect of JIT adoption on ﬁrms’ cash management. I start by showing the negative relationship between inventory and cash and estimating the fraction of the increased cash holdings associated with the reduced inventory. This in turn helps to infer the impact of implementing JIT on cash balance, given its role in eliminating inventory. I then focus on a sample of JIT adopters and non-adopters to directly investigate the diﬀerence between their preadoption and post-adoption cash saving behavior. At last, I show that similar patterns plotted in Figure 1 are also found in Japan, Germany and France. 1.2.1 Just-in-Time (JIT) Philosophy Before showing the evidence of the importance of JIT in explaining the rise in cash holdings, I give a brief introduction to the JIT philosophy. JIT is a philosophy of eﬃciency improvement, emphasizing the performance of activities based on immediate needs. Narrowly deﬁned, it strives to eliminate excess inventory resulting from overproduction and waiting. JIT philosophy can be applied to both the purchasing stage and the production stage. JIT purchasing involves the speedy delivery of materials from suppliers once they are ordered, and the requirement for purchasing comes from manufacturing process. JIT manufacturing involves the production of goods to meet current needs, rather than anticipate future demand. JIT purchasing is a must for ﬁrms that implement JIT manufacturing. A delay in material delivery will aﬀect the entire production process. JIT strategy was ﬁrst adopted by the Toyota manufacturing plants and then attracted a large number of followers in Japan by the mid 1970s. With Japanese manufacturing ﬁrms achieving high levels of international competitiveness in the early 1980s, JIT started capturing considerable attention in the U.S., and has been gradually adopted since then. Prior to the introduction of JIT, U.S. ﬁrms believed in Just-in-Case (JIC) philosophy. They 6 1.2. Empirical Evidence held buﬀer stocks at every stage in the production process in order to meet unexpected demand ﬂuctuations or production problems. 1.2.2 Cash and Inventory Stimulated by the time-series patterns of the cash and inventory ratios illustrated in Figure 1, the ﬁrst question I set out to answer is how important inventory is as an element linking with cash hoarding, after controlling for other factors that are usually taken into account to explain cash. To answer this question, I use the baseline cash regression in [11] and make three changes to it. First, I separate inventory holdings from net working capital, in order to explicitly gauge the importance of the former. Second, I replace industry level risk with ﬁrm speciﬁc risk and control for industry ﬁxed eﬀects, so that I can use within-industry variations to identify the eﬀect of risk on cash holdings. Lastly, I include cohort dummies which are constructed based on ﬁrms’ IPO listing dates as well as time dummies. The cohort ﬁxed eﬀects are motivated by the fact documented in [11] that most recent listed companies on average hold more cash than older cohorts, and the year ﬁxed eﬀects are used to capture the common macroeconomic shocks across ﬁrms. The cash regression is therefore speciﬁed as follows, cash = α0 + α1 f irm size + α2 risk + α3 inventory + α4′ X + + ∑ year + ∑ cohort + ϵ1 . ∑ industry (1.1) In this regression, cash is the ratio of cash and short-term investments to total asset; ﬁrm size is deﬁned as the natural logarithm of total asset; risk is computed as the standard deviation of annual operating cash ﬂow to total asset for the previous ﬁve years; and inventory is measured as the ratio of inventory to total asset. Other explanatory variables X include marketto-book ratio, ﬁrm’s operating cash ﬂows, working capital net of cash and inventory, capital investment, and so forth. A detailed description of these covariates is provided in Appendix A.1. The sample is constructed from Compustat Industrial Annual ﬁles, constituting an unbalanced panel of manufacturing ﬁrms (SIC 2000-3999) that covers 1980 to 2006.9 To control for the outliers in the sample, I delete ﬁrms 9 I use a pre-crisis sample to ensure that estimation results are not driven by the Great Recession. I also run the same cash regression for wholesale and retail trade (SIC 50005999) and services (7000-8999). Results are presented in Appendix A.2. 7 1.2. Empirical Evidence with negative total assets and negative sales, and winsorize continuous variables. Leverage, cash, and inventory ratios are winsorized between zero and one. R&D, acquisition and capital investment ratios are winsorized at the top and bottom 1%. Cash ﬂow ratio and net working capital are winsorized at the bottom 1%, and market-to-book ratio is winsorized at the top 1%. Table 1.1 reports descriptive statistics for these variables, which have similar characteristics to those in prior studies.10 Table 1.1: Summary Statistics Table 1.1 presents descriptive statistics for the variables used in the estimation, and reports the mean, median, standard deviation, 25th and 75th percentile, and number of observations of each variable for manufacturing. The sample is constructed from Compustat Annual Industrial ﬁles over the period 1980-2006. A detailed deﬁnition of variables is provided in Appendix A.1. Variables Mean Median Std. Dev. 25% 75% Obs. Cash 0.19 0.08 0.24 0.02 0.27 78055 Inventory 0.19 0.17 0.14 0.08 0.27 78006 Size 4.23 4.10 2.54 2.50 5.87 78055 Risk 0.11 0.04 0.22 0.02 0.10 53646 Market-to-Book 2.28 1.20 3.60 0.78 2.18 67494 Cash ﬂow -0.12 0.06 0.62 -0.06 0.12 78055 Net working capital -0.14 -0.04 0.50 -0.14 0.04 77288 Capital investment 0.06 0.04 0.06 0.02 0.07 77154 Leverage 0.26 0.21 0.25 0.05 0.38 77921 R&D 0.13 0.05 0.21 0.02 0.13 54216 Dividend dummy 0.31 0 0.46 0 1 78180 Acquisition 0.02 0 0.05 0 0 74801 Table 1.2 summarizes the estimation results of regression (1.1) and its alternative speciﬁcations. Column (1) reports the pooled OLS regression results controlling for 3-digit SIC industry ﬁxed eﬀects, year ﬁxed eﬀects and cohort ﬁxed eﬀects, whereas columns (2)-(4) re-estimate regression (1.1) with 4-digit SIC industry dummy variables, and ﬁrm ﬁxed eﬀects respectively. The variable of particular interest here is the inventory ratio. According to Column (1), a 1 percentage point decrease in inventory is correlated with a 0.69 percentage point increase in a ﬁrm’s cash holdings, which is statistically 10 See, for instance, [71]. 8 1.2. Empirical Evidence Table 1.2: The Regression Results on Corporate Cash Holdings Table 1.2 reports the estimation results of the cash regressions on ﬁrms’ characteristics, including ﬁrm size, risk, marketto-book ratio, cash ﬂow, inventory, and other commonly-included control variables. Industry, cohort and year ﬁxed eﬀects are included in the regressions. The heteroskedasticity-consistent standard errors reported in parenthesis account for possible correlation within a ﬁrm cluster. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1%, respectively. Column (5) shows the average changes for each explanatory variable and Column (6) reports the predicted changes in cash holdings by each explanatory variable, according to the regression estimates in Column (4). Variables Inventory Size Market-to-book Risk Cash ﬂow Net working capital Capital investment Leverage R&D Dividend Acquisition Industry FE (3-digit) Industry FE (4-digit) Firm FE Year FE Cohort dummy Observations R-squared (1) Pooled OLS (2) Pooled OLS (3) Fixed Eﬀect (4) Fixed Eﬀect (5) ∆ in Each Variable (6) Explained ∆ in cash (7) Explained Fraction -0.6928*** (0.0210) -0.0101*** (0.0008) 0.0067*** (0.0008) 0.0627** (0.0114) 0.0361*** (0.0058) -0.0096 (0.0067) -0.7484*** (0.0305) -0.2497*** (0.0105) 0.1397*** (0.0185) -0.0291*** (0.0047) -0.3894*** (0.0190) -0.6894*** (0.0201) -0.0107*** (0.0012) 0.0067*** (0.0008) 0.0604*** (0.0111) 0.0333*** (0.0056) -0.0077 (0.0067) -0.7330*** (0.0307) -0.2409*** (0.0105) 0.1160*** (0.0182) -0.0253*** (0.0045) -0.3767*** (0.0189) -0.7417*** (0.0137) -0.7299*** (0.0134) -0.0059*** (0.0018) 0.0047*** (0.0006) 0.0589*** (0.0089) 0.0174*** (0.0047) -0.0340*** (0.0052) -0.4033*** (0.0183) -0.1908*** (0.0068) -0.1160*** (0.0139) 0.0077*** (0.0025) -0.2571*** (0.0124) -0.1002 0.0731 (0.0013) -0.0065 (0.0020) 0.0063 (0.0008) 0.0061 (0.0009) -0.0043 (0.0012) 0.0042 (0.0006) 0.0107 (0.0005) 0.0036 (0.0001) -0.0092 (0.0011) -0.0011 (0.0004) -0.0014 (0.0001) 64.5% 1.104 1.364 0.104 -0.2456 -0.1235 -0.0265 -0.0191 0.0797 -0.1464 0.0054 -5.74% 5.56% 5.54% -3.80% 3.71% 9.45% 3.18% -8.12% -0.97% -1.24% Yes Yes Yes Yes Yes Yes 32,939 0.572 32,939 0.585 Yes Yes Yes Yes 32,939 0.770 32,939 0.799 and economically signiﬁcant. The coeﬃcients of other independent variables are consistent with those estimated in [11]. Larger ﬁrms, either because of economies of scale for transaction purposes or because of having easier access to external capital, hold less cash. Firms facing higher risks tend to save more cash because of precautionary motives. Firms expecting more future investment opportunities, proxied by market-to-book ratio and R&D spending, accumulate more cash. Also, paying oﬀ debt, investing in capital and distributing dividends consume cash. Results are robust with respect to diﬀerent speciﬁcations and regression methodologies. Columns (2)-(4) show quantitatively similar results to column (1). In particular, the coeﬃcient on inventory ratio varies within a fairly narrow interval [-0.69,-0.74]. To assess the importance of inventory, I estimate how much of the explained cash increase is associated with the changes in each explanatory 9 1.2. Empirical Evidence variable over the past thirty years. I ﬁrst compute the average value of each independent variable in 1980s and 2000s respectively. The diﬀerences between those two periods are the overtime changes of every ﬁrm characteristic, which are reported in Column (5). Then the product of the changes and their corresponding coeﬃcients reported in Column (4) gives us the contribution of each factor to the increase in the predicted cash. Results are summarized in Column (6) and (7). We can see that the most important factor related to the increased cash holdings is the decline in inventory, which accounts for 7.31 percentage points increase, or 64.5% of the observed cash increase. This result echoes the ﬁnding in [11]. 1.2.3 Cash and JIT I showed above that the inventory and cash ratios are negatively correlated and the decline in inventory is the primary driver behind the increase in cash holdings. Given the role of JIT playing in reducing inventory, we can infer its importance in understanding cash hoarding behavior. In this subsection, I provide direct evidence with a sample of JIT adopters and non-adopters. I examine the impact of JIT on corporate cash holdings by employing a diﬀerence-in-diﬀerences (DID) approach. The identiﬁcation comes from the diﬀerences between pre-adoption and post-adoption, within-ﬁrm diﬀerences of JIT adopters and non-adopters. The initial JIT adopter sample, along with the information on the adoption year for each ﬁrm, is kindly provided by the authors of [60].11 Of the 201 adopters, 14 ﬁrms are no longer available on Compustat; of the remaining 187 ﬁrms, I drop 18 non-manufacturing ﬁrms. My ﬁnal sample therefore includes 169 JIT adopters.12 I then pool these adopter with non-adopters from Compustat Industrial Annual ﬁles and extract ﬁnancial data for each ﬁrm. Table 1.3 presents the summary statistics for the relevant variables of the adopters. Relative to the average ﬁrm as shown in Table 1, adopters hold a similar level of inventory but less cash. They are larger in size, face lower cash ﬂow risks and have lower market-to-book ratios. They also have healthier operating cash ﬂows, higher net working capital and lower leverage ratios. In terms of expenses, adopters spend a similar rate on physical capital and acquisition, invest less in R&D and pay out more dividends. 11 12 Please refer to [60] for the detailed sample selection and screening procedures. The description of the JIT adopter sample is provided in Appendix A.3. 10 1.2. Empirical Evidence Table 1.3: Summary Statistics for 169 JIT Adopters Table 1.3 presents descriptive statistics for a sample of 169 JIT adopters, and reports the mean, median, standard deviation, 25th and 75th percentile, and number of observations of each variable used in the cash and inventory regressions. The sample covers the period 1980-2006. Detailed information about the adopters is provided in Appendix A.3. Variables 1.2.3.1 Mean Median Std. Dev. 25% 75% Obs. Cash 0.10 0.06 0.11 0.02 0.13 3314 Inventory 0.20 0.18 0.10 0.13 0.25 3314 Size 6.33 6.26 1.97 4.84 7.71 3314 Risk 0.04 0.24 0.06 0.01 0.04 3207 Market-to-Book 1.30 1.03 0.98 0.77 1.52 3265 Cash ﬂow 0.09 0.10 0.12 0.06 0.14 3314 Net working capital -0.01 -0.01 0.11 -0.06 0.05 3265 Capital investment 0.06 0.05 0.04 0.03 0.08 3281 Leverage 0.21 0.20 0.15 0.10 0.30 3309 R&D 0.05 0.04 0.06 0.02 0.07 2821 Dividend dummy 0.71 1 0.45 0 1 3314 Acquisition 0.02 0 0.05 0 0.01 3105 Sample Validation Before analyzing the eﬀect of JIT on cash holdings, I validate the adopter sample by examining whether adopters manage their inventory in a way consistent with JIT philosophy. To this end, I consider the following speciﬁcation, inventoryi,t = β0 + β1 Di,t + β2′ Xi,t + γi + σt + ϵi,t , (1.2) where Di,t is the dummy variable, taking the value one if ﬁrm i at time t implements JIT and zero otherwise, Xi,t are the control variables similar to those included in the cash regression, and γi and σt are ﬁrm ﬁxed eﬀects and year ﬁxed eﬀects, respectively. I identify β1 , the average eﬀect of JIT on inventory holdings, by assuming that (1) all the unobserved heterogeneity that leads to the correlation between adopting JIT and the error terms is captured by ﬁrm ﬁxed eﬀects, and that (2) the changes of the dependent variable due to changes in the macroeconomic environment are captured by year ﬁxed eﬀects, which are common to ﬁrms in both the treated and the control groups. 11 1.2. Empirical Evidence The estimation results for regression model (1.2) with the sample constructed from Compustat Industrial Annual ﬁles are reported in Columns (1)-(3) of Table 4. According to Column (1), after adopting JIT philosophy, ﬁrms on average reduce their inventory holdings by 4.25 percentage points, which is statistically signiﬁcant at the 1% level. In Columns (2) and (3), I report the results of regression (1.2) after controlling for other variables. The sign and signiﬁcance of the coeﬃcient estimate on JIT dummy variable remain the same. Table 1.4: Eﬀects of JIT Adoption on Corporate Inventory and Cash (Annual) Table 1.4 reports the estimation results of the inventory and cash regressions on JIT-adoption dummy and ﬁrms’ characteristics which include ﬁrm size, risk, market-to-book, cash ﬂow, net working capital, and other commonly-used control variables. Firm ﬁxed eﬀects and year ﬁxed eﬀects are included in the regressions, and the heteroskedasticity-consistent standard errors are reported in parenthesis. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1%, respectively. Columns (1)-(3) show the results of inventory regressions, and Columns (4)-(6) show the results of cash regressions. The sample is constructed from Compustat Industrial Annual ﬁles, covering the period 1980-2006. Variables JIT-adoption (1) inventory (2) inventory (3) inventory (4) cash (5) cash (6) cash -0.0425*** (0.0029) -0.0335*** (0.0029) -0.0178*** (0.0009) -0.0023*** (0.0003) -0.0221*** (0.0046) 0.0027 (0.0022) -0.0283*** (0.0031) -0.0222*** (0.0011) -0.0024*** (0.0003) -0.0303*** (0.0048) 0.0042 (0.0032) -0.0067** (0.0032) -0.0475*** (0.0121) 0.249*** (0.0038) -0.0046 (0.0082) 0.0072*** (0.0017) -0.306*** (0.0076) 0.0207*** (0.0039) 0.0114*** (0.0038) -0.0040*** (0.0016) 0.0064*** (0.0006) 0.0535*** (0.0090) 0.0507*** (0.0038) 0.0141*** (0.0043) 0.0103*** (0.0020) 0.0064*** (0.0007) 0.0810*** (0.0096) 0.0143*** (0.0051) -0.0291*** (0.0054) -0.3685*** (0.0201) -0.2089*** (0.0073) -0.1127*** (0.0152) 0.0024 (0.0027) -0.2347*** (0.0137) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 78,006 0.721 50,149 0.771 32,939 0.774 78,055 0.665 50,171 0.717 32,939 0.757 Size Market-to-book Risk Cash ﬂow Net working capital Capital investment Leverage R&D Dividend dummy Acquisition Firm FE Year FE Observations Adj R-squared I next estimate regression (1.2) with a sample constructed from Compu12 1.2. Empirical Evidence stat Industrial Quarterly. The advantage of using quarterly data, compared with annual data, is that it gives more information about how JIT adopters manage their inventory over time and therefore helps to avoid the possible lack of variations due to the small sample size of adopters. The corresponding results are reported in Columns (1)-(3) of Table 1.5. The coeﬃcient estimate of JIT dummy is quantitatively similar to the one reported in Table 1.4. Besides, the results are fairly robust to diﬀerent speciﬁcations. On average, JIT implementation leads ﬁrms to reduce inventory by 4.25 percentage points, according to Column (3) which controls a list of other explanatory variables aside from JIT adoption. Table 1.5: Eﬀects of JIT on Corporate Inventory and Cash (Quarterly) Table 1.5 reports the estimation results of the inventory and cash regressions on JIT-adoption dummy and ﬁrms’ characteristics which include ﬁrm size, risk, market-to-book, cash ﬂow, net working capital, and other commonly-used control variables. Firm ﬁxed eﬀects and year ﬁxed eﬀects are included in the regressions, and the heteroskedasticity-consistent standard errors are reported in parenthesis. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1%, respectively. Columns (1)-(3) show the results of inventory regressions, and Columns (4)-(6) show the results of cash regressions. The sample is constructed from Compustat Industrial Quarterly, covering the period 1980Q1-2006Q4. Variables JIT-adoption (1) inventory (2) inventory (3) inventory (4) cash (5) cash (6) cash -0.0461*** (0.0012) -0.0426*** (0.0015) -0.0185*** (0.0004) -0.0027*** (0.0001) -0.0402*** (0.0089) -0.0146*** (0.0040) -0.0425*** (0.0073) -0.0232*** (0.0009) -0.0025*** (0.0002) -0.0711*** (0.0143) -0.0209*** (0.0076) -0.0004 (0.0022) 0.0447*** (0.0102) 0.0338*** (0.0034) -0.0503*** (0.0131) 0.0068*** (0.0013) -0.0481*** (0.0146) 0.0250*** (0.0016) 0.0276*** (0.0021) 0.0052*** (0.0009) 0.0049*** (0.0003) 0.0760*** (0.0178) 0.1338*** (0.0073) 0.0285*** (0.0098) 0.0223*** (0.0019) 0.0050*** (0.0004) 0.1665*** (0.0303) 0.0338** (0.0148) -0.0222*** (0.0048) -0.1236*** (0.0202) -0.2150*** (0.0068) -0.2980*** (0.0299) 0.0005 (0.0028) -0.3094*** (0.0301) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 311,735 0.730 176,382 0.786 46,756 0.797 312,459 0.679 176,524 0.750 46,756 0.780 Size Market-to-book Risk Cash ﬂow Net working capital Capital investment Leverage R&D Dividend dummy Acquisition Firm FE Year FE Observations Adj R-squared The results shown above are from the speciﬁcation which assumes that the eﬀects of JIT on inventory are the same over time. Considering the facts that JIT implementation is a long-term process and ﬁrms possibly adjust their behaviour gradually, I also estimate the following speciﬁcation to allow 13 1.2. Empirical Evidence for heterogenous eﬀects of JIT during the post-adoption period.13 inventoryi,t = β0 + β1 Di, year 1−3 + β2 Di, year 4−6 + β3 Di, year 7−9 + β4 Di, year 10+ + β5′ Xi,t + γi + σt + ϵi,t , (1.3) where Di, year t1 −t2 is the dummy variable, taking the value one if ﬁrm i during the post-adoption period t1 to t2 operates in the JIT system and zero otherwise. The coeﬃcient of the dummy variable Di, year t1 −t2 measures the diﬀerence between pre-adoption inventory and inventory in the post-period t1 to t2 , which diﬀerences out the common shock aﬀecting both adopters and non-adopters. Columns (1)-(3) of Table 1.6 present the estimation results of regression (1.3). As expected, ﬁrms shed their inventory holdings progressively. Column (1) suggests that in the ﬁrst three years after adopting the JIT system, ﬁrms reduce inventory by 3.18 percentage points, while the reduction in inventory amounts to 5.69 percentage points after 10 years’ adoption. The same declining pattern of inventory, with similar magnitude, is found in Columns (2) and (3) after controlling for other explanatory variables. The inventory ratio drops by 2.7 and 2.2 percentage points in the ﬁrst 3 years to a total of 5.4 and 5.8 percentage points after 10 years, respectively. 1.2.3.2 The Eﬀect of JIT on Cash The results of regressions (1.2) and (1.3) shown in Tables 1.3-1.5 conﬁrm the validity of the JIT-adopter sample. In this subsection, I use it to analyze how JIT adoption aﬀects cash holdings. The speciﬁcation I consider is analogous to the regression equation (1.2), cashi,t = α0 + α1 Di,t + α2′ Xi,t + γi + σt + ϵi,t , (1.4) where of particular interest is the coeﬃcient estimate on the dummy variable Di,t . The variables in Xi,t are the ones used in the cash regression (1.1), including ﬁrm size, market-to-book ratio, cash ﬂow risk and so forth. The results of regression (1.4) are reported in Columns (4)-(6) of Table 1.4 and 1.5, which are estimated using annual and quarterly data, respectively. The coeﬃcient estimate of the dummy variable Di,t is positive and statistically signiﬁcant in all cases, suggesting that implementing JIT leads 13 There are ten management practices typically associated with the JIT system, which require changes throughout the entire organization, from the way that goods are ordered to the role of people working on the shop ﬂoor. These changes are not completed in one step but need continuous improvement and trial and error approach over many years. 14 1.2. Empirical Evidence Table 1.6: Eﬀects of JIT on Corporate Inventory and Cash (Quarterly) Table 1.6 reports the estimation results of the inventory and cash regressions on JIT-adoption dummies and ﬁrms’ characteristics which include ﬁrm size, risk, market-to-book, cash ﬂow, net working capital, and other commonly-used control variables. Firm ﬁxed eﬀects and year ﬁxed eﬀects are included in the regressions, and the heteroskedasticity-consistent standard errors are reported in parenthesis. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1%, respectively. Columns (1)-(3) show the results of inventory regressions, and Columns (4)-(6) show the results of cash regressions. The sample is constructed from Compustat Industrial Quarterly, covering the period 1980Q1-2006Q4. Variables (1) inventory (2) inventory (3) inventory (4) cash (5) cash (6) cash JITyear1−3 -0.0318*** (0.0014) -0.0394*** (0.0014) -0.0478*** (0.0015) -0.0569*** (0.0013) -0.0270*** (0.0018) -0.0366*** (0.0017) -0.0442*** (0.0018) -0.0540*** (0.0017) -0.0185*** (0.0004) -0.0027*** (0.0001) -0.0400*** (0.0089) -0.0144*** (0.0040) -0.0218*** (0.0052) -0.0325*** (0.0054) -0.0423*** (0.0056) -0.0578*** (0.0057) -0.0233*** (0.0009) -0.0024*** (0.0002) -0.0707*** (0.0143) -0.0208*** (0.0076) -0.0002 (0.0022) 0.0445*** (0.0101) 0.0336*** (0.0034) -0.0509*** (0.0131) 0.0065*** (0.0013) -0.0463*** (0.0146) 0.0124*** (0.0018) 0.0110*** (0.0020) 0.0202*** (0.0021) 0.0461*** (0.0020) 0.0184*** (0.0024) 0.0151*** (0.0026) 0.0202*** (0.0026) 0.0430*** (0.0026) 0.0052*** (0.0009) 0.0049*** (0.0003) 0.0759*** (0.0178) 0.1335*** (0.0073) 0.0122* (0.0074) 0.0193** (0.0084) 0.0234*** (0.0082) 0.0605*** (0.0086) 0.0223*** (0.0019) 0.0050*** (0.0004) 0.1659*** (0.0303) 0.0338** (0.0148) -0.0224*** (0.0048) -0.1229*** (0.0202) -0.2147*** (0.0068) -0.2973*** (0.0299) 0.0009 (0.0027) -0.3124*** (0.0301) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 311,735 0.730 176,382 0.786 46,756 0.797 312,459 0.679 176,524 0.750 46,756 0.780 JITyear4−6 JITyear7−9 JITyear10+ Size Market-to-book Risk Cash ﬂow Net working capital Capital investment Leverage R&D Dividend dummy Acquisition Firm FE Year FE Observations Adj R-squared ﬁrms to accumulate more cash and the conclusion is robust with respect to diﬀerent speciﬁcations and samples. According to Column (6) of Table 1.5, on average, adopting the JIT system induces ﬁrms to raise their cash balances by 2.85 percentage points. To interpret the magnitude of this effect, a 10 percentage point decline in inventory stock corresponds to a 6.7 percentage point increase in cash. JIT adoption therefore plays a signiﬁcant role in shaping corporate cash holdings. Also examined is the following speciﬁcation, a counterpart of regression (1.3) which allows for diﬀerent eﬀects of JIT on cash in the post-adoption 15 1.2. Empirical Evidence years, cashi,t = α0 + α1 Di, year 1−3 + α2 Di, year 4−6 + α3 Di, year 7−9 + α4 Di, year 10+ + α5 Xi,t + γi + σt + ϵi,t , (1.5) The estimation results are presented in Columns (4)-(6) of Table 1.6. The coeﬃcients on JIT dummies are all positive, statistically signiﬁcant and show an increasing trend, which suggest that after implementing the JIT philosophy, ﬁrms build up their cash reserves gradually over time. According to Column (6), they increase cash by 1.22 percentage points within the ﬁrst three years and this number climbs up to 6.05 percentage points after ten years, which is of economical signiﬁcance. Note that the impact of JIT on cash holdings that I measure here is the lower bound for the true impact. The ﬁrms included in the sample as non-adopters may or may not use JIT. If part of those so-called “nonadopters” do implement the new system, the estimated eﬀect should be biased downwards, which works against ﬁnding an impact. 1.2.4 International Evidence In this subsection, I investigate whether ﬁrms in Japan, the master and pioneer of JIT, manage their inventory and cash holdings in the same way as their U.S. counterparts. Also examined are Germany and France, Europe’s two largest economies following Japan in the 1980s to adopt JIT. Figure 1.2 plots the time series dynamics of the average cash ratio (line connected with circle), the average inventory ratio (line connected with triangle) and their sum (line connected with diamond) for manufacturing ﬁrms that are publicly traded in Japan, Germany and France. The sample of Japanese companies is constructed from PACAP for the period 1975 to 1990, while the samples for Germany and France are constructed from Compustat Global covering the period from 1988 to 2009. In Japan, JIT had gained extensive adoption by the mid 1970s. Since then, the inventory ratio has been eﬀectively reduced from 20% to 12%. From 1988, the inventory ratio stopped decreasing and became stable. The average cash ratio within the same period moved in the opposite direction, except for the last three years. The cash decrease starting from the end of 1980s is attributed to the weakened bank power ([75]). In Germany and France, the cash ratio and inventory ratio also go in opposite directions over time, with inventory trending down, cash trending up and the sum of those two ratios remaining stationary. More precisely, 16 1.2. Empirical Evidence .1 .2 ratios .3 .4 Japan 1975 1980 1985 1990 year average cash ratio average cash and inventory ratio average inventory ratio .1 .2 ratios .3 .4 Germany 1990 1995 2000 year average cash ratio average cash and inventory ratio 2005 2010 average inventory ratio .1 .15 ratios .2 .25 .3 .35 France 1990 1995 2000 year average cash ratio average cash and inventory ratio 2005 2010 average inventory ratio Figure 1.2: Average Cash and Inventory Ratios in Japan, Germany and France. The ﬁgure summarizes the average cash-to-asset ratio, average inventory-toasset ratio and sum of those two ratios over time in Japan, Germany and France. 17 1.3. Model the cash ratio more than doubles over the 12-year period in both countries, rising from 7.7% and 8.8% to 14.9% and 18.8% respectively. In the meanwhile, the inventory ratio drops from 23.4% and 22% to 17.5% and 11% correspondingly and the sum of the cash and inventory ratios ﬂuctuates roughly around 32% in both countries. 1.2.5 Discussion In summary, I in this section demonstrate the importance of JIT adoption in understanding the substantial rise in corporate cash holdings. I ﬁrst estimate the cash regression commonly used in the cash literature to show that the increase in cash holdings is mainly associated with the signiﬁcant inventory reduction. A 10 percentage point decline in inventory ratio is related with a 7.3 percentage point rise in cash ratio. Then employing the Diﬀerence-in-Diﬀerences (DID) analysis with a sample of JIT adopters and non-adopters, I ﬁnd that ﬁrms progressively shed inventory and accumulate cash after they implement JIT. After ten-year adoption, ﬁrms on average increase their cash reserves by over six percentage points. Lastly, I show that this cash-increase-inventory-decline phenomenon is also prevalent in Japan, Germany and France. All the empirical evidence supports the hypothesized relationship that JIT implementation is a major driver behind the observed increase in corporate cash balance. Caution is needed due to the small sample size of JIT adopters, as well as the possible endogeneity stemming from measurement error and simultaneity in the cash regression. To cope with these potential concerns, I next turn to a structural model to understand the eﬀect of JIT adoption on corporate cash holdings. In addition, with a structural model, I can analyze the mechanisms through which JIT implementation aﬀects corporate cash and quantify the impacts of those channels. 1.3 Model This section presents a partial equilibrium problem of a ﬁrm who faces uncertainty and ﬁnancing frictions. I introduce inventory holdings into an otherwise standard neoclassical model of capital investment and cash accumulation by modelling raw materials as factors of production.14 The lag in delivery of raw material purchases gives rise to the key diﬀerence between the JIC and JIT systems. In particular, the delivery lag under the 14 See, for instance, [79]. 18 1.3. Model JIC system leads ﬁrms to purchase raw materials in anticipation of future demand and to maintain inventory as buﬀers to meet uncertainties. Under the JIT system, since there is no lag in delivery, ﬁrms are able to respond contemporaneously to shocks, and they place orders based on current-period demand. I begin by specifying a ﬁrm’s production technology and ﬁnancing options. Then I describe the problems the ﬁrm faces when it operates as a JIC adopter and as a JIT adopter. Lastly, I characterize the ﬁrm’s optimal decision rules under the JIC and JIT environments and provide the economics behind the diﬀerent optimal choices on cash and inventory. 1.3.1 Technology I consider a discrete time model of an inﬁnitely lived ﬁrm. The ﬁrm combines physical capital k, labor l, and materials N , to produce output, and it faces a combination of demand and productivity shock, z. Maximizing labor out of the problem gives us the revenue function, F (z, k, N ), speciﬁed by a constant elasticity of substitution(CES) technology − ηθ F = z[αk −η + (1 − α)N −η ] . (1.6) Here, curvature θ < 1 captures decreasing returns to scale in production, or market power, or a combination of both; α < 1 is the share parameter, describing the weight of physical capital and materials in revenue function; and η controls the elasticity of substitution between these two inputs.15 The technology is subject to a revenue shock z, following an AR(1) process in logs with persistency ρ and innovation εz , ln z ′ = ρ ln z + ε′z . (1.7) A prime indicates a variable in the next period and no prime indicates a variable in the current period. The innovation εz has a normal distribution with mean 0 and variance σz2 , εz ∼ N (0, σz2 ). 15 The revenue function can be derived from a static optimization problem. Speciﬁcally, the ﬁrm faces a demand function y = z1 p−Θ and utilizes production technology y = − 1−β1 z2 (lβ1 [αk−η + (1 − α)N −η ] η )Φ to produce goods. Here, z1 and z2 are demand and productivity shocks, respectively. Given labor wage, optimization over the labor l yields a revenue function. 19 1.3. Model 1.3.1.1 Capital Every period, the ﬁrm augments its capital stock by capital investment, I, given as I = k ′ − (1 − δk )k. (1.8) The parameter δk is the capital depreciation rate, 0 < δk < 1. Adjusting capital by purchasing or selling it incurs adjustment costs, which are deﬁned by γ1 I 2 ( ) k. (1.9) 2 k This speciﬁcation includes only convex adjustment costs and the parameter γ1 > 0 captures the smoothing eﬀect. A(k, k ′ ) = 1.3.1.2 Inventory The ﬁrm makes a decision on whether or not to adjust material stock based on current state. The ﬁrm has two options: either pay a ﬁxed cost f and purchase is units of materials; or do not make adjustment this period. The ﬁxed adjustment costsf considered in this model consists of delivery cost, ordering and setup cost, preventive system maintenance cost and stockout cost. The speed of materials delivery depends on the manufacturing system adopted. Each period, after uncertainty is realized, the ﬁrm makes the inventory adjustment decision, and chooses how many materials, N , to use for production. Under the JIC environment, new purchased materials arrive after production. Therefore, no matter the ﬁrm decides to adjust its inventory stock or not, the decision N is constrained by the beginning of period material stock s, N < s. By contrast, under the JIT environment, new material orders get delivered before production starts. If the ﬁrm decides to make the adjustment, the materials available for current-period production become s1 = s + is . Alternatively, the ﬁrm can avoid the cost and enter production with its initial stock s1 = s. Materials fully depreciate in use and the unused ones are held as inventory and depreciate at a rate δs . The end of period inventory holdings are therefore given by s′ = (1 − δs )(s1 − N ), (1.10) with s1 = s + is if the ﬁrm makes new purchases this period and s1 = s otherwise. 20 1.3. Model 1.3.2 Financing To ﬁnance investment projects, ﬁrm has three sources: current operating cash inﬂow from sales, internal cash holdings and external funds. Internal cash balance, c, stored by the ﬁrm earns a risk-free rate r with the interest being taxed at a rate τc . The tax penalty is included to make sure the existence of an upper bound on cash holdings. External ﬁnancing takes the form of equity issuance in the model. Given the parsimonious set-up, this should not be interpreted literally. Despite in the form of equity issuance, it can be other sources of external ﬁnancing, such as debt ﬁnancing, line of credit from banks, and/or short-term credit from suppliers in the case of purchasing materials.16 Issuing equity incurs costs. The functional form I assume is linear-quadratic: 1 g(e) = ϕe (−λ1 e + λ2 e2 ). 2 (1.11) where e denotes dividends distributed to shareholders and a negative e indicates equity issuance. The indicator function ϕe equals zero if dividend e is non-negative, and one otherwise. Cost parameters λi , i = 1, 2, are positive. The functional form departs from the speciﬁcation used in [49], by excluding a ﬁxed cost term. This assumption is motivated by the fact that access to lines of credit or short-term trade credit is virtually no ﬁxed cost. The quadratic term is kept for capturing the eﬀect of debt ﬁnancing as well as simplifying numerical computation, though the estimate in [49] suggests that it is not signiﬁcantly diﬀerent from zero. The sources of funds available to ﬁnance inventory adjustment diﬀer under JIT and JIC, which stems from the lag in delivery of new purchases and the requirement to pay on receipt. Under JIC, options for funding material purchases are the same as those to ﬁnance capital investment, including operating cash ﬂows, cash balance and external funds. For JIT, however, cash ﬂows generated from current-period production are unavailable for purchasing materials, because now the payment is made before production begins. As such, the no lag in delivery and payment upon receipt induce an operational use of cash. That is, cash acts as a source of working capital to smooth production. 16 [61], based on a dataset on almost 30,000 trade credit contracts, suggest that trade credit is expensive for most buyers. The eﬀective annual interest rate ranges from 2% to 100% , with the average rate 54%. 21 1.3. Model 1.3.3 Firm’s Problem The risk-neutral ﬁrm’s objective is to maximize the equity value of the ﬁrm which is discounted at the risk-free rate r, by choosing between adjusting and not adjusting inventory, V (z, k, c, s) = max{V n (z, k, c, s), V a (z, k, c, s)}. (1.12) V n (z, k, c, s) denotes the ﬁrm’s value of inaction, and V a (z, k, c, s) is the value of adjusting inventory stocks, as a function of shock z, beginning-ofperiod capital stock k, cash balance c and inventory s. 1.3.3.1 Firm’s Problem under JIC Conditional on not adjusting inventory, the ﬁrm’s problem is: V n (z, k, c, s) = max {e1 − g(e1 ) + βEV (z ′ , k ′ , c′ , s′ )}, ′ ′ k ,c ,N (1.13) ′ ˆ e1 = F (z, k, N )−τc [F (z, k, N )−δk k−N ]−(c′ − Rc)−[k +A(k ′ , k)−(1−δk )k], 1 β= , 1+r c′ ≥ 0, s′ = (1 − δs )(s − N ) ≥ 0, ˆ = 1 + r(1 − τc ). R After shock is realized, the ﬁrm makes its decision on how many materials to use in producing goods, but constrained by the quantity available in stock. The remaining materials are stored as inventory. They depreciate and are transferred to next period. The ﬁrm also decides how much to invest in capital to build capital stock which can be utilized in next-period production, and decides how much to save in cash with non-negative cash balance constraint. If current period after-tax cash inﬂow is not suﬃcient to fund physical capital investment and cash saving, the ﬁrm issues equity and pays the corresponding issuance costs. If current resources are enough to cover those expenses, the ﬁrm distributes dividends. In the case of adjusting inventory, the ﬁrm solves a problem similar to the inaction scenario described above, except that now the ﬁrm decides how many new materials, is , to purchase as well. With everything else the same, the ﬁrm’s problem is modiﬁed as follows: V a (z, k, c, s) = max′ ′ {e1 − g(e1 ) + βEV (z ′ , k ′ , c′ , s′ )}, N,is ,k ,c (1.14) 22 1.3. Model where ′ ˆ e1 = F (z, k, N )−τc [F (z, k, N )−δk k−N ]−(c′ − Rc)−[k +A(k ′ , k)−(1−δk )k] −(f + is ), c′ ≥ 0, s ≥ N, s′ = (1 − δs )(s − N + is ). The ﬁrm makes inventory adjustment decision before production. However, the delivery lag leads to the unavailability of the newly-purchased materials for current production, N ≤ s, and the timing of the payment makes the current-period cash ﬂows available for funding material transactions. In other words, as a result of the lags in delivery and the requirement to pay upon receipt, the adjustment decision prior to the production is equivalent to the case of post-production adjustment. 1.3.3.2 Firm’s Problem under JIT In the case of inaction, the ﬁrm’s problem is the same as the one under JIC. Contingent on adjusting inventory, the problems that the ﬁrm faces under JIT and JIC are diﬀerent because of the delivery lag. Under the JIT system, the ﬁrm’s problem can be viewed as two stages. In the ﬁrst stage, after shock realization but before production, the ﬁrm makes a choice on material purchase. With the newly-purchased and immediatelydelivered materials, the ﬁrm enters the second stage in which it solves a problem the same as inaction but with diﬀerent levels of cash and inventory holdings, V a (z, k, c, s) = max{ϕe0 [e0 − g(e0 )] + V n (z, k, is >0 ϕe0 (1 − ϕe0 )e0 − f , s + is )}, ˆ R (1.15) ˆ − is , e0 = Rc { 1 if e0 ≤ 0 = 0 otherwise, ˆ − is reﬂecting the ﬁrst stage transaction, and is is the new where e0 = Rc purchase. The ﬁrm makes the payment out of internal cash balance. If internal funds are insuﬃcient to cover the costs (ϕe0 = 1), the ﬁrm resorts to equity ﬁnance and pays issuing costs and enters the second stage with 23 1.3. Model ˆ is zero cash. Otherwise, the remaining cash, e0 > 0 normalized by R, carried forward into the second stage. The materials available for production after the adjustment are the sum of the initial stock and the new purchase, s + is . Then the ﬁrm, with the new resource constraint, makes choices on material usage, capital investment and cash saving. Note that the ﬁxed inventory adjustment cost f is paid in the second stage, considering that the cost comprises not only delivery cost, but also ordering and setup cost, preventive system maintenance cost and stockout cost which are internal operating costs and not payable to suppliers. 1.3.4 Optimal Policy Rules In this subsection, I characterize the optimal decision rules for the ﬁrm’s problem under JIC and JIT and develop the intuition behind them. 1.3.4.1 Cash Holding under JIC Solving the optimization problem (10) for the JIC system gives the optimal cash level, which satisﬁes ′ ′ ′ ˆ + β RE[ϕ ˆ 1 + ϕe1 (λ1 − λ2 e1 ) = β R (λ1 − λ2 e1a )] V a ϕea 1 ′ ′ ′ ˆ + β RE[(1 − ϕV a )ϕen1 (λ1 − λ2 e1n )] + µ1 , (1.16) where ′ ϕV a = { 1 if the ﬁrm adjusts inventory next period, 0 otherwise, and ei , i = a, n, denotes dividends or equity issuance in the case of adjusting inventory (i = a) or inaction (i = n). The left hand side of equation (1.16) represents the marginal cost of saving an additional unit of cash, that is, forgone dividends, or the sum of equity issuance and its corresponding cost in case of issuing equity to ﬁnance cash saving in current period. The right hand side of the equation is the marginal beneﬁt of cash saving. It equals the sum of the discounted expected return (the ﬁrst term), the discounted expected reduction in the equity issuance costs in the case of adjusting inventory next period (the second term) and in the case of inaction next period (the third term). The last term of the right hand side in equation (1.16) is the Lagrange multiplier of nonnegativity constraint on cash and gives the shadow price of cash holdings. 24 1.3. Model Equation (1.16) implies that capital market imperfection leads the ﬁrm to accumulate internal funds. Without ﬁnancing frictions, λ1 = 0 and λ2 = 0, the ﬁrm never saves cash as there is a tax penalty on cash and accordingly ˆ < 1. In addition, optimal the return on cash is lower than risk free rate, β R saving is positively correlated with the probability of issuing equity. As the ﬁrm anticipates a higher likelihood of the insuﬃciency of internal funds to ﬁnance future capital and/or inventory investments, it retains more cash because more issuing costs are likely to be saved. 1.3.4.2 Cash Holding under JIT Under the JIT system, the optimal cash policy is given by ′ ′ ′ ′ ′ ′ ˆ ˆ 1+ϕe1 (λ1 −λ2 e1 ) = β R+β RE{ϕ (λ1 −λ2 e0a )+(1−ϕea0 )ϕea1 (λ1 −λ2 e1a )] V a [ϕea 0 ′ ′ ′ ˆ + β RE[(1 − ϕV a )ϕen1 (λ1 − λ2 e1n )] + µ1 , where ′ ϕea0 = { (1.17) ′ 1 if issue equity when adjusting inventory next period, e0a ≤ 0, 0 otherwise, and ei , i = 0, 1, stands for equity issuance in the ﬁrst stage (i = 0) and second stage (i = 1) under the JIT system, respectively. With everything else remaining the same, equation (1.17) diﬀers from ′ ′ equation (1.16) through having an extra term, ϕea (λ1 − λ2 e0a ). It character0 izes the scenario in which next period the ﬁrm will issue equity to augment inventory stock in the ﬁrst stage when internal cash balance is not suﬃcient to make the transaction. Why does a ﬁrm hold more cash under JIT than under JIC? Comparing equation (1.16) with equation (1.17) reveals that the ﬁrm tending to save more cash is because of the higher probability of issuing equity under JIT. Adjusting inventory prior to production brings in a disconnection between material purchases (use of funds) and cash ﬂows generated this period (source of funds). As a result, cash ﬂows, an important source of internal liquidity, are not available to pay for materials. In order to avoid tapping into expensive external funds, the ﬁrm carries cash forward to make the payment and smooth future production. Equation(1.17) shows that under JIT, cash serves two purposes: ﬁnancing investment opportunities and facilitating production. The former is the precautionary motive for cash holdings, or rather, cash is held as a hedge against risk that future cash ﬂow shortfalls and capital market frictions result 25 1.3. Model in underinvestment. The latter is the transaction motive for cash holdings, that is, cash is needed in ordinary daily operations to make payments for material purchases in the presence of ﬁnancing frictions. Cash becomes a perfect substitute for inventory, acting as a source of working capital. The second motive for cash holding is absent under JIC. 1.3.4.3 New Purchase under JIC and JIT Systems I next turn to the optimal material purchase decision conditional on adjusting. Under the JIC system, the condition is given by 1 + ϕe1 (λ1 − λ2 e1 ) = (1 − δs )βE ∂V ′ . ∂s′ (1.18) The marginal cost of purchasing one more unit of good, shown by the left hand side, is the foregone dividend (or the sum of issued equity and issuing costs) this period, while its marginal beneﬁt is the expected present marginal value of an additional unit of end-of-period inventory in next period, but net of depreciation. Equation (1.19) describes the optimal condition of material purchase under the JIT system, ϕe0 [1 + ϕe0 (λ1 − λ2 e0 )] + (1 − ϕe0 )[1 + ϕe1 (λ1 − λ2 e1 )] = (1 − δs )βE ∂V ′ + µ2 , ∂s′ (1.19) with { ϕea0 = 1 if issue equity while adjusting inventory this period, ea0 ≤ 0, 0 otherwise, and µ2 denoting the Lagrange multiplier associated with the constraint s + is − N ≥ 0. The marginal cost under JIT relies on the suﬃciency of funds in the ﬁrst stage to purchase an additional unit of materials. If there is a shortage of internal liquidity, the marginal cost is the issued equity and its costs in the ﬁrst stage; or, the marginal cost is the same as that under the JIC system. The marginal beneﬁt for purchasing one more unit of materials is greater than that in the JIC case, captured by the extra term µ2 which appears in the equation as a result of the availability of newly-purchased materials in current-period production. Under JIT, the ﬁrm makes new purchases not only for building up inventory to avoid stockouts next period, but also for meeting current needs. 26 1.3. Model 1.3.4.4 Material Usage The optimal material use under JIT and JIC systems is characterized by the same equation, [1 + ϕe1 (λ1 − λ2 e1 )][(1 − τc )FN + τc ] = (1 − δs )βE ∂V ′ + µ2 . ∂s′ (1.20) Here, FN denotes the marginal revenue generated by an additional unit of material inputs and µ2 is the multiplier of the constraint that material use cannot exceed inventory available in stock. The left-hand side of equation (1.20) represents the marginal beneﬁt of using one additional unit of materials in production, which is the increased dividend payment generated from the marginal revenue and the deduced tax from material expenses. The right-hand side shows the corresponding marginal cost, which is the expected present marginal value of an additional unit of end-of-period inventory for next period, but net of depreciation. 1.3.4.5 Capital Investment Lastly, I discuss the optimal capital investment policy which is described by the following equation, ′ ′ [1+ϕe1 (λ1 −λ2 e1 )][1+Ak′ (k, k )] = βE{[1+ϕe1 (λ1 −λ2 e′1 )][(1−τc )Fk′ +τc δk ′ ′′ + (1 − δk ) − Ak′ (k , k )]}, (1.21) in which Fk′ is the marginal revenue next period generated by an additional unit of capital investment. The Euler equation is the same as the one we generally see in neo-classical framework, other than extra terms on equity issuing costs. Investing in capital this period incurs both direct costs and adjustment costs, which cuts current dividend payments to the shareholders. But it also beneﬁts shareholders next period: the expected discounted revenue earned from extra outputs produced by one additional unit of capital, the reduced tax payment from capital depreciation, the expected discounted value of that additional unit of capital net of depreciation, and the reduction in adjustment costs from that increment to capital stock. 27 1.4. Simplified Model with Analytical Solution 1.4 Simplified Model with Analytical Solution In the previous section, I present a dynamic stochastic model and characterize a ﬁrm’s optimal decisions under JIT and JIC systems. To illustrate and stress the main intuition behind the substitution between cash and inventory holdings and to examine whether my model is able to deliver a negative correlation of a similar magnitude to that observed in the data, I in this section present a simpliﬁed model that yields closed form solutions for cash and inventory holdings. I simplify the problem by assuming that there are no ﬁxed costs of inventory adjustment, f = 0, and no uncertainty, ρ = 0 and σz = 0, and I show analytically that adopting JIT leads ﬁrm to reallocate resources from inventory to cash. 1.4.1 Steady State under JIC System Without the ﬁxed inventory adjustment costs, a ﬁrm adjusts its inventory stock each period, and the ﬁrm’s problem is reduced to an adjusting case. I restate the ﬁrm’s problem in the JIC environment as follows, V (k, c, s) = max is >0,N,k′ ,c′ {e1 − g(e1 ) + βV (k ′ , c′ , s′ ), } where ′ ˆ e1 = F (0, k, N )−τc [F (0, k, N )−δk k−N ]−(c′ −Rc)−[k +A(k ′ , k)−(1−δk )k]−is , c′ ≥ 0, s ≥ N, ′ s = (1 − δs )(s − N + is ). Rewriting the problem with multipliers, the ﬁrst order conditions of cash, inventory holdings and new purchases are given by: V (k, c, s) = max′ ′ {e1 −g(e1 )+βV (k ′ , c′ , (1−δs )(s−N +is ))+µ1 c′ +µ2 (s−N )}, N,is ,k ,c ′ ˆ + β R[ϕ ˆ e (λ1 − λ2 e′ )] + µ1 , c′ : 1 + ϕe1 (λ1 − λ2 e1 ) = β R 1 1 ′ N : [1 + ϕe1 (λ1 − λ2 e1 )][(1 − τc )FN + τc ] = (1 − δs )β ∂V ∂s′ + µ2 , ′ is : 1 + ϕe1 (λ1 − λ2 e1 ) = (1 − δs )β ∂V ∂s′ , 28 1.4. Simplified Model with Analytical Solution s : ∂V ∂s ′ = (1 − δs )β ∂V ∂s′ + µ2 . I characterize the steady state below. Let a star (*) indicate the steady state value of a variable. With the Euler equations derived above, I can show that in equilibrium, a ﬁrm carries positive inventory but zero cash under JIC. Proposition 1 In the steady state of the JIC system, s∗ = N ∗ = (1−δs )i∗s > 0. Proof. The proof is by contradiction. Suppose µ∗2 = 0 and therefore i∗s > 0, the ﬁrst order condition of s, ∂V ∂V = (1 − δs )β ∗ + µ∗2 , ∗ ∂s ∂s ∂V implies that ∂s ∗ = 0. Substituting this into the Euler equation of is gives 1 + ϕ∗e1 (λ1 − λ2 e∗1 ) > 0. This in turn implies i∗s = 0, because the marginal cost is larger than the marginal beneﬁt of purchasing new materials. This contradicts with i∗s > 0. Therefore, in the steady state, µ∗2 > 0, and we can conclude that s∗ = N ∗ = (1 − δs )i∗s . The ﬁrst equality is derived from complementary slackness µ∗2 (s∗ − N ∗ ) = 0 and the second equality is obtained from the law of motion in s, s∗ = (1 − δs )(s∗ − N ∗ + i∗s ). Proposition 2 In the steady state of the JIC system, c∗ = 0. Proof. The proof is by contradiction. Suppose c∗ > 0, complementary slackness implies that µ∗1 = 0. From the Euler equation of cash holdings, ˆ + β R[ϕ ˆ ∗e (λ1 − λ2 e∗1 )] + µ∗1 , 1 + ϕ∗e1 (λ1 − λ2 e∗1 ) = β R 1 ˆ This contradicts with the tax penalty on cash savthis implies that 1 ≤ β R. ˆ therefore c∗ = 0. ings, 1 > β R, 1.4.2 Steady State under JIT The ﬁrm’s problem under the JIT environment is rewritten as V (k, c, s) = max is >0,N,k′ ,c′ {ϕe0 [e0 − g(e0 )] + [e1 − g(e1 )] + βV (k ′ , c′ , s′ )}, where ˆ − is , e0 = Rc 29 1.4. Simplified Model with Analytical Solution ′ ˆ e1 = F (0, k, N )−τc [F (0, k, N )−δk k−N ]−(c′ − Rc)−[k +A(k ′ , k)−(1−δk )k], c′ ≥ 0, s + is ≥ N, s′ = (1 − δs )(s − N + is ). Again, the Bellman equation with multipliers can be reformulated as V (k, c, s) = max is >0,N,k′ ,c′ {ϕe0 [e0 −g(e0 )]+[e1 −g(e1 )]+βV (k ′ , c′ , (1−δs )(s−N +is )) + µ1 c′ + µ2 (s + is − N )}, and the corresponding Euler equations are ′ ′ ˆ + β R[ϕ ˆ c′ : 1 + ϕe1 (λ1 − λ2 e1 ) = β R e0 (λ1 − λ2 e0 ), ′ ′ ′ + (1 − ϕe0 )ϕe1 (λ1 − λ2 e1 )] + µ1 ′ N : [1 + ϕe1 (λ1 − λ2 e1 )][(1 − τc )FN + τc ] = (1 − δs )β ∂V ∂s′ + µ2 , ′ is : 1 + ϕe0 (λ1 − λ2 e0 ) = (1 − δs )β ∂V ∂s′ + µ2 , s : ∂V ∂s ′ = (1 − δs )β ∂V ∂s′ + µ2 . Diﬀerent from the JIC case, under JIT, the ﬁrm chooses to hold cash to facilitate operations. The next two propositions exhibit that the steady state values of cash and inventory are positive and zero, respectively. Proposition 3 In the steady state of the JIT system, s∗ = 0. Proof. The proof is similar to the one in Proposition 2. It is straightforward to show that µ∗2 > 0 by contradiction. From the constraint s∗ + i∗s − N ∗ = 0 and the law of motion of inventory holdings s∗ = (1 − δs )(s∗ − N ∗ + i∗s ), we can derive that in the steady state s∗ = 0 and i∗s = N ∗ > 0. ∗ ∗ Proposition 4 In the steady state of the JIT system, c∗ → iˆs = Nˆ > 0 if R R external ﬁnancing is expensive relative to internal funds (λ1 ≥ r). Proof. The proof is by assuming the contrary. First, in steady state the ﬁrm does not issue equity at the second stage, otherwise ﬁrm value would turn out to be negative. That is, ϕ∗e1 = 0. Next, suppose c∗ = 0, which implies that the ﬁrm does not have internal funds to ﬁnance new purchases at the 30 1.4. Simplified Model with Analytical Solution ﬁrst stage and therefore the ﬁrm has to issue equity, ϕ∗e0 > 0. Substituting both ϕ∗e0 > 0 and ϕ∗e1 = 0 into Euler equation of cash holdings, we have ˆ + λ1 − λ2 e∗0 ) + µ∗1 . 1 < β R(1 1−µ∗ ∗ 1 Because λ1 ≥ r > ˆ − 1 + λ2 e0 , the marginal beneﬁt of holding an βR additional unit of cash is greater than its marginal cost, which indicates that the ﬁrm would have an incentive to increase cash holdings. This contradicts with the assumption c∗ = 0 in steady state, and therefore c∗ > 0. To determine the optimal cash holdings, I next suppose c∗ > 0 but ∗ ϕe0 = 0. That is, the ﬁrm holds some cash, which however is insuﬃcient to pay for the entire newly-purchased materials. The Euler equation becomes ˆ + λ1 − λ2 e∗ ). 1 < β R(1 0 Because external funds are costly, λ1 > r, the marginal beneﬁt of holding an additional unit of cash remains larger than the marginal cost, the ﬁrm therefore continues to save cash. Would the ﬁrm hold enough cash so that it does not need to resort to external sources? I assume ϕ∗e0 = 0. Then the Euler equation is reduced to ˆ 1 > β R. In this case, the marginal cost of holding an additional unit of cash is greater than its marginal beneﬁt, so that∗ the ﬁrm tends to save less and the optimal cash holdings is given by c∗ → iˆs . R 1.4.3 Discussion The results established in Propositions (1)-(4) allow us to understand the diﬀerent incentives for carrying cash and inventory under JIT relative to JIC. In the steady state, the ﬁrm operating in the JIC situation holds zero cash, because cash in this environment serves for a precautionary purpose only. In the absence of uncertainty, the ﬁrm has full information on future liquidity needs and plans manufacturing to generate enough cash ﬂows to ﬁnance investments, both capital and inventory. Therefore, the ﬁrm has no incentive to save cash. Contrary to cash, inventory under this system works as working capital, and is not adjustable next period before production. Anticipating the level of material uses, the ﬁrm holds inventory forward to smooth future manufacturing. 31 1.5. Quantitative Analysis By contrast, under JIT, cash is saved in spite of the absence of uncertainty, whereas inventory holdings are zero. This is because that implementing JIT allows the ﬁrm to ﬂexibly adjust its inventory stock before organizing production each period. The ﬁrm, hence, no longer has motives to store inventory which is very costly. Instead, the ﬁrm chooses to transfer exact amount of cash forward to purchase materials required to produce goods. Relative to the JIC system, cash in this environment takes the place of inventory, acting as working capital to facilitate operation. The simpliﬁed model gives intuition on how implementing JIT drives the substitution between cash and inventory. More importantly, it generates a very similar magnitude of substitutability to that presented in the empirical section. Unconditionally, as shown in Figure 1, the substitution ratio between cash and inventory is approximately 1. After controlling for other variables, the ratio drops slightly to 0.73. Comparing the steady state values of cash and inventory under the two diﬀerent systems, we can see that cash ∗ goes up to Nˆ from 0, while inventory decreases from N ∗ to 0. The implied R substitution ratio lies in the range [0.73,1] obtained from data. To summarize, the model provides an explanation for the substitutability between cash and inventory and reproduces a reasonable substitution ratio close to the data. 1.5 Quantitative Analysis This section reports numerical results of the dynamic stochastic model built in Section 3. I begin by considering three main features of the JIT system and evaluating their impacts on ﬁrms’ cash policy and inventory management. I then conduct a comparison between the explanation often-discussed in the literature and mine on the performance of explaining cash increase. Lastly, I use my model as a laboratory to study whether the adoption of JIT increases ﬁrm value. To examine the model’s quantitative predictions, I ﬁrst calibrate parameters based on the JIC environment with a sample of manufacturing ﬁrms from Compustat. Then I parameterize the JIT environment with the same set of values to assess the impact of a change in the delivery lag, one of the major diﬀerences between JIC and JIT philosophy, on cash holdings by isolating other possible factors. I also re-calibrate some parameters to reﬂect other main features of JIT and analyze how those changes inﬂuence a ﬁrm’s cash saving as well as other decisions. 32 1.5. Quantitative Analysis 1.5.1 Calibration The time period t in my model corresponds to one year.17 The parameter values are therefore calibrated to match ﬁrm-level and aggregate-level moments at annual frequency in the early 1980s, or chosen based on parameter values that are standard in the literature whenever possible. The calibration strategy is discussed below. Firm’s Revenue Function. The curvature of the revenue function θ captures both return to scale in production and a ﬁrm’s market power. [48], [49], [32] and [41] estimate the parameter to be 0.551, 0.627, 0.59 and 0.64, respectively. I set it to be 0.55, the lower bound of the interval. The parameter η is set equal to 2, reﬂecting the elasticity of substitution between capital and materials is roughly 0.33. [28] estimates this parameter with post war U.S. aggregate data and suggests a low elasticity of substitution between capital and inventory with the two-standard-error interval approximately [0,0.63]. [15] calibrate this parameter with Compustat data and ﬁnd η = 0.5, that is, the elasticity of substitution is 0.67. My choice of the elasticity of substitution lies in the range established in these two studies. The last parameter in the revenue function needed to be determined is the share parameter α. I will return to this discussion below. Stochastic Process. To parameterize the stochastic process by choosing the persistence parameter ρ and the standard deviation σ, I draw on a large literature that estimates the process with Compustat data. At annual frequency, [39] and [4] estimate a serial correlation around 0.24-0.45 with the corresponding standard deviation around 0.15-0.45, while [40], [48], [49] and [55] have a serial correlation and the standard deviation in the ranges 0.62-0.74 and 0.12-0.15, respectively. Considering model similarities as well as shock speciﬁcations, I choose persistency ρ = 0.7 which falls in the upper end of the estimates, and its corresponding standard deviation σ = 0.15. Capital Adjustment. Using simulated method of moment (SMM) estimation and Compustat data, Nikolov and Whited (2010) estimate γ1 , the quadratic capital adjustment cost, to be in the range of 0.41-0.71 with four slightly diﬀerent models. Following their study, I set γ1 to be 0.7. The capital depreciation rate δk is set equal to the average of depreciation to the gross capital stock, 0.12, a value derived from Compustat manufacturing industries in the early 1980s. Inventory Adjustment. The parameters governing the inventory dynamics are the ﬁxed adjustment costs f and the inventory depreciation rate 17 Solving the model at annual frequency instead of quarterly frequency reduces computational time substantially, but hardly changes the results. 33 1.5. Quantitative Analysis δs . I calibrate them together with the capital-share parameter α by matching the average cash to asset ratio, average inventory to asset ratio and average capital to sales ratio in manufacturing in 1980 in Compustat.18 The ﬁrst target, the average cash to asset ratio, is informative about the ﬁxed costs f . The simpliﬁed model presented in Section 4 shows that without ﬁxed adjustment costs, a ﬁrm adjusts its inventory stocks each period and has no need to hold cash. The presence of the ﬁxed cost along with uncertainty makes the ﬁrm cautious. It chooses to adjust inventory less frequently and accumulate cash to pay for the non-convex cost. The second target, the average inventory to asset ratio, is informative about the inventory carrying costs which is captured by the depreciation rate δs .19 Intuitively, the ﬁrm is reluctant to hold inventory if it faces high carrying costs. The last moment, the average capital to sales ratio, provides information on the share parameter α, given the elasticity of substitution between capital and materials in production and the curvature of the revenue function. Financing, Corporate Income Tax and Interest Rate. [49] use structural estimation to infer the magnitude of external ﬁnancing costs. They estimate the linear cost of equity issuance to be 0.091, when allowing for non-convex equity issuance costs. When excluding the non-convex term, [73] estimate the linear cost to be approximately within the range [0.13,0.18]. Considering that external ﬁnancing in my model does not incur non-convex cost and can be other cheaper sources other than equity issuance, I set the parameter to be 0.10, which is lower than the estimates in [73] but slightly higher than the one in [49]. The quadratic equity issuance cost λ2 is selected equal to 0.0004, following the estimate in [49].20 The real risk-free rate r is set at 4%, a diﬀerence between the average annual Treasury-bill rate in the early 1980s and the average annual inﬂation rate in the same period. This corresponds with the discount factor β=0.96. The corporate income tax τc is set at 46%, according to the corporation income tax brackets and rates reported by Internal Revenue Service (IRS).21 18 The average ﬁrm-level capital-to-sales ratio ﬂuctuates over time. As an attempt to control the business cycles in the early 1980s recessions, I compute the average capital-tosales ratio for the period 1980-1983. 19 Given the theoretical setup in which I focus on input inventory only, I calibrate the model to match average input inventory to asset ratio. All conclusions/results presented in the previous Empirical Evidence section also hold for input inventory. Results available upon request. 20 A ﬁrm’s cash holding decision appears to be insensitive to the parameter λ2 . See also [79]. 21 Between 1980 to 1986, a tax rate of 46% is applied on corporate income bracket over $100,000. 34 1.5. Quantitative Analysis Table 1.7: Model Parameterizations Table 1.7 summarizes the parameters used to solve the model. Panel A reports the parameters specifying revenue function and governing revenue shocks, the parameters characterizing the evolvement of physical capital and inventory stock, and the parameters describing a ﬁrm’s external ﬁnancing conditions as well as corporate income tax rate and interest rate. Panel B presents calibration results. The data moments are computed based on the sample period of the early 1980s. Panel C reports the model predictions on the mean and variance of capital investment ratio under the JIC system. Panel A: Assigned Parameters Technology and Shock Process curvature (θ) 1 elasticity of substitution between capital and material ( 1+η ) standard deviation of shock (σz ) persistency of shock (ρ) Capital and Inventory quadratic capital adjustment cost (γ1 ) capital depreciation rate (δk ) Financing linear equity cost (λ1 ) quadratic equity cost (λ2 ) risk-free rate (r) corporate income tax (τc ) Panel B: Calibration Calibrated Parameters capital share (α) ﬁxed adjustment cost (f ) inventory holding cost (δs ) Moments Used for Calibration average capital to revenue (k/F (z, k, N )) average cash ratio (c/(k + c + s)) average inventory ratio (s/(k + c + s)) Panel C: Predictions under JIC Capital Investment (I/(k + c + s)) average investment variance of investment 0.55 0.33 0.15 0.7 0.7 0.12 0.10 0.0004 0.04 0.46 Data 0.838 0.098 0.193 0.90 0.055 0.32 JIC 0.835 0.099 0.200 Data 0.076 0.001 JIC 0.082 0.003 Table 1.7 presents the parameter values used for solving the JIC model. Panel A summarizes the parameters borrowed from other studies. The ﬁrst set of parameters describes a ﬁrm’s revenue function and the exogenous stochastic process that the ﬁrm faces. The second set of parameters speciﬁes the dynamics of physical capital stock. And the last set characterizes the ﬁrm’s external ﬁnancing conditions, corporate income tax rate and interest rate. Panel B of Table 1.7 reports the data moments I select to target, their model counterparts and the calibration results. The estimated capital share α is 0.90, a value between those suggested in [28] and [15]. The inventory carrying costs amount to 32% of a ﬁrm’s inventory holdings, which falls 35 1.5. Quantitative Analysis within the range estimated by [78]. The ﬁxed adjustment costs f is required to be 0.055 in order for a ﬁrm to have strong incentives to hold the level of cash we observe in the data. The value is approximately equivalent to 14% of average revenue. The ﬁxed adjustment costs considered in this model consist of delivery costs, ordering and setup costs (such as costs for unloading and internal handling), preventive system maintenance costs and stockout costs. 1.5.2 Quantifying Key Features of JIT With the calibrated parameters, I now examine the quantitative implications of my JIT model to see if it has the potential to provide a powerful explanation for the cash hoarding phenomenon. I quantify the impacts of JIT on cash holdings by decomposing it into three main changes — no lag in delivery, increased inventory adjustment costs, and long-term contracts with reliable suppliers — and then examining the impact of each. 1.5.2.1 No Lag in Delivery The most important change introduced by JIT philosophy is the speedy delivery and therefore operation ﬂexibility. It allows ﬁrms to free resources tied up in materials and products sitting in warehouses, and to save utilities and space costs incurred in keeping inventory from perishing. To estimate the eﬀect of this change, I set all parameters in the JIT environment to their values in the JIC environment. The results are reported in the column titled No Lag in Delivery of Table 1.8. In this environment, ﬁrms on average have cash-asset ratio of 24.7%, which implies that the production ﬂexibility alone leads ﬁrms to increase cash ratio by 15 percentage points or hold cash balance roughly 2.5 times as large as that in the JIC environment. Moreover, ﬁrms on average hold zero inventory. The reason for the zero inventory but skyrocketing cash balance is that operation ﬂexibility makes materials much cheaper. Firms therefore decide to employ more materials in production compared with the JIC environment. Because of the complementarity of materials and physical capital in producing products, ﬁrms also augment their capital stock, which along with materials boosts sales/revenue. Relative to revenue, ﬁxed inventory adjustment costs now are negligible, and ﬁrms choose to purchase materials each period. As a result, ﬁrms hoard cash instead of carrying inventory forward. This in turn makes cash and inventory ratios nearly invariable, indicated by their zero variances. 36 1.5. Quantitative Analysis Table 1.8: Implications under the JIT System Table 1.8 presents the parameter values used in simulating models under JIC and JIT and their corresponding simulated moments. Panel A summarizes the parameter values used in the JIC model and the models with diﬀerent frictions under JIT. In Panel B, Column (2) reports the data moments computed based on a sample of manufacturing ﬁrms in 1980 from Compustat. Column (3) reports the simulated moments generated in the benchmark JIC model. Column (4) reports the data moments computed based on a sample of manufacturing ﬁrms in post-2000 period from Compustat. Columns (5)-(7) report the simulated moments under JIT environments with diﬀerent frictions. Data 1980 Panel A: Parameters ﬁxed adjustment cost (f ) linear equity cost (λ1 ) Panel B: Moments Cash Average cash Variance of cash JIC Data 2000s 0.055 0.10 No Lag in Delivery Increased Fixed Adjustment Costs Long-term Contracts 0.055 0.10 0.101 0.10 0.101 0.04 0.098 0.099 0.251 0.010 0.247 0.000 0.231 0.022 0.203 0.027 Inventory Average inventory Variance of inventory 0.193 0.200 0.082 0.001 0.000 0.000 0.077 0.006 0.084 0.007 Investment Average investment Variance of investment 0.076 0.082 0.044 0.001 0.089 0.001 0.082 0.001 0.086 0.001 1.5.2.2 Increased Inventory Adjustment Costs As shown in the previous subsection, the immediate delivery of parts enables ﬁrms to gain eﬃciency through operation ﬂexibility and inventory carrying cost reduction. It leads ﬁrms to devote more resources to production, using more material inputs and investing more in physical capital. Therefore, the material order quantity conditional on adjusting inventory stocks also endogenously grows. Intuitively, larger order sizes incur higher inventory adjustment costs, by requiring more labor to unload and handle purchased materials and higher shipping charges. To take into account this change, I increase the ﬁxed inventory adjustment cost. I compare the average material order size is under JIC and JIT environments, and raise the ﬁxed inventory adjustment cost f by the same proportion as the changes in the average order size. I therefore set f = 0.101 under JIT. All other parameters take the same values as the ones used in assessing the eﬀect of speedy material delivery. 37 1.5. Quantitative Analysis The corresponding results are reported in the column titled Increased Fixed Adjustment Costs of Table 1.8. Although the stockout motive for inventory holdings is absent, to economize on the ﬁxed adjustment costs, ﬁrms choose to carry some level of inventory forward instead of replenishing the stock before production at the beginning of each period. This drives the average inventory level up by 7.7 percentage points and leads to more volatile inventory ratio. The presence of inventory stock also makes cash holdings less important in facilitating operation and production. Firms on average lower cash balance by 1.6 percentage points, with the ratio dropping from 24.7% to 23.1%. 1.5.2.3 Long-term Commitment Contracts with Suppliers [87] ﬁnds that companies implementing JIT usually try to develop a nurturing and reliable long-term relationship with their suppliers. More friendly terms and conditions of trade credit may be obtained by entering this kind of long-term commitment. I capture this feature by lowering borrowing costs. Due to the lack of information on trade credit contracts between JIT adopters and their suppliers, I set the linear equity issuance cost to 0.04 which is as low as the risk-free rate. Note that the number obtained will give the lower bound of the eﬀect of JIT on corporate cash, considering that the borrowing cost should be at least equal to the risk-free rate. All other parameter values remain the same as those in the economy with increased inventory adjustment costs. The results are summarized in the column Long-term Contracts of Table 1.8. Relaxed ﬁnancial frictions lower the value of ﬁnancial ﬂexibility provided by internal cash, and encourage ﬁrms to resort to external ﬁnancing. The reduced borrowing cost, from 10% to 4%, leads to a drop in average cash ratio by 2.8 percentage points, from 23.1% to 20.3%. Besides, cheaper external ﬁnancing leads ﬁrms to invest more in inventory and capital. Compared to the economy with increased inventory adjustment costs, both the inventory and capital investment ratios rise. 1.5.2.4 Discussion I quantify the impacts of the JIT system on corporate cash holdings by considering three main changes and adding each to the benchmark model sequentially. The results reported in Table 8 suggest that if all ﬁrms in the economy switch their operating system from JIC to JIT, (i) operation ﬂexibil38 1.5. Quantitative Analysis ity contributes to a 14.8 percentage-point increase in cash ratio, (ii) increased inventory adjustment costs, along with speedy delivery, lead to a 13.2 percentage-point increase in total, and (iii) conditional on both speedy delivery and increased inventory adjustment costs, the long-term commitment contract between producers and their suppliers causes the average cash ratio to drop by at most 2.8 percentage points. Overall, the implementation of JIT results in at least 10.5 percentage-point increase in cash holdings. My model also does a good job in explaining the decline in inventory holdings. It predicts that ﬁrms on average have inventory ratios of 8.4% in the post-adoption period, which is a 11.6-percentage-point reduction relative to the JIC system. However, the simulated capital investment ratio overshoots its data counterpart, 8.6% vs. 4.4%. I conjecture that this discrepancy arises because in the real world manufacturers gradually opted to outsource more and more of their production to suppliers abroad for cost reduction purposes, which in turn reduces their capital investment commitments. Also, the model fails to generate reasonable variances of cash and inventory ratios. The simulated variances are nearly three and seven times the data. One possible reason for the inaccurate predictions is that it is hard to observe lumpiness in inventory investment with ﬁrm-level data. 1.5.3 JIT and the Rise in Corporate Cash The subsequent exercise is to quantify the fraction of the observed cash increase attributed to the JIT implementation. To perform this analysis, we need to know the percentage of ﬁrms using JIT in the economy and make appropriate adjustments to the results derived above. According to the report Physical Risks to the Supply Chain provided by CFO Research Services in collaboration with FM Global, nearly two-thirds of manufacturing ﬁrms have implemented Just-in-Time inventory practices by 2008.22 1.5.3.1 A Weighted-average Approach I suppose that shifting from JIC to JIT requires a one-time ﬁxed cost and the switchover is irreversible. The one-period ﬁxed cost is heterogenous among ﬁrms and stochastic. Each period, ﬁrms operating under the JIC system draw their ﬁxed costs from a distribution. The prospective adopters will choose JIT if and only if they receive a better draw such that the beneﬁt of adopting outweighs the cost. Since the cost is stochastic, the decision to 22 CFO research Services conducted a survey among senior ﬁnance executives in North America in the fall of 2008. 39 1.5. Quantitative Analysis Table 1.9: The Role of JIT in Explaining the Rise in Corporate Cash Table 9 presents the adjusted results that take into account the adoption rate of JIT and control for self-selection bias. Panel A summarizes the data moments for the sample periods, 1980 and post-2000, as well as the changes of cash and inventory ratios between those two periods. In Panel B, the ﬁrst two columns report the simulated moments generated under the JIC and JIT environments. Column (3) considers an economy in which two-thirds of the ﬁrms implement JIT, and adjusts the results with a weighted-average approach. Column (4) computes the diﬀerence between Column (3) and Column (1). Panel C repeats the exercises in Panel B, except that it controls for self-selection bias when making the adjustment. Panel A: Data Moments Average cash Average inventory Panel B: Weighted-Average Average cash Average inventory Panel C: Self-Selection Average cash Average inventory Data (1980) Data (post-2000) Change 0.098 0.193 0.251 0.082 0.153 -0.111 (1) (2) (3) (4) JIC JIT Adjusted Results Change 0.099 0.200 0.203 0.084 0.168 0.123 0.069 -0.077 (1) (2) (3) (4) JIC JIT Adjusted Results Change 0.097 0.200 0.205 0.085 0.171 0.127 0.074 -0.073 implement JIT is random and uncorrelated with cash holdings. A weighted average is therefore a reasonable approach to make the adjustments. Given that two-thirds of ﬁrms have switched from JIC to JIT, the adjusted cash and inventory ratios for the economy are 16.8% ( 13 × 0.099 + 23 × 0.203) and 12.3% ( 13 × 0.200 + 23 × 0.084), respectively. Table 1.9 summarizes the results, with Panel A reporting data moments and Panel B reporting the model counterparts. In the data, the average cash ratio has increased by 15.3 percentage points since 1980, from 9.8% to 25.1%. During the same period, the average inventory ratio has decreased by 11.1 percentage points, from 19.3% to 8.2%. Panel B suggests that the implementation of JIT can explain a large share of the observed cash and inventory changes, 45.1% of the cash increase and 69.4% of the input inventory reduction. 40 1.5. Quantitative Analysis 1.5.3.2 Controlling for Self-selection In the subsection above, I suppose that the adoption cost is random and use a weighted-average to evaluate the contribution of JIT adoption on cash hoarding. In this subsection, I perform a robustness check by assuming that the one-period ﬁxed cost, C, is identical to all ﬁrms and modelling the adoption decision to control for the induced self-selection bias. Each period, after the realization of revenue shocks, ﬁrms operating under JIC weigh the expected beneﬁts of the switch-over against the cost and make decisions. If the adoption beneﬁts are greater relative to the cost C, ﬁrms decide to implement JIT. Switching back from JIT to JIC is an unavailable option.23 Assessing the contribution of JIT requires information on the one-period adoption cost C. I calibrate C to match the adoption rate in the data. More speciﬁcally, I simulate a sample of 1000 ﬁrms for 100 periods, by starting from the same initial state {z1 , k1 , c1 , s1 } and drawing 1000 sequences of revenue shocks εz from the same distribution N (0, σz2 ). For the ﬁrst 50 periods, all ﬁrms are restricted to operating under JIC. From the period 51, ﬁrms are allowed to select between JIC and JIT. Once switchover, ﬁrms operate under JIT permanently. Prospective adopters make adoption decisions each period. I choose C = 0.55 such that two-thirds (approximately 67%) of ﬁrms are JIT adopters after 28 periods since the JIT system becomes an option.24 Among those adopters, a large portion are large ﬁrms who have more resources to aﬀord adoption costs. This model implication is consistent with the empirical ﬁndings suggested in [88]. According to a survey conducted among small and large U.S. manufacturers, they ﬁnd that large ﬁrms are more likely to implement JIT systems than small ones. Results are summarized in the column Adjusted in the Panel C of Table 1.9. Those moments are computed based on the simulated data in the postadoption periods (period 71 to period 78). Quantitatively, they are very similar to the ones obtained with the weighted-average method. After controlling for self-selection, JIT contributes 48.4% and 65.8% to the observed 23 The irreversibility assumption can be justiﬁed by the fact that implementing JIT involves physical plant changes as well as changes throughout the whole organization. 24 The lump sum cost C = 0.55 is equivalent to 125% of the average total asset (measured by the sum of capital stock, cash and inventory holdings) under JIC. This number measures both the direct costs of system adoption and the indirect costs of adapting to the new system, for example, including high initial investment (purchasing more sophisticated equipment, information system and softwares, and changing the layout of facilities) and extensive training for employees on diﬀerent machines and in diﬀerent functions. 41 1.5. Quantitative Analysis rise in cash and reduction in inventory.25 1.5.4 Comparison with the Risk-based Explanation Previously, I investigate whether JIT is related to changes in cash holdings and quantify its contribution. I next rely on the model to evaluate the role of increased idiosyncratic risk in explaining the observed cash growth, which is highlighted in previous cash hoarding studies ([11]). I then conduct a comparison between the risk-based explanation (precautionary motive for cash holdings) and the one I propose in this paper (transaction motive for cash holdings). Table 1.10: JIT-based Explanation vs. Risk-based Explanation Table 1.10 summarizes the comparison results between the JIT-based explanation and the riskbased explanation. Panel A presents the parameter values used in each model, while Panel B reports their corresponding simulated moments as well as data moments. Data 1980 Panel A: Parameters ﬁxed adjustment cost (f ) linear equity cost (λ1 ) standard deviation (σ) Panel B: Moments Cash Average cash Variance of cash JIC Data 2000s 0.055 0.10 0.15 JIT-based (JIT) Risk-based (JIC) 0.101 0.04 0.15 0.055 0.10 0.35 0.098 0.099 0.251 0.010 0.203 0.027 0.107 0.029 Inventory Average inventory Variance of inventory 0.193 0.200 0.082 0.001 0.084 0.007 0.166 0.007 Investment Average investment Variance of investment 0.076 0.082 0.044 0.001 0.086 0.001 0.080 0.012 25 The simulated mean and within-ﬁrm variance of cash and inventory ratios under JIC and JIT are reported in the columns JIC and JIT. These moments are derived by setting cost C = 0 in the simulations and dropping the ﬁrst 20 periods under each system to exclude the possible bias introduced by the initial states and transitional periods. That is, the moments in the columns JIC and JIT are computed with simulated data for preadoption periods (period 21 to period 50) and post-adoption periods (period 71 to period 100). 42 1.5. Quantitative Analysis To this end, I recalibrate the standard deviation σ under the JIC system and compute the corresponding average cash ratio as well as other simulated moments. [31] measure the median of ﬁrm-level risk by 10-year centered rolling standard deviation of sales growth. Their measure of risk has grown from 0.15 to 0.21 in the past three decades. Following their approach, the average ﬁrm-level risk measured with my sample increases from 0.26 in the 1980s to 0.32 in the post-2000 period, and the risk measure used in my empirical section goes from 0.08 to 0.18 within the same period. All these three risk measures climb up over time, with the largest increase by a factor of approximately 2. I therefore set the standard deviation σ to 0.35, and keep all other parameters the same as their values in JIC. The results are reported in the last column of Table 1.10. In the economy with a more-than-doubled risk, ﬁrms have an average cash ratio 10.7%, raising the ratio by 0.8 percentage points from the benchmark case. Relative to JIT-implementation which explains 8.1 percentagepoint increase, the rise in ﬁrm level uncertainty accounts for a small share of the cash growth observed in the data. The diﬃculty in generating precautionary cash saving has been discussed in the literature. In the JIC setup, a ﬁrm’s cash ﬂow (source of funds) is perfectly positively correlated with the ﬁrm’s investment opportunities and expenses (use of funds). The ﬁrm therefore has a low incentive to save cash. The risk-based model also underpredicts the declining in the average inventory ratio, missing by a factor of 2. On the other dimensions, since risk doubles and investment opportunity is perfectly correlated with productivity, it is not surprising to see that the risk-based model performs poorly in volatility-related moments as well. The variances of cash, inventory and investment ratios overshoot their data counterparts by a factor of 3, 7 and 12, respectively.26 My ﬁndings on cash are in line with the results found in [19]. They focus on a risk-based explanation. They consider two possible channels through which risk aﬀects cash holdings, by introducing two sources of uncertainty — revenue shocks and expense shocks. They ﬁnd that the rise in cash is mostly attributable to current-period liquidity needs (liquidity/transaction motive) rather than future prospects (precautionary motive). My work derives the same conclusion as [19], but distinguishes itself by specifying what the “liquidity needs” are. More importantly, my work models cash as work26 As a robustness check, I set the standard deviation to 0.45 by tripling the value in the benchmark setup and compute the corresponding average cash and inventory ratios. The former goes up to 14.3% and the latter drops to 14.8%, both of which still undershoot the changes observed in the data. Besides, the trippled risk dramatically drives up the volatility of cash, inventory and investment ratios. 43 1.6. Conclusion ing capital and implies that even if in the absence of uncertainty, ﬁrms would hold cash to facilitate transactions. 1.5.5 Impacts of JIT on Firm Values Can implementing JIT enhance ﬁrm performance? This question is of particular interest to managers and practitioners. Empirical studies that examine this relationship have reported mixed results, but with more and more recent research providing evidence of the success of JIT in improving ﬁrm’s ﬁnancial performance ([54]; [10]; [21]; [60]; [38]; and [68]). I in this section reinvestigate this question by using the model to assess the ﬁrm value changes due to the JIT implementation. One of the main values of this exercise added to the existing JIT literature is its ability to isolate all other elements possibly aﬀecting a ﬁrm’s performance aside from JIT, and to deliver quantitative evaluations. Moreover, previous empirical studies use a three-year or ﬁve-year window to analyze the impact of JIT adoption. Given the long-run nature of the implementation process, it takes a longer period of time for JIT to prove its advantages. This may provide a partial explanation of limited empirical support of positive eﬀects of JIT on ﬁrm performance in early stages. Here, I focus on the long-run impact of JIT implementation and gauge it by comparing the stationary distributions in JIC and JIT systems. Lastly, this exercise also helps to answer the question how ﬁrm characteristics generate heterogeneity in the beneﬁts from JIT implementation, especially for diﬀerent ﬁrm sizes. To quantify the extent to which JIT implementation aﬀects ﬁrm performance, I compare the stationary equity values under JIC and JIT.27 Specifically, I simulate a sample of 1000 ﬁrms for 50 periods under JIC and JIT systems respectively and compute the percentage change in the stationary equity values under each system for each ﬁrm. The ratio represents the value change as a result of JIT. The results are summarized in Table 1.11. 1.6 Conclusion In the past three decades, the U.S. corporate sector has gradually shifted resources from inventory to cash. In this paper, I propose an explanation— the implementation of Just-in-Time (JIT) inventory system — to understand the observed high substitution rate between cash and inventory, and 27 For JIC, I consider the benchmark model, while for JIT, I consider the model taking into account all three changes. 44 1.6. Conclusion Table 1.11: JIT Adoption and Changes in Firm Value Table 1.11 reports the percentage changes in ﬁrm value as a result of JIT adoption and the correlation between the change and ﬁrm size. It summarizes the minimum, mean and maximum value change in three environments: No Lag in Delivery, Increased Inventory Adjustment Costs and Long-term Contracts. mean change minimum change maximum change corr(∆V, k) No Lag in Delivery 110% 71.5% 154% −0.75 Increased Adjustment Costs 46.9% 13.3% 81.7% −0.66 Long-term Contracts 47.4% 14.1% 81.9% −0.67 in turn to shed light on corporate cash hoarding behavior which has attracted extensive attention recently. I begin by providing strong evidence for the importance of Just-in-Time (JIT) system adoption in understanding inventory reduction and cash accumulation. I then develop a structural model to explore how JIT inﬂuences inventory and cash policies and quantify its eﬀects. In the model, I emphasize the transaction motive for cash savings. Adopting JIT helps ﬁrms to eliminate non-value-added inventory; it also leads ﬁrms to allocate released resources to cash, in order to purchase production materials and facilitate operations without tapping into expensive external borrowing. I show that the model reproduces a high negative correlation between cash and inventory, and ﬁnd that JIT adoption can account for at least 45% of the cash increase and 69% of the inventory reduction observed in the data. There is a lively debate on the causes of corporate cash hoarding, raising concerns about possible resource misallocation from physical capital to cash. My results suggest that at least 45% of the accumulated cash can be rationalized as a normal and positive investment. I also use my model to examine whether JIT improves ﬁrm performance. I ﬁnd that all ﬁrms beneﬁt from implementing JIT in the long run, regardless of ﬁrm characteristics. The average beneﬁcial eﬀect is a 47% increase in ﬁrm value. In addition, the magnitude of the eﬀects decreases in ﬁrm size. Smaller ﬁrms proﬁt more from the introduced production ﬂexibility. These ﬁndings suggest that there will be potentially signiﬁcant eﬃciency gains available for the corporate sector from implementing JIT. This paper provides an explanation for the corporate cash hoarding behavior within a partial equilibrium model. In subsequent research, I would like to extend the work to a general equilibrium framework and study the eﬀects of corporate cash holdings on the real economy. 45 Chapter 2 Competition and Risk: When the Choice between Cash and Inventory Matters 2.1 Introduction Cash and inventory are two important components of a ﬁrm’s assets. They account for 33% of total assets for U.S. non-ﬁnancial, non-utility publicly traded ﬁrms in 2009.28 Cash facilitates the daily operation of ﬁrms and hedges against adverse shocks. Inventory provides protection from uncertainty in product demand and is a reversible store of liquidity as it can be converted into cash when necessary. This paper analyzes cash and inventory within a uniﬁed framework to shed new light on corporate liquidity management. Although empirical studies in corporate ﬁnance treat inventory as a substitute to cash by including working capital net of cash as an explanatory variable in cash regressions (see for example, [74], and [11]), almost nothing is known regarding the resource allocation choices between these two margins. This paper contributes to the literature by providing theoretical foundation and empirical support for the cash-inventory tradeoﬀ. This is particularly important to understand in light of the substantial changes in ﬁrms’ portfolio choices in the past thirty years. Figure 2.1 depicts how cash-to-asset and inventory-to-asset ratios vary over time for U.S. nonﬁnancial, non-utility publicly traded ﬁrms. The average cash ratio soars from 9.6% in the mid-1970s to 23.5% in 2009, simultaneously there is a signiﬁcant decline in the average inventory ratio from 24.4% to 9.5%, while the sum of these two ratios is approximately constant over time. The former, cash hoarding behavior, has recently attracted considerable attention from ﬁnancial economists. By contrast, macroeconomists have focused on the 28 This ratio stays roughly constant over the last three decades, as shown in Figure 1 below. 46 .1 .15 ratios .2 .25 .3 .35 2.1. Introduction 1970 1980 1990 year average cash ratio average cash and inventory ratio 2000 2010 average inventory ratio Figure 2.1: Average Cash and Inventory Ratio from 1975 to 2009 in the U.S.. The ﬁgure summarizes the average cash-to-asset ratio, average inventory-to-asset ratio and the sum of those two ratios over time. The sample includes all Compustat ﬁrm-year observations from 1975 to 2009 with positive values for the book value of total assets and sales revenue for non-ﬁnancial and non-utility ﬁrms incorporated in the United States. The cash ratio is measured as the ratio of cash, cash equivalents and short term investments to the book value of total assets. The inventory ratio is measured as the ratio of inventory to the book value of total assets. latter, the signiﬁcant reduction in inventory, and emphasized it as central to an understanding of the Great Moderation.29 Each of these two parallel but separate streams of studies oﬀers an incomplete picture of the systematic portfolio restructuring in corporate sector.30 Analyzing cash and inventory jointly helps to generate important insight neglected in previous studies. In this paper I investigate how ﬁrms manage internal liquidity by allocating resources between cash and inventory to avoid raising costly external ﬁnancing. Using a three-period model, I show that competition in the product market and risk are two key determinants of the tradeoﬀ between saving cash and carrying inventory forward. More speciﬁcally, I set up a model in which a risk-neutral ﬁrm operates in a product market with some degree of market power (captured by the price elasticity of demand) and faces stochas29 The Great Moderation in economics refers to a decline in the U.S. macroeconomic volatility since the mid-1980s. 30 One exception is [11]. They highlight the importance of inventory in understanding the cash dynamics in the past thirty years. 47 2.1. Introduction tic productivity shocks. The ﬁrm has an investment opportunity which can be ﬁnanced through three channels: cash ﬂows generated from selling inventory and newly-produced products, cash holdings, and external equity. The capital market is assumed to be imperfect and modelled by equity issuance costs. To avoid raising expensive external funds, the ﬁrm attempts to meet liquidity needs by saving cash and/or holding inventory. Cash earns a lower rate of interest than the risk-free rate, while inventory earns a rate contingent on the future state of the world and depreciates when carried forward to next period. How do ﬁrms choose between cash and inventory when they face the tradeoﬀ described above? My model implies that the choice between these two margins depends largely on a ﬁrm’s market power and the risk the ﬁrm faces. Firms with greater market power place more weight on inventory relative to cash, and ﬁrms operating in a riskier environment rely more on cash. The intuition for these results is as follows. When ﬁrms face relatively inelastic demand, that is, have greater market power, ﬁrms are able to charge a higher product price without losing customers, which in turn yields a higher expected return from holding inventory. Therefore ﬁrms facing less elastic demand allocate more resources in inventory than those with more elastic demand. When ﬁrms operate in a more uncertain environment, ﬁrms are more likely to experience extreme states of the world, including both low-productivity states and high-productivity states. As a result of the risk aversion induced by costly external borrowing, ﬁrms are more concerned about a liquidity shortage (i.e., low-productivity states of the world). To avoid resorting to expensive external equity, riskier ﬁrms save more liquid assets, mainly in the form of cash. This is because inventory is particularly valuable in fairly low-productivity states, while cash delivers the same return in all states in which there is a liquidity need. Consequently, relative to less risky ﬁrms, riskier ﬁrms have higher cash ratios and lower inventory ratios. The model also provides new insight into inventory holdings. Firms not only hold inventory to avoid stockouts, or to expect capital gains from a price increase, but also treat inventory as an important part of their liquidity instruments. Hence, holding all other relevant factors constant, ﬁnanciallyconstrained ﬁrms choose to hold more inventory, so that they have more internal resources to draw down whenever necessary. I test model implications using a sample of manufacturing ﬁrms constructed from Compustat Industrial Annual ﬁles over the period 1980-2009. When examining how market power and uncertainty aﬀect the choice between cash and inventory, I use the cash-to-inventory ratio as the dependent variable to capture the relative use of cash versus inventory in liquidity man48 2.1. Introduction agement, as suggested in [83]. In order to exclude the possibility that the results found are solely driven by one of these two margins, I also regress cash-to-asset ratio and inventory-to-asset ratio on controls separately. Subsequently, I analyze whether ﬁnancially-constrained ﬁrms behave diﬀerently when managing inventory. This tests the liquidity motive for inventory holdings. The results are strongly supportive of the model predictions. Speciﬁcally, ﬁrms with higher ability to set price over cost rely more heavily on inventory relative to cash when managing liquidity needs, while ﬁrms exposed to more risks act in the opposite way. Financially-constrained ﬁrms on average store more inventory than ﬁnancially-unconstrained ﬁrms with the same pricing power, implying that inventory is indeed treated as a source of funds. The ﬁrst two ﬁndings shed light on the substantial portfolio restructuring shown in Figure 1. They suggest that the rise in market competition (see, for instance, [31] and [56]) and the rise in risk (see, for instance, [22] and [30]) can partially explain the observed resource reallocation from inventory to cash. This paper contributes in several ways to the literature. First, it adds to the previous studies on liquidity management by taking into account an additional form of internal liquidity besides cash, inventory holdings, and explores the interaction between them. There is a rich literature focusing on the precautionary motive for corporate cash. It suggests that in the face of ﬁnancing frictions and uncertainty, ﬁrms accumulate cash to fund future investments and/or to cover current expenses (see, for example, [1], [79], [19] and [53]).31 My paper extends these analyses by incorporating another source of internal funds — inventory, and examines how ﬁrms make choices between inventory and cash in the presence of costly external ﬁnance. To the best of my knowledge, this is the ﬁrst paper to analyze the tradeoﬀs between cash and inventory. Second, this paper helps to understand corporate cash hoarding behav32 ior. [71] and [82] examine the eﬀects of product market competition on 31 Besides the precautionary motive, other main motives for cash saving identiﬁed in the literature include facilitating transaction ([12]), agency issues ([73]), smoothing R&D ([20]), avoiding double taxation of foreign income ([37]), and managing competition from product market (e.g. [46], [71]). In this paper, I only consider the precautionary motive of cash holding. 32 [11] ﬁrst formally document this phenomenon and emphasize inventory, accounts receivable, idiosyncratic risk and R&D investment as four major drivers behind it. Other related papers include [19] and [5]. [19] analyze revenue uncertainty and expense uncertainty, and quantify their eﬀects on corporate cash balance. They ﬁnd that the rise in cash holdings is mainly because of current-period liquidity needs rather than future prospects. 49 2.1. Introduction cash and link cash hoarding to increased market competition. My paper derives the same conclusion, but distinguishes itself from those two studies by presenting an additional mechanism through which market competition shapes cash policy. The above papers suggest that market competition increases the option value of remaining active in the market and therefore ﬁrms hold cash to avoid ineﬃcient closure. My model provides a more concrete story. Increased market competition limits ﬁrms’ pricing ability and in turn reduces ﬁrms’ proﬁtability. This aﬀects cash through two channels: ﬁrst, it lowers cash ﬂows and increases needs for internal resources (cash and inventory) to preserve liquidity; second, it lowers the return on inventory and makes inventory a less desirable source of funds relative to cash. Therefore, as the product market becomes more competitive, ﬁrms prefer to raise cash for liquidity management. I test this hypothesized mechanism and ﬁnd that inventory is indeed positively related to market power. This is consistent with the model prediction. Lastly, this paper contributes to the inventory literature in two dimensions. First, it proposes new explanations for the inventory declining observed in the data. The substantial reduction in inventory holdings is most commonly attributed to Just-in-Time (JIT) philosophy implementation. My analysis suggests that changes in the business environment (market competitiveness and uncertainty) also contribute to the pattern. Second, this paper models and tests a new motive for holding inventory. In macroeconomics, a range of motives have been identiﬁed, including production-smoothing motive ([52], [17] and [64]), stockout-avoidance motive ([57] and [58]), and cost shock-hedge motive ([66], [18] and [80]). This paper models inventory as a source of funds and explores the liquidity motive for inventory holdings. [36], in their empirical analysis, ﬁnd that ﬁnancially-constrained ﬁrms sell inventory to smooth capital investment. In this paper, I present new evidence on the liquidity motive for inventory holdings by showing that ﬁnanciallyconstrained ﬁrms on average keep more inventory. The remainder of the paper is structured as follows. In Section 2, I develop a three-period model, characterize optimal ﬁnancial policies and derive implications. Section 3 contains empirical tests and Section 4 concludes. [5] study ﬁrms’ ﬁnancing choices among debt, equity and ﬁnancial assets and highlight the dividend tax cut as an important factor contributing to the corporate saving increase. 50 2.2. Model 2.2 Model In this section, I present a model in which a risk-neutral ﬁrm builds its liquidity reserves to avoid raising costly external ﬁnancing. Firm’s liquid ˆ which assets can be composed of cash and inventory. Cash earns a rate R is lower than the gross risk-free rate, while inventory depreciates at a rate δ when transferred to next period and earns a rate contingent on the price change between the current and next period. This tradeoﬀ determines the composition of the ﬁrm’s optimal liquidity buﬀers. 2.2.1 Structure The model has three periods, denoted by t = 1, 2 and 3. At period 1, the ﬁrm is endowed with w1 , w1 = 1, units of good, as well as a risky asset (capital) which produces random f (z1 ) units of good in period 2 only. The good is divisible. At period 2, after the realization of a productivity shock z1 , the ﬁrm faces an investment opportunity which costs I, I = 1. This investment opportunity allows the ﬁrm to invest in a risk-free asset which produces deterministic units of good, w3 , at period 3. The ﬁrm pays I out of either internal or external funds or a combination of both. The ﬁrm has two tools to transfer resources internally from period 1 to period 2. One is to save cash out of cash ﬂows. This option earns a rate ˆ > 1, which is lower than the risk-free rate. The alternative is to of return R carry inventory forward. Inventory depreciates at the end of period 1 at a rate δ, 0 < δ < 1, and can be sold to generate cash ﬂows in period 2. If the ﬁrm has insuﬃcient internal resources to ﬁnance the investment in period 2, it can issue equity which incurs unit costs λ. The risky asset produces output in period 2, which is given by f (z1 ) = ez1 , where the productivity shock z1 has a normal distribution with mean µ and variance σ 2 , z1 ∼ N (µ, σ 2 ). The ﬁrm operates in a monopolistically competitive market and faces demand with the following speciﬁcation: Q = P −θ . Here P denotes the price charged by the ﬁrm, while Q is the quantity demanded. Product demand exhibits a constant price elasticity θ, θ > 1.33 To 33 For an equilibrium to exist, the price elasticity of demand must be greater than one. 51 2.2. Model simplify the model, the ﬁrm is only allowed to set prices in period 1, given the endowment level w1 . The unsold products are stored as inventory and carried forward to period 2. In period 2, I assume that all products must be sold. Therefore, price is no longer a choice variable, but a function of the beginning-of-period inventory holdings and the products produced by the risky asset in place. In period 3, product price is exogenous and equal to 1. 2.2.2 Firm’s Problem In period 1, the ﬁrm allocates endowed resources into three choices, cash saving c2 , inventory holding s2 and dividend payment d1 . Given the endowment w1 , the ﬁrm makes its decision on how many units of good to sell to generate cash ﬂows, under a constraint on the quantity of products currently available for sale, w1 . The unsold products are stored as inventory, s2 . They depreciate and are transferred to period 2. The ﬁrm also decides how much ˆ denotes the to save as cash, c2 , out of the cash ﬂows. The parameter R eﬀective rate of return on cash saving. d 1 = p1 q 1 − where c2 , ˆ R q1 = min{w1 , p−θ 1 }, −1 p1 = q 1 θ , c2 ≥ 0, s2 = (1 − δ)(w1 − q1 ) ≥ 0. In period 2, the ﬁrm has an opportunity to invest in a risk-free asset which costs I. If the ﬁrm does not have suﬃcient internal resources to cover the cost, that is, cash saving c2 plus the realized cash ﬂows p2 q2 is less than I, the ﬁrm issues costly equity. If the ﬁrm can aﬀord the cost with the available internal resources, the remaining funds are distributed as dividend. d2 = (1 + λϕ)[p2 q2 + c2 − I], where q2 = f (z1 ) + s2 , −1 p2 = q 2 θ , { 1 if d2 ≤ 0, ϕ= 0 otherwise. 52 2.2. Model The ﬁrst two constraints correspond to the assumption that all products are sold in period 2. The quantity q2 and the price p2 therefore are determined by the shock realization z1 and the beginning-of-period inventory holdings s2 . The indicator function ϕ equals to 1 if the ﬁrm needs to access to the capital market and issue equity, and 0 otherwise. Note that here I assume that w3 is so large that the ﬁrm chooses to invest in period 2 with probability 1. In the last period, the dividend distributed is the cash ﬂows generated by the investment, w3 units of products valued at a price of 1: d3 = w3 . The risk neutral ﬁrm’s objective is to maximize the expected discounted value of future stream of dividends, by choosing optimal inventory holdings s2 and cash saving c2 . The ﬁrm’s problem can be written as follows: max s2 ≥0,c2 ≥0 d1 + βE1 d2 + β 2 E1 d3 , (2.1) 1 where the discount factor β equals to 1+r , r is the risk-free rate, and dividends d1 , d2 and d3 are speciﬁed as above. 2.2.3 Optimal Policy Rules In this subsection, I characterize optimal decision rules for the ﬁrm’s problem and develop the intuition behind them. 2.2.3.1 Cash Solving the optimization problem (2.1) gives the optimal cash saving, which satisﬁes ˆ + λRβEϕ ˆ 1 = Rβ + µ1 , (2.2) { where ϕ= 1 if d2 ≤ 0, 0 otherwise. The left-hand side of equation (2.2) represents the marginal cost of saving an extra unit of cash, that is, forgone dividend in period 1. The right-hand side of the equation is the marginal beneﬁt of cash saving, the sum of discounted expected return (the ﬁrst term) and discounted expected reduction in the cost of equity issuance (the second term). The last term of the righthand side in equation (2.2) is the Lagrange multiplier of the nonnegativity constraint on cash and gives the shadow price of cash holdings. 53 2.2. Model Equation (2.2) shows that the optimal cash saving is positively correlated with the cost of issuing equity and the probability of issuing equity to ﬁnance investment opportunities. Without costly equity issuance, that is, λ = 0, the ﬁrm never saves cash as the return on cash is lower than risk-free rate. As the equity issuance cost and the probability of issuing equity rise, the ﬁrm retains more cash because more money can be saved if it needs to issue equity next period. 2.2.3.2 Inventory Next, I turn to the optimal inventory policy, or equivalently, the ﬁrm’s pricing rule or optimal quantity choice. The optimality condition is given by θ − 1 − θ1 ∂d2 q1 = (1 − δ)βE + µ2 , θ ∂s2 (2.3) where parameter θ is price elasticity of demand. The left-hand side of equation (2.3) gives the marginal cost of carrying one additional unit of −1 θ good forward to period 2, which is θ−1 dollars of forgone revenue and in θ q1 turn foregone dividends in period 1. The right-hand side shows the marginal beneﬁt, which is the expected present marginal value of an additional unit of end-of-period inventory for period 2 after depreciation. The parameter µ2 is the Lagrange multiplier of the nonnegativity constraint on inventory. Substituting the demand function into equation (2.3), I can rewrite the optimal condition in a more familiar form. In the case of an interior solution, equation (2.3) becomes p1 = ∂d2 θ [(1 − δ)βE ]. (θ − 1) ∂s2 (2.4) The above equation describes the optimal pricing rule of a monopolistically competitive ﬁrm. That is, the ﬁrm charges a constant markup over marginal θ cost. Here, the constant markup is θ−1 . The marginal cost is the ﬁrm’s 2 marginal value of an additional unit of inventory, (1 − δ)βE ∂d ∂s2 . Under the assumption that all products must be sold in period 2, equation (2.4) can be written as follows, which relates the charged price in period 1 with the expected price in period 2, θ−1 θ−1 θ−1 p1 = (1 − δ)βE{ p2 } + λ(1 − δ)βE{ϕ p2 }. θ θ θ (2.5) 54 2.2. Model According to equation (2.5), the model presents two reasons for holding inventory. The ﬁrst is a stockout-avoidance motive, captured by the ﬁrst term on the right-hand side of the equation. The ﬁrm makes the carrying decision based on the prospects for beneﬁting from a price increase between the current and next period. This might happen when the ﬁrm expects a large negative productivity shock which will drive up next period’s product price and create an expectation of a gain from holding inventory. The second motivation of holding inventory is to reduce equity issuance costs, as indicated by the second term on the right-hand side of equation (2.5). Inventory is ready for sale to generate cash ﬂows. As a source of funds, it can save expected costs of issuing equity, playing the same role as cash. 2.2.4 2.2.4.1 Choice between Cash and Inventory Deterministic Environment As shown in the model, both cash and inventory are sources of funds used to ﬁnance capital investment and/or cover operation costs. To illustrate and understand how the ﬁrm allocates resources between cash and inventory when managing its liquidity needs, I in this subsection simplify the model by assuming that the risky asset produces ex1 < 1 units of good, which is insuﬃcient to ﬁnance the investment opportunity in period 2 with probability 1. This assumption guarantees that the ﬁrm will transfer resources over period, with the resources composed of cash and/or inventory, and makes it easy to characterize and analyze the cash-inventory tradeoﬀ. The optimality of the ﬁrm’s ﬁnancial policy requires the ﬁrm to be indiﬀerent between cash and inventory. Accordingly, the ﬁrm chooses to carry more inventory when the price change net of depreciation is greater or equal to the eﬀective rate of return on cash holdings: −1 q2 θ (1 − δ) −1 q1 θ ˆ ≥ R. Combined with model assumptions, it can be written as ˆ − (ex1 + s2 )− θ (1 − δ) ≥ R[(1 1 s2 − 1 ) θ ]. 1−δ This inequality implies that the optimal inventory holding is given by s∗ = max{0, a − ex1 a }, 1 + 1−δ (2.6) 55 2.2. Model θ where a = ( 1−δ ˆ ) . Since the return on cash is lower than the gross risk-free R rate, the ﬁrm never chooses to hold cash more than the unﬁlled liquidity shortage. The optimal cash holding then is as follows, c∗ = max{0, 1 − (ex1 + s∗ )1− θ }. 1 (2.7) I next analyze the properties of the equilibrium and explore how diﬀerent parameter values aﬀect the ﬁrm’s optimal ﬁnancial policy. Proposition 5 summarizes the results, which I prove in Appendix B.1. x1 Proposition 5 Deﬁne the cutoﬀ value of θ, θ¯ = ˆ . A ﬁrm’s optiln(1−δ)−ln R mal ﬁnancial policy depends on the extent of its market power, holding costs and future prospects: ¯ s∗ = 0. Firms use cash for liquidity management. 1. If θ ≥ θ, ¯ the optimal inventory holdings s∗ and optimal cash holdings 2. If 1 < θ < θ, ∗ c are characterized by equations (2.6) and (2.7), respectively. 3. The optimal inventory holdings s∗ decrease in the price elasticity of demand θ, inventory holding costs δ and, the eﬀective return on cash holdings ˆ while the optimal cash holdings c∗ increase in θ, δ and R. ˆ R, ¯ 4. The cutoﬀ value θ decreases in inventory holding costs δ, the eﬀective ˆ and the productivity realization x1 in period 2. return on cash holdings R, Proposition 5 states that a ﬁrm will cover its liquidity needs fully with cash when the ﬁrm’s market power is lower than a threshold. If the ﬁrm’s market power is greater than the critical value, the ﬁrm will carry some level of inventory forward, and the rest of the liquidity shortage will be ﬁnanced by cash. Furthermore, the optimal level of inventory holdings increases in the ﬁrm’s market power, while cash moves in the opposite direction. Lastly, ˆ make inventory inventory carrying costs δ, and the cash holding return R undesirable and expensive to hold relative to cash. One point that deserves attention here is that these properties remain unchanged if adding a demand shock to the model.34 The intuition is that the decision on inventory holdings 34 In the presence of a demand shock, the demand function can be written as Q = ez2 P −θ . Assuming that the realization of the demand shock is x2 in period 2, the corresponding optimal inventory holding is given by s∗ = max{0, a − ex1 a }, 1 + 1−δ x1 −x2 θ ¯ with a = ex2 ( 1−δ ˆ ) . The cutoﬀ value θ becomes ln(1−δ)−ln R ˆ . Following the proof in R Appendix B.1, we can show that equilibrium properties stated in Proposition 1 still hold. 56 2.2. Model relies on the price change between two periods, which is determined by the net eﬀect of shifts in demand and supply. Therefore, in the absence of a demand shock, the productivity shock in the model can be interpreted as the relative supply. Also worth noting is that implied by the fourth statement in Proposition 5, in the case of extremely small productivity realization x1 in period 2, or equivalently, extremely large demand realization in period 2, a perfectly competitive ﬁrm (θ¯ → ∞) may transfer inventory across periods. The discussion above highlights the idea that one of the key determinants of a ﬁrm’s choice between cash and inventory to meet future liquidity needs is the ﬁrm’s market power, or the elasticity faced by the ﬁrm. The intuition is that, the degree of market power largely aﬀects ﬁrms’ proﬁtability as well as the return on inventory. When the elasticity of demand is relatively low, ﬁrms are able to set a high price over cost. Therefore they have high gross margins and acquire a high return on holding inventory. As demand becomes more responsive to prices, both the proﬁt and the prospect of gaining from holding inventory drop. To preserve ﬁnancial ﬂexibility, ﬁrms start switching to cash. 2.2.4.2 Stochastic Environment In the following, I remove the assumption made in the previous subsection and analyze the role of risk in determining a ﬁrm’s decisions on carrying inventory and holding cash. I also reexamine the statements presented in Proposition 1 by plotting the comparative statics of the equilibrium with respect to key parameters in the model. To understand how uncertainty aﬀects a ﬁrm’s cash decision, I rewrite the Euler equation for cash by incorporating the assumed uncertainty structure: ln b − µ ˆ + λRβΦ( ˆ 1 = Rβ ) + µ1 , (2.8) σ where Φ(·) is cumulative distribution function (CDF) of standard normal θ distribution, and the parameter b equals to (1 − c∗ ) θ−1 − s∗ . Parameters µ and σ are the mean and standard deviation of the productivity shock z1 . Equation (2.8) suggests that the standard deviation σ shapes the optimal cash holdings by changing the probability of tapping into external funds next period Φ( ln b−µ σ ). As a result of the external ﬁnancing costs, the ﬁrm’s problem becomes strictly concave, which in turn makes the ﬁrm behave like a risk-averse agent. In response to an increase in uncertainty which drives up the probability of issuing equity in the future, the ﬁrm chooses to cut 57 2.2. Model the current-period dividend payment and accumulate more precautionary savings. Proposition presents the result, which is proved in Appendix B.2. Proposition 6 The optimal cash holding increases in a ﬁrm’s risk σ. To analyze the impact of risk on the optimal inventory policy, I express equation (2.5) as θ−1 θ−1 θ−1 p1 = (1−δ)β E{p2 }+λ(1−δ)β {EϕEp2 +Cov(ϕ, p2 )}. (2.9) θ θ θ A rise in risk generates a strong incentive for the ﬁrm to carry more inventory forward. It raises the probability of equity issuance in the future Eϕ, creats µ σ2 a higher expectation on future price Ep2 = e− θ + 2θ2 , and also increases the covariance of the indicator function of equity issuance and product price next period Cov(ϕ, p2 ) (i.e., the ﬁrm is more likely to face ﬁnancial diﬃculties when the productivity is low and thus the product price is high). This result is formalized in Proposition 7 and proved in Appendix B.3. Proposition 7 The optimal inventory holding increases in a ﬁrm’s risk σ. Figure 2.2 illustrates how the ﬁrm responds to changes in the risk σ by allocating resources between cash and inventory diﬀerently. It shows that the eﬀects of uncertainty on the optimal level of cash and inventory are monotone. An increase in risk σ results in simultaneous rises in the cash and inventory holdings. Furthermore, the magnitude of the rise in cash dominates that of inventory increase (i.e., an increases in risk raises the share of cash), because a rise in uncertainty drives up the marginal beneﬁt of inventory holdings chieﬂy through the probability of equity issuance in the future Eϕ, rather than its relative return compared to cash, Ep2 + Cov(ϕ,p2 ) ˆ − R. Eϕ Intuitively, as the productivity shock becomes more volatile, a liquidity shortage is more likely to occur in the future. To avoid raising costly outside funds, the ﬁrm chooses to accumulate more inside resources as a combination of cash and inventory. Inventory is more valuable than cash in extremely low states, that is, when future productivity is lower than some threshold value. Since the probability of future productivity dropping below the cutoﬀ increases as risk rises, the ﬁrm holds more inventory. For the rest states of the world, in which the probability of issuing equity remains positive, the ﬁrm chooses to raise cash balance, by selling inventory today and saving 58 2.2. Model 0.2 optimal inventory and cash 0.18 0.16 0.14 0.12 0.1 inventory cash 0.08 0 0.05 0.1 0.15 0.2 σ 0.25 0.3 0.35 0.4 Figure 2.2: The Eﬀects of Changes in Risk on the Optimal Cash and Inventory Holdings. This ﬁgure depicts the optimal cash and inventory holdings as a function of uncertainty σ. the proceeds of the sale. In this manner, the ﬁrm transfers resources from future states with relatively high cash ﬂows to the present and then channels resources back into all future states. An increase in risk therefore induces the ﬁrm to hoard liquid assets mainly through accumulating cash. One point worth noting is that Proposition 7 is an artifact of my choice of the speciﬁcation of demand function. If a linear demand is assumed, an increase in risk will lower the expected future price by driving up the 1 2 expect future supply eµ+ 2 σ . Then the overall eﬀect of a risk increase on the optimal level of inventory holdings is ambiguous. It depends on which eﬀect dominates — the decline in the expected future price, or the increase in the covariance between the equity issuance probability and the expected price. However, the impact of risk on inventory level plotted in Figure 2.2 is the upper bound, so the conclusion about the changes in the share of cash and inventory in internal liquidity due to a risk increase, which will be tested in the empirical section, is still true. In the presence of uncertainty, all properties of the equilibrium described in Proposition 5 still hold. Figure 2.3 provides a graphical description of the impacts of the price elasticity of demand (θ), inventory carrying costs ˆ and the mean of productivity (δ), the eﬀective return on cash holdings (R), 59 0.2 0.1 0 inventory cash 1 1.5 2 θ 2.5 3 inventory cash 0.4 0.3 0.2 0.1 0 0.05 0.1 δ 0.15 0.2 optimal inventory and cash 0.5 0.3 0.2 inventory cash 0.15 optimal inventory and cash optimal inventory and cash 0.4 optimal inventory and cash 2.2. Model 0.1 1 1.01 1.02 ˆ R 1.03 1.04 0.4 inventory cash 0.3 0.2 0.1 0 −2 −1 µ 0 1 Figure 2.3: Comparative Statics in a Stochastic Environment. This ﬁgure illustrates the impacts of product market environment θ (the top left panel), the eﬀective ˆ (the top right panel), inventory carrying costs δ (the bottom left panel) return on cash R and the expected future productivity µ (the bottom right panel) on the optimal cash and inventory holdings. innovation (µ) on the tradeoﬀ between cash and inventory under a stochastic environment. Illustrated in the upper left panel of Figure 2.3, an increase in θ leads the ﬁrm to reallocate resources from inventory to cash. As θ increases, the demand function becomes ﬂatter, and the degree of responsiveness in demand quantity with respect to price rises. The ﬁrm therefore has lower pricing power and is more likely to experience a cash ﬂow shortfall. To avoid raising expensive external funds, the ﬁrm needs to hold more liquid assets. Considering the ﬁrm has a lower degree of control over price, holding inventory becomes less desirable, so the ﬁrm holds more cash and less inventory. ˆ drives up the value of A rise in the eﬀective rate of return on cash, R, cash saving in each state of the world. The ﬁrm therefore shifts resources from inventory to cash, as shown in the upper right panel. Similarly, a higher inventory carrying cost δ makes it more expensive to hold inventory. As a result, the ﬁrm chooses to transfer less inventory and more cash over 60 2.3. Estimation and Results periods to meet future liquidity needs. The lower right panel of Figure 2.3 illustrates the eﬀect of the expected future supply (captured by the mean productivity shock µ) on the cash and inventory decisions. The larger the mean value of productivity shocks, the higher a cash ﬂow is expected to arrive in the future, and also less likely the ﬁrm will need to issue equity next period. Accordingly, the ﬁrm has a weaker incentive to accumulate liquid assets and reduces both inventory and retaining earnings. In addition, an increase in the mean value of future relative supply drives down the expected price increase and in turn the value of inventory. The ﬁrm thus reduces inventory holdings even further. 2.2.5 Testable Hypothesis The model produces three testable hypotheses on corporate cash and inventory holdings, which I formally test in the next section. Hypothesis 1 A ﬁrm’s choice between cash and inventory for liquidity management is largely driven by its market power. Firms with greater market power rely more heavily on inventory than on cash. Hypothesis 2 Risk is another key determinant of the tradeoﬀ between cash and inventory. As risk rises, ﬁrms respond by storing more resources mainly in the form of cash. Therefore, the share of cash in ﬁrms’ total asset increases, while the share of inventory declines. Aside from these two hypotheses, equation (2.5) suggests that ﬁrms carry inventory forward not only for avoiding stockout or expecting a capital gain, but also for reducing expected external ﬁnancing costs. The former is identiﬁed as one of main motives for holding inventory in the macroeconomics literature. I test and highlight the latter— the liquidity motive for inventory. This implication is depicted in Figure 2.4. Hypothesis 3 Given the same degree of market power, ﬁnancially constrained ﬁrms hold more inventory than non-ﬁnancially-constrained ﬁrms. 2.3 Estimation and Results In Subsection 2.2.4, I prove and illustrate through simulations how a ﬁrm’s cash holding and inventory carrying decisions depend on the price elasticity of demand and uncertainty. In this section, I perform formal empirical tests on those two predictions. Also tested is the model’s implication on the liquidity motive for inventory holdings. 61 2.3. Estimation and Results 0.3 inventoryNFC 0.28 inventoryFC optimal inventory and cash 0.26 0.24 0.22 0.2 0.18 0.16 0.14 0.12 0.1 1 1.5 2 θ 2.5 3 Figure 2.4: Financial Constraints and Inventory Holdings. This ﬁgure shows the optimal inventory holdings for ﬁnancially-constrained ﬁrms (λ > 0) and non-ﬁnanciallyconstrained ﬁrms (λ = 0), as a function of market power. 2.3.1 2.3.1.1 Methodology The Choice between Cash and Inventory I now turn to testing the central insight provided by this model that the choice between cash and inventory can be driven by ﬁrms’ market power and the risk they face. More speciﬁcally, ﬁrms with higher market power (lower price elasticity of demand θ in the model) tend to hold more inventory and less cash relative to ﬁrms with lower market power; and ﬁrms facing higher uncertainty keep more cash and less inventory. To implement a test of these implications, I build on the baseline empirical model in [11] and specify the regression as cash = α0 +α1 market power+α2 risk+α3 f irm size+α4 market to book inventory ∑ ∑ ∑ +α5 cash f low+α6′ X+ industryj + yeart + cohortl +ϵ1 . j t l (2.10) In this regression, following [83], I use cash-to-inventory ratio as the dependent variable to capture the relative use of cash versus inventory in liquidity management. The ratio is calculated as the natural logarithm 62 2.3. Estimation and Results of the ratio of cash and short-term investments to inventory. To proxy a ﬁrm’s market power, I use two measures. One is pseudo markup, reﬂecting a ﬁrm’s ability to make proﬁts. It is measured by sales over the sum of cost of goods sold (COGS) and selling, general and administrative expense (XSGA). The other proxy is a ﬁrm’s market share.35 I construct this variable by computing a ﬁrm’s sales as a share of total sales in the 4-digit Standard Industrial Classiﬁcation (SIC) industry where the ﬁrm operates. I take a natural logarithm of the two proxies of market power. Using a log form for both cash-to-inventory and market power, rather than a level form, is to simplify the interpretation of results. The coeﬃcient α1 can be interpreted as the elasticity, that is, a 1% change in market power results in α1 % change in cash.36 To measure ﬁrm risk, I compute the standard deviation of the annual operating cash ﬂow ratio over the previous ﬁve years, and I also use a log form of it in the regression. Firm size is the natural logarithm of total assets. Market-to-book ratio is measured as the ratio of the sum of market equity and book debt over the book value of total asset. Cash ﬂow is the ratio of operating cash ﬂows over total asset. X are variables commonly used in the cash regression, including capital investment, net working capital, leverage, R&D expenditures, a dividend dummy, and acquisition expenses. A detailed description of these covariates is provided in Appendix B.4. I control for industry ﬁxed eﬀects to use intra-industry variations to identify the eﬀects of diﬀerent characteristics on cash holdings. I also include year ﬁxed eﬀects which are used to capture the common macroeconomic shocks across ﬁrms, and cohort dummies which are constructed based on ﬁrms’ IPO listing dates. The inclusion of cohort ﬁxed eﬀects are motivated by the fact documented in [11] that most recent listed companies on average hold more cash than older cohorts. I use lagged values for variables in X, testing the hypothesis that the beginning-of-period information (that is, the state of one period) has impacts on a ﬁrm’s cash-inventory management. However, it is possible that the error terms are serially correlated, which results in an endogeneity problem and makes coeﬃcient estimates inconsistent. The reason is as follows. 35 The constructed market share is immune to the criticism that focusing on publiclytraded ﬁrms in Compustat instead of all public and private ﬁrms in the industry may provide misleading results on the eﬀect of market competition. The reason is that changes in denominator have no impacts on the intra-industry relative market shares, which are the variations employed to identify the eﬀect of market power on corporate cash and inventory policies. 36 As a robustness check, I also run regressions with both variables expressed in levels. There is no qualitative change in the coeﬃcient of market power that we are interested in here. Results are available upon request. 63 2.3. Estimation and Results In light of the simultaneity between cash, inventory and other investment decisions, although controls are beginning-of-period information, serial correlation in the disturbances corr(ϵt−1 , ϵt ) ̸= 0 leads to a correlation between the error term and covariates, corr(X, ϵt ) ̸= 0, and in turn causes OLS coeﬃcient estimates to be biased. Therefore, serially uncorrelated disturbances are required to ensure the coeﬃcient estimates consistent. I assume corr(ϵt−1 , ϵt ) = 0 for now. In the last subsection, I perform robustness checks by coping with the possible serial correlation in the error terms and by implementing dynamic models. As an attempt to alleviate the concern that the results of regression (2.10) are driven solely by one side, that is, market power and risk aﬀect cash or inventory only, I next examine them separately. I estimate the following two equations: cash = α0 + α1 market power + α2 risk + α3 f irm size + α4 market to book ∑ ∑ ∑ + α5 cash f low + α6′ X + industryj + yeart + cohortl + ϵ2 , t j l (2.11) and inventory = β0 +β1 market power+β2 risk+β3 f irm size+β4 market to book ∑ ∑ ∑ +β5 cash f low+β6′ X+ industryj + yeart + cohortl +ϵ3 . j t l (2.12) The standard errors are corrected by allowing for heteroscedastic and crossequation correlated residuals, in addition to the possible correlation within a ﬁrm cluster. The dependent variables, cash and inventory, are measured as the natural logarithm of the cash-to-asset ratio and inventory-to-asset ratio. Also modiﬁed is the corresponding net working capital included in X. It is now deﬁned as the ratio of working capital net of cash and net of inventory over total asset, respectively. All other covariates are the same as those included in the regression equation (2.10). 2.3.1.2 Liquidity Motive for Inventory Holdings Not only does my model provide insight into the relationship between market power, risk and working capital management (cash and inventory), but it also predicts that, in addition to avoiding future stockout and proﬁting 64 2.3. Estimation and Results from a price increase, ﬁrms carry inventory over time to preserve ﬁnancial ﬂexibility and keep themselves from resorting to costly external funds. This leads to our third testable implication: ﬁnancially-constrained ﬁrms hold more inventory than ﬁnancially-unconstrained ﬁrms with the same degree of market power. The test speciﬁcation is as follows: inventory = γ0 +γ1 market power+γ2 D+γ3 risk+γ4 size+γ5 market to book ∑ ∑ ∑ industryj + yeart + cohortl +ϵ4 . +γ6 cash f low+γ7′ X+ j t l (2.13) Holding other explanatory variables to be the same as those in the baseline regression of inventory (equation [2.12]), I add a dummy variable D to capture the intensity of ﬁnancial constraints. The dummy variable is equal to one if the ﬁrm is classiﬁed as ﬁnancially constrained, and zero otherwise. This dummy variable is a variable of particular interest here. Based on the model prediction that ﬁnancial constraints lead ﬁrms to hold more inventory, we should expect that the coeﬃcient γ2 is positive. To reﬂect the stockout-avoidance motive of corporate inventory holdings, which is commonly emphasized in macroeconomics literature, an alternative speciﬁcation controlling for sales growth is examined as well. To test this model implication, measures of intensity of ﬁnancial frictions are required. Following [44], I use two measures: ﬁrm size deﬁned as the natural logarithm of total asset and Size-Age (SA) index which is computed as: SA = −0.737 ∗ size + 0.043 ∗ size2 − 0.04 ∗ age. To construct the ﬁnancially constrained ﬁrm indicator, for each year, I rank ﬁrms based on their ﬁrm size and SA index and identify ﬁnancially constrained (unconstrained) ﬁrms as being in the bottom (top) 25% of the size distribution and the top (bottom) 25% of the Size-Age (SA) index distribution, respectively.37 2.3.2 Data and Summary Statistics I construct the sample from Compustat Fundamentals Annual ﬁles for the period 1980 to 2009. I focus on manufacturing industry (SIC 2000-3999) and process data following [11]. 37 Also, [49] suggest that ﬁrm size is best suited as a proxy for high costs of external funds which is the deﬁnition of ﬁnancial frictions in my model, while Cleary, White-Wu and Kaplan-Zingales indices seem best suited as proxies for a high need for external funds. 65 2.3. Estimation and Results Table 2.1 reports summary statistics for regression variables. The mean and median cash-to-asset ratio are 20% and 9% respectively, implying a signiﬁcantly positive skew. In contrast to cash ratio, inventory-to-asset ratio is much more symmetric, with the mean 19% slightly greater than median 16%. The explanatory variables of particular interest, markup and market share, exhibit diﬀerent distributions. The former is relatively symmetrically distributed, with the mean equal to 1.03 and the median equal to 1.10. The latter, regardless of levels of industry ﬁneness, is positively skewed. Table 2.1: Summary Statistics Table 2.1 presents descriptive statistics for the variables used in the estimation. The sample is constructed from Compustat Annual Industrial ﬁles over the period 1980-2009. Table 1 reports the mean, median, standard deviation, 25th and 75th percentile, and number of observations of each variable for manufacturing. A detailed deﬁnition of variables is provided in Appendix B.4. Variables Mean Median Std. Dev. 25% 75% Obs. Cash Inventory Markup Market share (4-digit) Market share (3-digit) Size Risk Market-to-Book Cash ﬂow Net working capital (cash) Net working capital (inventory) Capital investment Leverage R&D Dividend dummy Acquisition Sale growth 0.20 0.19 1.03 0.07 0.04 4.31 0.12 2.31 -0.13 0.03 0.04 0.06 0.26 0.13 0.30 0.02 0.10 0.09 0.16 1.10 0.008 0.002 4.20 0.04 1.20 0.06 0.12 0.08 0.04 0.20 0.05 0 0 0.08 0.24 0.14 0.33 0.17 0.12 2.58 0.23 3.71 0.66 0.62 0.65 0.06 0.25 0.22 0.46 0.05 0.47 0.02 0.08 1.00 0.001 0.0001 2.57 0.02 0.78 -0.08 -0.01 -0.05 0.02 0.04 0.02 0 0 -0.05 0.28 0.27 1.17 0.05 0.02 6.00 0.10 2.18 0.12 0.26 0.28 0.07 0.38 0.14 1 0 0.22 86054 85991 79348 86187 86188 86054 59936 74919 86054 85278 85232 85148 85893 60615 86189 82553 77112 2.3.3 2.3.3.1 Results The Choice between Cash and Inventory Columns (1)-(3) in Table 2.2 present pooled-OLS estimates of regression (2.10) and its alternative speciﬁcations, which are identiﬁed with intraindustry variations. Column (1) reports the regression results for a simpliﬁed model, controlling for basic explanatory variables as well as 4-digit SIC industry ﬁxed eﬀects, year ﬁxed eﬀects and cohort ﬁxed eﬀects. Column (2) extends the basic cash model by adding a list of other control variables, whereas Column (3) controls for 3-digit instead of 4-digit SIC industry ﬁxed eﬀects, allowing for a suﬃcient number of ﬁrms in each industry so as to avoid a possible lack of variations in estimation. To address the concern 66 2.3. Estimation and Results over the possibility of spurious regression due to trending variables, Column (4) reports estimation results with Fama-MacBeth approach. Columns (5)(8) repeat the exercises in Columns (1)-(4), using market share as a proxy for market power.38 All heteroskedasticity-consistent standard errors are clustered by ﬁrm. Table 2.2: The Choice between Cash and Inventory Holdings Table 2.2 reports the results of regressions of cash-to-inventory ratio on a ﬁrm’s market power, risk and other commonly-included control variables. Industry, year and cohort ﬁxed eﬀects are included in the regressions and the heteroskedasticity-consistent standard errors reported in parenthesis account for possible correlation within a ﬁrm cluster. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1%, respectively. Variables Markup (1) Pooled OLS (2) Pooled OLS -0.74*** (0.08) -0.57*** (0.07) Dependent variable: Cash-to-Inventory (3) (4) (5) (6) Pooled Fama Pooled Pooled OLS MacBeth OLS OLS -0.57*** (0.07) Size Market-to-Book Cash ﬂow 34,289 0.44 34,289 0.43 34,800 0.48 Yes 32,803 0.30 32,803 0.37 32,803 0.35 33,300 0.40 34,289 0.38 Acquisition Observations R-squared Yes Yes Yes Yes Yes Yes Dividend Industry FE (3-digit) Industry FE (4-digit) Year FE Cohort Yes Yes Yes Yes Yes R&D 0.15*** (0.01) 0.16*** (0.01) 0.16*** (0.01) 0.64*** (0.08) -0.01 (0.07) -1.48*** (0.32) -2.54*** (0.11) 0.58*** (0.25) -0.11*** (0.02) -0.80*** (0.26) Yes Yes Yes Yes Yes Yes Leverage -0.96*** (0.02) 0.15*** (0.01) 1.05*** (0.02) 0.12*** (0.01) 0.84*** (0.10) -0.12* (0.06) -1.47*** (0.31) -2.33*** (0.11) 1.04*** (0.19) -0.02 (0.02) -1.03*** (0.24) -0.62*** (0.02) 0.18*** (0.02) 0.72*** (0.02) 0.10*** (0.01) 0.46*** (0.04) -0.10* (0.06) -1.37*** (0.31) -2.34*** (0.10) 1.35*** (0.16) -0.07 (0.05) -1.46*** (0.18) 0.18*** (0.02) 0.18*** (0.01) 0.12*** (0.01) 0.57*** (0.06) -0.03 (0.06) -1.50*** (0.34) -2.55*** (0.11) 1.08*** (0.20) -0.12*** (0.05) -1.12*** (0.18) Capital investment -0.76*** (0.03) 0.18*** (0.02) 0.87*** (0.03) 0.10*** (0.01) 0.50*** (0.04) -0.11* (0.06) -1.36*** (0.32) -2.40*** (0.10) 1.52*** (0.17) -0.08* (0.05) -1.43*** (0.18) -0.69*** (0.03) 0.15*** (0.02) 0.75*** (0.03) 0.11*** (0.01) 0.57*** (0.04) 0.13*** (0.02) 0.14*** (0.01) 0.13*** (0.01) 0.84*** (0.06) Net working capital 0.19*** (0.02) 0.19*** (0.01) 0.12*** (0.01) 0.58*** (0.06) -0.02 (0.06) -1.44*** (0.36) -2.63*** (0.11) 1.43*** (0.19) -0.16*** (0.05) -1.11*** (0.19) (8) Fama MacBeth -0.30** (0.12) Market share Risk (7) Pooled OLS Yes Yes Yes Yes In accordance with model predictions, the results in Column (1) suggest that cash-to-inventory ratio declines with ﬁrms’ ability to set price over cost, and increases with uncertainty that ﬁrms face. A 1% decrease in a ﬁrm’s markup leads to an approximate 0.74% increase in a ﬁrm’s cash-to-inventory ratio, while a 1% increase in a ﬁrm’s risk pushes the ratio up by 0.13%. The results are statistically and economically signiﬁcant. 38 4-digit and 3-digit market shares are constructed separately. 67 2.3. Estimation and Results Besides, the estimation results are robust with respect to diﬀerent speciﬁcations, sources of variations and estimation methods. The estimated coeﬃcient on risk vary little across columns, moving between the lowest 0.13 and the highest 0.19. The estimates of the coeﬃcient on market power are also quantitatively similar. In Column (2), after controlling for a host of other explanatory variables, the value slightly changes to −0.57. With 3-digit SIC industry ﬁxed eﬀects, the coeﬃcient of markup is unchanged compared to the one estimated in Column (2). In Columns (5)-(7), the economic magnitude of the coeﬃcient on market share is larger relative to Columns (1)-(3), from −0.62 to −0.76. Estimating the regression equation (2.10) with Fama-MacBeth method conﬁrms the robustness of the results. As a further test of model implications, I regress cash and inventory separately on control variables included in the regression (2.10) and then conduct a joint test of whether market power and risk aﬀect cash and inventory holdings in a way consistent with the theory. This helps to exclude the possibility that the results obtained in Table 2.2 are mostly aﬀected by one margin rather than both. The standard errors are heteroskedasticity and autocorrelation consistent (HAC) and taken into account contemporaneous cross-equation correlations of the error terms. The results are reported in Tables 2.3-2.4. Table 2.3 uses markup to measure market power. Estimation results support model predictions: a ﬁrm’s cash holdings decrease with its market power but increase with risk, while inventory behaves in an opposite manner. The coeﬃcients on markup and risk in the cash regression have the correct signs and are statistically signiﬁcant at the one percentage level in all speciﬁcations. The results suggest that the elasticity of cash holdings to market power lies in the range of 0.19% to 0.34% and the elasticity of cash holdings to risk ranges from 0.05% to 0.10%. This implies that cash policy is at least three times more sensitive to product market environment than to risk. The estimated coeﬃcients on other covariates conform with those in previous studies, except for ﬁrm size. The diﬀerence can arise due to the information on market power contained in the variable ﬁrm size. Column (2) reports the results for inventory. The coeﬃcient on markup indicates that a 1% decrease in a ﬁrm’s markup results in a 0.42% drop in a ﬁrm’s inventory stock. The inclusion of other covariates (Column [4]) and the change of across-group variations (Column [6]) overall hardly aﬀect the estimation results. The elasticity drops only slightly to 0.36% and 0.35%, respectively. The response of inventory to risk is also consistent with the model prediction. On average, a 1% rise in risk leads to a 0.08% drop in inventory holdings. 68 2.3. Estimation and Results Table 2.3: OLS Regression of Cash and Inventory: by Mark-up Table 2.3 reports the results of regressions of cash and inventory holdings separately on a ﬁrm’s market power proxies by markup, risk and other commonly-included control variables. Industry, year and cohort ﬁxed eﬀects are included in the regressions and the heteroskedasticity-consistent standard errors reported in parenthesis account for possible correlation within a ﬁrm cluster. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1%, respectively. Variables Markup Risk Size Market-to-Book Cash ﬂow (1) cash (2) inventory (3) cash (4) inventory (5) cash (6) inventory -0.34*** (0.04) 0.05*** (0.02) 0.02** (0.01) 0.09*** (0.01) 0.53*** (0.04) 0.42*** (0.05) -0.07*** (0.01) -0.12*** (0.01) -0.04*** (0.00) -0.26*** (0.03) -0.19*** (0.04) 0.09** (0.01) 0.06*** (0.01) 0.07*** (0.01) 0.40*** (0.04) -0.48*** (0.05) -2.48*** (0.24) -2.45*** (0.08) 0.64*** (0.12) -0.08** (0.04) -1.70*** (0.15) 0.36*** (0.04) -0.08*** (0.01) -0.11*** (0.01) -0.04*** (0.00) -0.12*** (0.03) -0.41*** (0.04) -1.12*** (0.17) -0.05 (0.05) -0.23** (0.12) 0.04* (0.02) -0.59*** (0.08) -0.20*** (0.04) 0.10*** (0.01) 0.06*** (0.01) 0.07*** (0.01) 0.41*** (0.04) -0.50*** (0.05) -2.53*** (0.25) -2.54*** (0.08) 0.84*** (0.12) -0.11*** (0.04) -1.72*** (0.15) 0.35*** (0.04) -0.08*** (0.01) -0.12*** (0.01) -0.04*** (0.00) -0.12*** (0.03) -0.44*** (0.04) -1.23*** (0.17) -0.07 (0.05) -0.34*** (0.11) 0.05** (0.02) -0.61*** (0.08) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Net working capital Capital investment Leverage R&D Dividend Acquisition Industry FE (3-digit) Industry FE (4-digit) Year FE Cohort Observations R-squared Joint test: α1 =0, β1 =0 Joint test: α2 =0, β2 =0 33,590 33,590 0.25 0.33 p-value=0.00 p-value=0.00 33,590 33,590 0.35 0.36 p-value=0.00 p-value=0.00 33,590 33,590 0.33 0.34 p-value=0.00 p-value=0.00 Table 2.4 replicates the analysis in Table 2.3 with market share as the proxy for market power. The results suggest that the relationship between market power, risk and working capital management still holds. The estimated coeﬃcients of market share and risk have the expected signs and are stable across various regressions. Compared to the estimates reported in Table 2.3, the results are quantitatively similar. As robustness checks, I next turn to ﬁrm ﬁxed eﬀect and Fama-MacBeth approaches. In Table 2.5, I replace industry dummy variables in regression equations (2.11) and (2.12) with ﬁrm-speciﬁc dummy variables. The corresponding results are identiﬁed from within-ﬁrm over-time variations. Table 6 presents results with Fama-MacBeth method which uses cross-sectional variations to estimate coeﬃcients. Columns (1) and (2) in each table report the results for the cash model with market power proxied by markup and 69 2.3. Estimation and Results Table 2.4: Pooled OLS Regression of Cash and Inventory: by Market Share Table 2.4 reports the results of regressions of cash and inventory holdings separately on a ﬁrm’s market power proxies by market share, risk and other commonly-included control variables. Industry, year and cohort ﬁxed eﬀects are included in the regressions and the heteroskedasticityconsistent standard errors reported in parenthesis account for possible correlation within a ﬁrm cluster. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1%, respectively. Variables Market share Risk Size Market-to-Book Cash ﬂow (1) cash (2) inventory (3) cash (4) inventory (5) cash (6) inventory -0.30*** (0.01) 0.06*** (0.01) 0.29*** (0.02) 0.07*** (0.00) 0.37*** (0.03) 0.37*** (0.02) -0.09*** (0.01) -0.44*** (0.01) -0.03*** (0.00) -0.11*** (0.02) -0.24*** (0.01) 0.08*** (0.01) 0.26*** (0.01) 0.06*** (0.00) 0.32*** (0.03) -0.38*** (0.04) -2.29*** (0.22) -2.19*** (0.08) 0.68*** (0.08) -0.05 (0.04) -1.88*** (0.15) 0.33*** (0.01) -0.09*** (0.01) -0.40*** (0.01) -0.03*** (0.00) -0.05*** (0.02) -0.32*** (0.03) -1.01*** (0.15) -0.07 (0.05) -0.46*** (0.10) 0.02 (0.02) -0.38*** (0.08) -0.27*** (0.01) 0.09*** (0.01) 0.30*** (0.02) 0.06*** (0.00) 0.34*** (0.03) -0.40*** (0.04) -2.37*** (0.23) -2.27*** (0.08) 0.77*** (0.08) -0.07** (0.04) -1.89*** (0.15) 0.42*** (0.02) -0.08*** (0.01) -0.49*** (0.02) -0.03*** (0.00) -0.07*** (0.02) -0.30*** (0.03) -1.08*** (0.15) -0.07 (0.05) -0.49*** (0.09) 0.02 (0.02) -0.40*** (0.08) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Net working capital Capital investment Leverage R&D Dividend Acquisition Industry FE (3-digit) Industry FE (4-digit) Year FE Cohort Observations R-squared Joint test: α1 =0, β1 =0 Joint test: α2 =0, β2 =0 36,383 36,383 0.32 0.42 p-value=0.00 p-value=0.00 36,383 36,383 0.41 0.44 p-value=0.00 p-value=0.00 36,383 36,383 0.39 0.44 p-value=0.00 p-value=0.00 market share respectively. Columns (3) and (4) show results for inventory regressions. In Table 2.5, the estimated eﬀects of market power and risk are qualitatively the same as those in Tables 2.3 and 2.4, although smaller in magnitude. The coeﬃcients on some other explanatory factors yield slightly diﬀerent results in sign and signiﬁcance, including R&D expenditure, dividend distribution dummy and acquisition expenses. Table 2.6 delivers the same pattern with only one exception. Besides, the identiﬁed impacts of product market environment and uncertainty on cash and inventory are even stronger, relative to those shown in Table 2.3-2.5. In summary, Tables 2.2-2.6 provide strong evidence in support of model predictions on the relations between market power, risk, and the cashinventory management. I next turn to the last hypothesis about the motive 70 2.3. Estimation and Results Table 2.5: Firm Fixed Eﬀect Regression of Cash and Inventory Table 2.5 presents the results of regressing cash and inventory ratios on a ﬁrm’s market power and risk with ﬁrm ﬁxed eﬀect speciﬁcations. Standard errors reported in parenthesis account for clustering by ﬁrm. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1%, respectively. A: Cash Variables (1) Markup -0.10*** (0.04) Market share Risk Size Market-to-Book Cash ﬂow Net working capital Capital investment Leverage R&D Dividend Acquisition Firm FE Year FE Observations R-squared (2) (3) B: Inventory (4) 0.28*** (0.05) 0.06*** (0.01) 0.04* (0.02) 0.05*** (0.00) 0.25*** (0.04) -0.31*** (0.04) -2.00*** (0.20) -1.48*** (0.08) -0.08 (0.12) 0.00 (0.04) -1.22*** (0.13) -0.16*** (0.01) 0.06*** (0.01) 0.13*** (0.02) 0.04*** (0.00) 0.24*** (0.03) -0.22*** (0.04) -1.90*** (0.19) -1.28*** (0.07) 0.02 (0.07) 0.00 (0.04) -1.22*** (0.12) -0.05*** (0.01) -0.18*** (0.01) -0.01*** (0.00) -0.12*** (0.03) -0.13*** (0.03) -0.13 (0.12) -0.09** (0.04) -0.08 (0.11) 0.05*** (0.02) 0.01 (0.05) 0.21*** (0.01) -0.05*** (0.01) -0.32*** (0.02) -0.01** (0.00) -0.07*** (0.02) -0.09*** (0.03) -0.13 (0.11) -0.11*** (0.04) -0.09 (0.09) 0.06*** (0.02) 0.02 (0.05) Yes Yes Yes Yes Yes Yes Yes Yes 33,388 0.09 36,176 0.09 33,005 0.15 34,496 0.18 for holding inventory. 2.3.3.2 Inventory and Financing Frictions In the model, the presence of ﬁnancial frictions generates a distinct reason for holding inventory from those identiﬁed and stressed in the previous literature (e.g., the production-smoothing, the stockout-avoidance and the transaction motives), namely, the liquidity motive. To test this implication, I run regression (2.13) and report the results in Tables 2.7-2.8. I pay particular attention to the coeﬃcient on ﬁnancially constrained ﬁrm dummy. In the regression, ﬁnancing frictions are measured by ﬁrm size and SA index as suggested in [44]. The dummy variable is equal to one if a ﬁrm is classiﬁed as ﬁnancially constrained, and is equal to zero otherwise. According to the ﬁrst two rows of Column (1) in Table 2.7, on average, 71 2.3. Estimation and Results Table 2.6: Fama-MacBeth Regression of Cash and Inventory Table 2.6 presents the results of regressing cash and inventory ratios on a ﬁrm’s market power and risk with Fama-MacBeth approach. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1%, respectively.. A: Cash Variables (1) Markup 0.05 (0.09) Market share Risk Size Market-to-Book Cash ﬂow Net working capital Capital investment Leverage R&D Dividend Acquisition Industry FE (4-digit) Year FE Cohort Observations R-squared (2) (3) B: Inventory (4) 0.29*** (0.05) 0.05*** (0.01) 0.03*** (0.01) 0.09*** (0.01) 0.51*** (0.06) -0.95*** (0.13) -2.82*** (0.20) -2.54*** (0.11) -0.67*** (0.14) -0.08*** (0.02) -1.46*** (0.20) -0.33*** (0.02) 0.06*** (0.01) 0.35*** (0.02) 0.07*** (0.01) 0.66*** (0.09) -0.82*** (0.12) -2.63*** (0.20) -2.31*** (0.11) 0.71*** (0.10) -0.04*** (0.02) -1.56*** (0.21) -0.08*** (0.01) -0.11*** (0.01) -0.04*** (0.01) -0.10*** (0.04) -0.62*** (0.06) -1.32*** (0.16) -0.15*** (0.04) 0.02 (0.14) 0.03*** (0.01) -0.62*** (0.09) 0.48*** (0.02) -0.07*** (0.01) -0.55*** (0.02) -0.03** (0.00) -0.11*** (0.03) -0.44*** (0.05) -1.13*** (0.16) -0.11*** (0.03) -0.25** (0.11) -0.01 (0.01) -0.45*** (0.08) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 33,886 0.43 36,691 0.46 33,508 0.42 35,014 0.51 ﬁrms have a regular tendency to raise their inventory by 0.27% in response to a 1% increase in market power; furthermore, holding market power constant, ﬁnancial frictions contribute to 1.08 percentage points increase in inventory ratio. These results are again robust when using 4-digit SIC industry ﬁxed eﬀects in Column (2), controlling for stockout-avoidance motive in Column (3), estimating with Fama-MacBeth approach in Column (4), and using another proxy for market power in Columns (5)-(8). The coeﬃcients on the ﬁnancially-constrained ﬁrm dummy variable are all positive and statistically signiﬁcant. I re-estimate model (2.13) with an alternative measure of ﬁnancial friction intensity in Table 2.8. The results are quantitatively similar. The results shown in Tables 2.7 and 2.8 support the model prediction that inventory is an important component of a ﬁrm’s ﬁnancial policy. [36] derive a similar conclusion by showing the signiﬁcant negative impacts of inventory investment on capital investment. They argue that in the presence 72 2.3. Estimation and Results Table 2.7: Inventory For Constrained and Unconstrained Firms: by Firm Size Table 2.7 presents the results of regressing inventory on a ﬁnancially-constrained ﬁrm dummy. Financing frictions are measured by ﬁrm size. Robust standard errors reported in parenthesis are clustered at ﬁrm level. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1% respectively. Variables Markup (1) Pooled OLS (2) Pooled OLS (3) Pooled OLS (4) Fama MacBeth (5) Pooled OLS (6) Pooled OLS (7) Pooled OLS (8) Fama McBeth 0.27*** (0.05) 0.29*** (0.05) 0.29*** (0.05) 0.29*** (0.06) 0.08*** (0.03) -0.09*** (0.01) 0.09** (0.04) -0.10*** (0.01) -0.11*** (0.01) -0.04*** (0.01) -0.12*** (0.04) -0.28*** (0.04) -0.71*** (0.23) -0.02 (0.06) -0.30** (0.14) 0.10*** (0.03) -0.62*** (0.11) -0.11*** (0.01) -0.04*** (0.01) -0.12*** (0.03) -0.27*** (0.04) -0.78*** (0.21) 0.02 (0.06) -0.19 (0.14) 0.10*** (0.03) -0.54*** (0.10) 0.10** (0.04) -0.10*** (0.01) 0.04 (0.03) -0.11*** (0.01) -0.04*** (0.01) -0.13*** (0.04) -0.27*** (0.04) -0.78*** (0.21) 0.02 (0.06) -0.20 (0.14) 0.10*** (0.03) -0.56*** (0.10) 0.06*** (0.01) -0.09** (0.01) 0.35*** (0.02) 0.07*** (0.02) -0.08*** (0.01) 0.28*** (0.02) 0.19*** (0.04) -0.09*** (0.01) 0.11*** (0.01) 0.09*** (0.02) -0.08*** (0.01) -0.10*** (0.01) -0.04*** (0.00) 0.08* (0.05) -0.56*** (0.07) -0.95*** (0.20) -0.12*** (0.04) -0.00 (0.16) 0.08*** (0.01) -0.55*** (0.10) -0.43*** (0.02) -0.02*** (0.01) -0.06*** (0.03) -0.18*** (0.03) -0.69*** (0.20) -0.01 (0.06) -0.37*** (0.12) 0.07*** (0.03) -0.44*** (0.10) -0.34*** (0.02) -0.02*** (0.01) -0.06** (0.03) -0.20*** (0.03) -0.71*** (0.19) 0.02 (0.06) -0.37*** (0.12) 0.07*** (0.03) -0.39*** (0.09) 0.28*** (0.02) 0.19*** (0.04) -0.10*** (0.01) 0.03 (0.03) -0.34*** (0.02) -0.02*** (0.01) -0.06** (0.03) -0.20*** (0.03) -0.72*** (0.19) 0.02 (0.06) -0.38*** (0.12) 0.08*** (0.03) -0.41*** (0.10) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 17,627 0.36 18,310 0.39 18,310 0.39 17,889 0.44 18,407 0.42 19,112 0.43 19,112 0.43 18,678 0.44 Market share FC dummy Risk Sales growth Size Market-to-Book Cash ﬂow Net working capital Capital investment Leverage R&D Dividend Acquisition Industry FE (3-digit) Industry FE (4-digit) Year FE Cohort Observations R-squared Yes -0.18*** (0.01) -0.03*** (0.00) 0.05 (0.03) -0.50*** (0.07) -0.73*** (0.21) -0.11*** (0.04) -0.26* (0.15) 0.08*** (0.01) -0.48*** (0.10) Yes of ﬁnance constraints, ﬁrms decumulate inventory to smooth capital investment for a limited source of funds. My work complements [36] by providing a piece of new evidence on the ﬁnancing role of inventory. That is, caeteris paribus, ﬁrms facing ﬁnance constraints store more inventory so that they have more internal resources to draw down whenever necessary. 2.3.4 Robustness Checks As mentioned in Subsection 2.3.1, if the error terms in equation (2.10) are serially correlated, the OLS coeﬃcient estimates obtained are biased. In this subsection, I undertake robustness checks by allowing for serially correlated error terms. I also estimate a dynamic model for the cash-to-inventory 73 2.3. Estimation and Results Table 2.8: Inventory For Constrained and Unconstrained Firms: by SA Index Table 2.8 presents the results of regressing inventory on a ﬁnancially-constrained ﬁrm dummy. Financing frictions are measured by Size-Age (SA) index. Robust standard errors reported in parenthesis are clustered at ﬁrm level. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1% respectively. Variables Markup (1) Pooled OLS (2) Pooled OLS (3) Pooled OLS (4) Fama MacBeth (5) Pooled OLS (6) Pooled OLS (7) Pooled OLS (8) Fama McBeth 0.31*** (0.06) 0.33*** (0.05) 0.32*** (0.05) 0.30*** (0.04) 0.09** (0.04) -0.09*** (0.01) -0.11*** (0.01) 0.08** (0.04) -0.09*** (0.01) -0.11*** (0.01) -0.04*** (0.01) -0.13*** (0.04) -0.30*** (0.04) -0.86*** (0.23) -0.03 (0.06) -0.33** (0.15) 0.08*** (0.03) -0.50*** (0.10) -0.03*** (0.01) -0.12*** (0.04) -0.29*** (0.04) -0.81*** (0.21) 0.01 (0.06) -0.22 (0.15) 0.08*** (0.03) -0.49*** (0.10) 0.08** (0.04) -0.09*** (0.01) -0.11*** (0.01) 0.05* (0.03) -0.04*** (0.01) -0.12*** (0.04) -0.29*** (0.04) -0.81*** (0.21) 0.02 (0.06) -0.24 (0.14) 0.08*** (0.03) -0.52*** (0.10) 0.06 (0.04) -0.08*** (0.01) -0.10*** (0.00) 0.36*** (0.02) 0.11*** (0.04) -0.08*** (0.01) -0.42*** (0.02) 0.29*** (0.02) 0.12*** (0.04) -0.09*** (0.01) -0.36*** (0.02) 0.10*** (0.01) 0.11*** (0.04) -0.07*** (0.01) -0.16*** (0.01) -0.04*** (0.01) -0.09* (0.04) -0.56*** (0.07) -1.08*** (0.24) -0.12*** (0.04) 0.02 (0.18) 0.07*** (0.01) -0.57*** (0.13) -0.02*** (0.01) -0.07*** (0.03) -0.19*** (0.03) -0.82*** (0.20) -0.02 (0.06) -0.42*** (0.13) 0.05** (0.03) -0.34*** (0.10) -0.02*** (0.01) -0.05* (0.03) -0.21*** (0.03) -0.73*** (0.19) 0.02 (0.06) -0.38*** (0.13) 0.06** (0.03) -0.32*** (0.10) 0.29*** (0.02) 0.13*** (0.04) -0.09*** (0.01) -0.36*** (0.02) 0.04 (0.03) -0.02*** (0.01) -0.05* (0.03) -0.21*** (0.03) -0.73*** (0.19) 0.02 (0.06) -0.40*** (0.13) 0.06** (0.03) -0.35*** (0.10) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 18,139 0.35 18,770 0.39 18,770 0.39 18,427 0.44 18,893 0.42 19,544 0.44 19,544 0.44 19,158 0.44 Market share FC dummy Risk Size Sales growth Market-to-Book Cash ﬂow Net working capital Capital investment Leverage R&D Dividend Acquisition Industry FE (3-digit) Industry FE (4-digit) Year FE Cohort Observations R-squared Yes -0.03*** (0.00) -0.05 (0.04) -0.50*** (0.07) -0.84*** (0.23) -0.10*** (0.04) -0.25 (0.16) 0.06*** (0.01) -0.47*** (0.13) Yes regression, considering the possibility that the model presented in previous sections are misspeciﬁed.39 With the assumption that the disturbances in equation (2.10) follow an AR(1) process, the cash-to-inventory model can be written as cash = α0 + α1 market power + α2 risk + α3 f irm size + α4 market to inventory ∑ ∑ book + α5 cash f low + α6′ X−1 + industryj + yeart j t 39 For instance, [34] estimate a target adjustment model to analyze the evolution of corporate cash holdings. 74 2.3. Estimation and Results + ∑ cohortl + ϵ4 , (2.14) l where ϵ4,t = ρϵ4,t−1 + νt and νt is independent and identically distributed. Also estimated is the dynamic model speciﬁed as follows: cash cash = α0 +α1 market power+α2 risk+α3 +α4 f irm size inventory inventory −1 +α5 market to book + α6 cash f low + α7′ X + α8′ X−1 + ∑ j industryj + ∑ t yeart + ∑ cohortl + ϵ5 , (2.15) l Regression equation (2.15) is the extension of equation (2.10), with the incash clusion of lagged dependent variable, inventory , and lagged explanatory −1 variables X, X−1 , on the right-hand side to account for the dynamics of ﬁnancial policy and time adjustment process. The coeﬃcient α1 indicates the short-term impact of market power on corporate cash and inventory α1 decisions, while the long-run eﬀect is given by 1−α . 3 Table 2.9 summarizes the results. In Column (1), I model AR(1)-inerrors and follow the Cochrane-Orcutt procedure to estimate the persistence parameter ρ. Column (2) presents the results of the dynamic regression (equation [2.15]) with 4-digit SIC industry ﬁxed eﬀects and time ﬁxed eﬀects. In Column (3), I consider an alternative speciﬁcation by adding more lagged dependent and independent variables as explanatory factors to regression (2.15). Columns (4)-(6) re-estimate the regressions in Columns (1)-(3) with market share. As shown in Column (1) of Table 2.9, allowing for serial correlation in the error terms gives qualitatively similar results to those reported in Tables 2.2. The coeﬃcient on markup is still negative and signiﬁcant at the 1% level. In Column (2), the coeﬃcient on markup captures the short-term eﬀect of market power. The result suggests that a 1% decline in markup raises cash-inventory ratio by 0.12% in the short run, while in the long run, the overall impact is a 0.46% increase. Changing the lag structure by including more lagged dependent and independent variables barely changes short-run and long-run impacts of markup on cash-inventory ratio. According to the estimates in Column (3), the short-term eﬀect of markup drops slightly to -0.11%; overall, a 1% decrease in markup drives up 0.50% of cash-inventory ratio. The estimated eﬀect remains unchanged when market share is used as the alternative measure of market power. 75 2.3. Estimation and Results Table 2.9: Holdings Robustness Checks: The Choice between Cash and Inventory Table 2.9 reports the results of dynamic cash-inventory models. The heteroskedasticity-consistent standard errors reported in parenthesis account for possible correlation within a ﬁrm cluster. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1% respectively. Variables Markup (1) Cochrane-Orcutt (2) Dynamic (3) Dynamic -0.40*** (0.05) -0.12*** (0.04) -0.11*** (0.03) Market share Risk 0.09*** (0.01) 0.05*** (0.01) 0.74*** (0.01) 0.16*** (0.01) 0.04*** (0.01) 0.37*** (0.05) -0.02 (0.05) -0.65*** (0.19) -0.81*** (0.07) 0.96*** (0.15) -0.05 (0.03) -0.16 (0.10) 0.05*** (0.00) 0.05*** (0.00) 0.39*** (0.04) -0.14*** (0.06) -0.88*** (0.19) -0.11 (0.08) 0.69*** (0.17) 0.03 (0.03) -0.64*** (0.12) Lagged cash (ﬁrst order) Lagged cash (second order) Size Market-to-Book Cash ﬂow Net working capital Capital investment Leverage R&D Dividend Acquisition Lagged X (ﬁrst order) Lagged X (second order) Industry FE (4-digit) Year FE Cohort Observations R-squared Yes 0.04*** (0.01) 0.62*** (0.01) 0.16*** (0.01) 0.04*** (0.00) 0.04*** (0.00) 0.40*** (0.04) -0.12** (0.06) -1.20*** (0.20) -0.20*** (0.08) 0.36** (0.16) -0.02 (0.03) -1.06*** (0.12) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 32,803 0.15 30,568 0.72 28,481 0.73 (4) Cochrane-Orcutt (5) Dynamic (6) Dynamic -0.47*** (0.02) 0.09*** (0.01) -0.19*** (0.01) 0.05*** (0.01) 0.73*** (0.01) 0.56*** (0.02) 0.04*** (0.01) 0.28*** (0.04) -0.06 (0.05) -0.70*** (0.18) -0.70*** (0.07) 0.83*** (0.12) -0.03 (0.03) -0.07 (0.10) 0.22*** (0.01) 0.04*** (0.00) 0.42*** (0.03) -0.19*** (0.05) -0.82*** (0.19) -0.12 (0.07) 0.77*** (0.14) -0.02 (0.03) -0.69*** (0.12) -0.18*** (0.01) 0.04*** (0.01) 0.61*** (0.01) 0.15*** (0.01) 0.21*** (0.01) 0.04*** (0.00) 0.42*** (0.03) -0.17*** (0.05) -1.12*** (0.19) -0.19*** (0.07) 0.51*** (0.13) 0.01 (0.03) -1.10*** (0.12) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 34,289 0.19 31,863 0.74 29,627 0.75 The eﬀect of risk is also similar to that found in Table 2.2. The estimates of the coeﬃcient on risk are statistically signiﬁcant and positive in all cases. The estimate in Column (2) indicates that a 1% increase in risk contributes to a 0.05% rise in the cash-inventory ratio in the short run and around a 0.19% increase in the long run, which is of economic signiﬁcance. 2.3.5 Trends in Cash and Inventory Holdings Figure 2.1 shows that in the past three decades, corporate sector has been hoarding cash and at the same time decumulating its inventory stock. This paper presents insight into the joint dynamics of cash and inventory ratios. 76 2.3. Estimation and Results The model suggests that riskier and increasingly competitive product market environments lead ﬁrms to reallocate resources from inventory stock to cash reserve. A decline in market power limits ﬁrms’ ability to raise product prices over marginal costs without losing customers, and in turn reduces ﬁrm proﬁtability and the value of inventory. Firms therefore shift resources from inventory to cash in order to maintain ﬁnancial ﬂexibility. In addition, an increase in risk raises the probability of occurrence of extreme events. Costly external ﬁnance leads ﬁrms to concern more about low-productivity states (negative outcomes) and ﬁrms respond to a riskier environment by saving more out of current resources. Inventory is highly valued in fairly low productivity states of the world, while cash, delivering the same rate of return in all states, is valued whenever the probability of failing to meet liquidity needs is positive. Consequently, ﬁrms transfer resources from future high productivity states (by drawing down inventory) to present, store the proceeds and use them to fund all states in the future if necessary. Overall, an increase in risk leads ﬁrms to put more weight on cash than on inventory when managing liquidity. It is empirically well documented that product market competition and ﬁrm-level risk have increased over time. [31] use turnover of leaders within industry to proxy market competition and identify industry leaders based on a ﬁrm’s operating income and market value. They ﬁnd that there is a steady increase in turnover among leaders over time. Similarly, [56] analyze the time trend of product market competitiveness. They employ three variables, industry turnover, return on assets, and the market share of foreign competitors, to measure market competition. They also ﬁnd that product market becomes increasingly competitive and domestic producers lose market shares to their foreign rivals. The increase in ﬁrm-level volatility for publicly traded ﬁrms is ﬁrst comprehensively documented in [22]. They use stock return data in CRSP to measure ﬁrm-level volatility and ﬁnd that there is a signiﬁcant positive time trend in idiosyncratic risk. [30] use the volatility of the growth rate of sales to measure ﬁrm-level risk and ﬁnd the same upward trend after controlling for sample composition eﬀects. Therefore, my model provides possible explanations for the patterns shown in Figure 2.1: increases in risk and market competitiveness can contribute to the simultaneous changes in corporate cash and inventory holdings. Similar conclusions on cash holdings are derived in [71] and [82]. They use a real option model to show that both market competition and risk drive up the option value of exit by aﬀecting the closure threshold and ﬁrms hold cash to prevent ineﬃcient exit. This paper distinguishes itself from those two studies by proposing an additional mechanism relating product market 77 2.4. Conclusion environment, risk and cash holdings: the reason for which ﬁrms prefer raising cash rather than other alternative sources of internal ﬁnance for liquidity management, when the total internal liquidity demand is higher due to a tougher business environment. 2.4 Conclusion While previous studies have focused on cash holdings, I in this paper consider an additional source of internal liquidity which is often neglected: inventory. I analyze the tradeoﬀ ﬁrms face when choosing between holding cash and inventory to manage liquidity needs. Using a model that assumes uncertain productivity shocks and imperfect capital markets, I show that a ﬁrm’s market power and exposure to risk aﬀect the choice between cash and inventory. Firms facing relatively inelastic demand and thus having greater pricing power ﬁnd themselves more likely to beneﬁt from carrying inventory than saving cash. Firms operating in a riskier environment have a greater chance of experiencing a cash ﬂow shortfall and thus a liquidity shortage. They respond by raising internal liquidity, mainly in the form of cash. I verify the model implications by showing that ﬁrms with higher market power accumulate more inventory but less cash and riskier ﬁrms act in the opposite way, after controlling for other determinants. These implications in turn suggest that the dramatic change in the composition of ﬁrms’ current assets observed in the data can be explained by increased market competition and increased ﬁrm-level risk. This paper also adds to the existing inventory literature by modeling inventory as a source of funds and presents evidence on it. The empirical results show that ﬁnancially-constrained ﬁrms hold more inventory than their counterparts, which supports the liquidity motive for inventory holdings. An interesting extension to the work in subsequent research is to consider an inﬁnite-horizon model. The current framework limits the role of market power in aﬀecting cash and inventory. In the three-period setup, to simplify the model I assume that ﬁrms have to sell all products in the second period, which implies that ﬁrms are not allowed to manage cash ﬂows by (dis)investing in inventory. In other words, an inﬁnite-horizon model allowing for price setting and therefore inventory investment in each period can endogenize cash ﬂows (one of the most important internal liquidity) and, to some degree, absorb the adverse eﬀects of uncertainty. Besides, an inﬁnitehorizon model can be used to quantitatively evaluate the impacts of market power and uncertainty on cash and inventory. 78 Chapter 3 Limited Participation in International Business Cycle Models: A Formal Evaluation 3.1 Introduction A number of existing studies have shown that access to international borrowing and lending is important for international business cycles. We verify this result formally by means of a novel statistical procedure applied to several versions of a standard two-country two-good model. We show that a model with no cross-border asset trades (ﬁnancial autarky) is the speciﬁcation that outperforms other models in matching the data and that the diﬀerences in performance are statistically signiﬁcant. We then propose a competing model that allows for within country household heterogeneity in participation in asset markets. Using our test we show that this amended model does signiﬁcantly better in matching the data than a representative agent benchmark. Furthermore, when limited participation is introduced, complete markets speciﬁcation performs as well as ﬁnancial autarky in ﬁtting the data. How does access to diﬀerent ﬁnancial assets aﬀect the functioning of the economy? In a seminal work, [7, 8], BKK hereafter, document the key business cycle regularities in industrial countries related to volatilities of consumption, output, investment and their cross-country co-movements, and develop an international business cycles model with complete asset markets in an attempt to rationalize the data facts. They show that only some of the regularities can be explained by the model. The BKK model fails in three key dimensions. First, while cross-country consumption correlations tend to be similar to cross-country output correlations in the data, the model predicts consumption correlations far exceeding those for outputs. This is 79 3.1. Introduction the so-called “quantity” puzzle [7]. Second, investment and employment are positively correlated across countries while the model predicts a negative correlation. This data-model disconnect is usually referred to as the “international comovement” puzzle [13]. Third, the model generates signiﬁcantly less volatility in the terms of trade and the real exchange rate relative to the data. The model also predicts a positive correlation between real exchange rate and the ratio of domestic to foreign consumption, again contrary to the data.40 To account for the disconnect between the model and data, [14], [62], [6], [33] study economies in which the only asset traded internationally is a non-contingent bond. They show that these economies admit diﬀerent allocations from those arising under complete asset markets only if productivity shocks are very persistent and do not spill over across countries. [47] develop this argument further by considering an economy in which no international assets are traded. They call it ﬁnancial autarky. They ﬁnd that the equilibrium dynamics under ﬁnancial autarky are similar to those in the data. Their conclusion, however, is based primarily on an “eyeball” comparison of various moments predicted by the model with those of competing models and with the data. In fact, such informal moment comparison is standard practice in the literature. There are several important shortcomings of the “eyeball” approach. First, it does not inform whether diﬀerences in the model performance are statistically signiﬁcant. Namely, an “eyeball” approach cannot credibly distinguish between the systematic diﬀerences in model performance (in the sense that the model uncovers important relationships between the variables at the population level) and, therefore, is likely to be found in other data sets; and diﬀerences arising due to random variations in the data. Second, often model comparison is hindered by the fact that one model performs better in matching some moments, but competing models perform better in matching other moments. Without a metric that aggregates across various moments of interest, informal model comparison remains inconclusive. In this paper we propose a testing procedure that allows the researcher to assess the statistical signiﬁcance of results when comparing DSGE models to the data. The procedure builds upon [50] and is a version of Vuongtype tests for misspeciﬁed models [86] adopted for the DSGE framework. Suppose that the researcher is interested in evaluating whether a newly proposed economic structure is important for explaining some chosen data 40 A detailed recent discussion of various puzzles in the international business cycles models can be found in [69]. 80 3.1. Introduction patterns. For that purpose, a Vuong-type procedure compares the empirical ﬁt of the new model with that of a leading benchmark model, and tests the null hypothesis that they are equal. If the null hypothesis is accepted, then the researcher must conclude that there is no suﬃcient empirical evidence in favor of the new model. On the other hand, if the null hypothesis is rejected in favor of the new model, the researcher can credibly argue that the new economic structure provides a superior explanation to the data patterns over that of the benchmark model. The procedure does not require that either of the competing models be correctly speciﬁed, and therefore the conclusions are robust to misspeciﬁcation.41 The procedure consists of several steps. In the ﬁrst step we determine the values of the deep structural parameters in each of the competing models. This can be done either informally by setting the parameters to their values typically used in the literature or through formal estimation where the values for the parameters are chosen to match certain characteristics of the data. In the second step, we compute the weighted Euclidean distance between the vectors of model-predicted characteristics and their estimates from the data. We then obtain the test statistic as the diﬀerence between the estimated measures of ﬁt of the two competing models as well as its standard error. The standard error has to take into account how the values for the structural parameters were obtained in the ﬁrst step. Lastly, we reject the null hypothesis of equal ﬁts if the studentized diﬀerence in ﬁts exceeds a standard normal critical value. We apply the methodology to a popular class of models in the international business cycles literature in order to determine which of the asset market structures used extensively in that literature has the strongest explanatory power for observed empirical regularities. More precisely, we compare three key models: ﬁnancial autarky, single risk-free bond economy, and an economy with complete asset markets. Our comparison is based on a set of standard data characteristics: variances of key macroeconomic aggregates, such as consumption, investment, labor input, etc.; correlations of these aggregates with output, and their cross-country co-movements. Our procedure recognizes that diﬀerent data characteristics have diﬀerent scales (i.e. variances can take any non-negative values, while correlations are restricted by [−1, 1] interval). This makes model comparison based on the equally-weighted aggregation of characteristics problematic. Instead, we 41 We deﬁne a structural model to be misspeciﬁed if it cannot predict the population values of the chosen data characteristics for any combination of the deep structural parameters. See [50] for details. 81 3.1. Introduction propose a data-dependent weighting scheme which allows us to normalize various characteristics by their data counterparts and aggregate them easily. We show that based on both sets of moments (variances and correlations) our test indeed picks ﬁnancial autarky as the winning speciﬁcation – consistent with the informal conclusion in [47]. We then propose a competing model that allows for agent heterogeneity. We focus on a simple dimension of heterogeneity – asset market participation. In our competing model there are two groups of agents in each country: those with access to international and domestic ﬁnancial markets (participants) and non-participants. We characterize the business cycle properties of the model with limited asset market participation (LAMP) and then apply our test to evaluate the ability of this amended model relative to models with a representative agent in matching the properties of the data. As before, we consider three speciﬁcations for international asset markets: ﬁnancial autarky, single risk-free bond economy, and an economy with complete asset markets, except that in the economy with LAMP these ﬁnancial regimes apply to participants only. We show that in the setup with LAMP, ﬁnancial autarky remains a preferred model if the comparison is based on volatilities of key macroeconomic aggregates. However, if the comparison is performed based on co-movements with output and cross-country correlations, then a complete markets economy is chosen as the winner. This is mainly due to the fact that LAMP improves the performance of the model for cross-country correlations: it signiﬁcantly raises the cross-country correlation in hours of work and investment. Thus, it improves on the “international comovement” puzzle. Adding LAMP also raises the cross-country correlation of output, and lowers the corresponding correlation for consumption, thus bringing the two closer together. Therefore, our models with LAMP also improve on the “quantity” puzzle. Lastly, based on the overall performance (variances and correlations), we ﬁnd that a complete markets model with LAMP performs no worse than ﬁnancial autarky and outperforms all other models. In a majority of cases the improvements are statistically signiﬁcant. Adding LAMP alters the behavior of a representative agent benchmark in three key ways. First, non-participants are hand-to-mouth consumers whose consumption closely tracks their income. Therefore, LAMP raises the sensitivity of aggregate consumption to income shocks, in line with the data. Second, non-participants’ only income is from their labor earnings, making their hours inelastic. Therefore, LAMP reduces the sensitivity of aggregate labor supply to productivity shocks. Third, with consumption responding more and labor supply responding less to shocks, investment becomes less sensitive to productivity shocks. These modiﬁcations improve the model’s 82 3.1. Introduction performance in some dimensions (i.e. cross-country correlations, volatility of consumption and its comovement with output), but worsen its performance in other dimensions (i.e. volatilities of output, investment, hours, etc.). Thus, ex-ante, the overall contribution of LAMP is ambiguous. This result highlights the need for a procedure that allows to compare models formally. We contrast the overall performance of LAMP against the model with a representative agent by aggregating the ﬁts across all three ﬁnancial regimes. We ﬁnd that LAMP class of models signiﬁcantly outperforms the original BKK model class. We verify the robustness of our results to the presence of investment speciﬁc technology shocks and with respect to the elasticity of substitution between labor inputs of participants and non-participants. Overall, our results indicate that adding LAMP to a standard international business cycles model signiﬁcantly improves its ability to match business cycle facts and can overturn the existing result that ﬁnancial autarky provides a better ﬁt to the data. To the best of our knowledge, ours is the ﬁrst paper to perform a statistical model evaluation and comparison based on the agents’ heterogeneity over a large set of international business cycle statistics. We believe that our model with LAMP provides a simple, but empirically important extension of the standard business cycle framework. The fact that only a small fraction of households participate in the stock market has been documented by [70], who showed that only 24% of US households owned equities in 1984; in 2007 this fraction was 51.1% based on the Survey of Consumer Finance.42 Limited asset market participation has received attention in the theoretical asset pricing literature (see 76; 85 and others). [27] provide its quantitative evaluation and show that a model with LAMP (and incomplete markets) can account for high volatility of equity risk premium in the data. [26] investigate the implications of LAMP and heterogeneous trading technologies for asset prices and wealth distribution and show that such a model matches well the high volatility of returns and the low volatility of the risk-free rate. Implications of LAMP for monetary policy have been studied by [43], [24], [2], [16]. They show that LAMP improves model performance for nominal aggregates. [84] studies the importance of LAMP and borrowing constraints for cross-country consumption correlations and welfare. 42 The share of US households who own equities, while increasing dramatically since 1984, has remained relatively stable at around 50% in the past 15 years, based on the Survey of Consumer Finances. Thus, based on the Survey, the share was 31.6% in 1989, in 1992 – 36.7%, in 1995 – 40.4%, in 1998 – 48.9%, in 2001 – 52.2%, in 2004 – 50.2%, and in 2007 – 51.1%. 83 3.2. Model Economies Relative to the above papers, the key contribution of our work is to statistically examine the consequences of LAMP for a large set of business cycle moments as well as formally evaluate its performance relative to alternative models popular in the literature. Our results build the case for LAMP further by showing its importance for international business cycles. The paper also makes methodological contributions by extending [50] in two respects. First, we extend the procedure to account for simulation uncertainty. The complexity of DSGE models often makes exact calculations of model predicted moments very cumbersome. In such cases, it is convenient to resort to simulations as we do in this paper. We show how the standard error of the model comparison test statistic can be adjusted to account for simulation uncertainty, which can be used to ensure that no power is lost due to simulations. Second, we propose a class-based test that allows one to compare the overall performances of classes of model with several models in each class. Such an extension is useful when, as in our case, each model has several structurally diﬀerent versions. Our paper is related to the growing literature on the comparison of structural models by means of statistical methods. To name a few, comparison of misspeciﬁed DSGE models has been studied from the Bayesian perspective by [81]. The method we employ in this paper can be viewed as a frequentist counterpart of the [81] procedure.43 More recently, [59] proposed a Vuongtype test for comparison of misspeciﬁed asset pricing models in terms of the Hansen-Jagannathan distance. The remainder of the paper is organized as follows. Section 3.2 presents our model economies. We discuss calibration and model solution in Section 3.3. Section 3.4 describes our testing methodology. Section 3.5 presents our results, and Section 3.6 concludes. 3.2 Model Economies To study the role of asset market structure in capturing the properties of international business cycles, we consider a sequence of three economies: an economy in which there are no markets for international asset trades (we refer to it as ﬁnancial autarky, fa); an economy in which a single noncontingent bond is traded – bond economy, be; and an economy with complete markets, cm. The structure of these economies follows closely that 43 While in [81] a structural model that achieves the lowest average posterior loss is selected, we follow the approach of [86] and test the null hypothesis that two competing models have equal losses. 84 3.2. Model Economies proposed by [7, 8] and studied in [47]. For completeness, we present it here as well. To study the role of investors heterogeneity we extend the three versions of the model to incorporate limited asset market participation. Aside from asset market structure and investors’ heterogeneity, all our economies have common structure. We describe it next. We consider the world consisting of two symmetric countries, h and f, each specializing in the production of its intermediate good. Each country is populated by a continuum of ﬁrms and households. 3.2.1 Firms Firms are perfectly competitive and reside in two sectors: intermediategoods sector and ﬁnal-goods sector. Firms in the intermediate goods sector (i-ﬁrms) hire domestically-located capital, k j , and labor, nj , j = {h,f}, to produce intermediate goods. The i-ﬁrms in country h specialize in the production of good a, while i-ﬁrms in country f specialize in the production of good b. Period t production by a representative i-ﬁrm in country j is F (ztj , ktj , njt ) ( )θ ( )1−θ ktj njt , =e ztj (3.1) with θ > 0, and ztj being the exogenous state of productivity in country j. Let wtj and rtj denote the real wage and rental rate on capital in country j in period t, measured in terms of the domestic intermediate good. The problem facing i-ﬁrms in country j then becomes max F (ztj , ktj , njt ) − wtj njt − rtj ktj , subject to njt > 0, ktj > 0, and equation (3.1). The intermediate goods produced by h and f i-ﬁrms can be freely traded in the international goods markets and can be costlessly transported between countries. Under these conditions, the law of one price must prevail to eliminate arbitrage opportunities. Households, who are the owners of the i-ﬁrms, sell their holdings of intermediate goods to domestic ﬁnal goods producing ﬁrms (f -ﬁrms), and use the proceeds for consumption, cjt and investment, xjt . Investment adds to the stock of physical capital available for production next period according to j kt+1 = (1 − δ)ktj + xjt , where δ is the depreciation rate. 85 3.2. Model Economies The f -ﬁrms are also perfectly competitive and produce ﬁnal goods from the h and f intermediate goods using constant returns to scale (CRS) technology: σ [ ( ) σ−1 ] σ−1 ( ) ( j ) σ−1 σ σ j j j j j G(at , bt ) = ω at + 1−ω bt , (3.2) where ω j is the weight that f -ﬁrms from country j assigns to the intermediate goods produced in country h. When ω j > 0.5 there is home bias in the production of ﬁnal goods in country j. The elasticity of substitution j j between h and f-produced intermediate goods is σ > 0. Let qa,t and qb,t denote the prices of intermediate goods a and b in country j in units of the ﬁnal good produced in country j. Then, the problem facing f -ﬁrms in country j is j j j max G(ajt , bjt ) − qa,t ajt − qb,t bt , subject to njt > 0, ktj > 0, and equation (3.2). Productivity in intermediate good sectors is governed by an exogenous process. In particular, we assume that the vector zt ≡ [zth , ztf ]′ follows an AR(1) process: zt = αzt−1 + et , (3.3) where et is a (2 × 1) vector of independently normally distributed, mean zero shocks with covariance Ωe . 3.2.2 Households Each country is also populated by a continuum of households, whose preferences are deﬁned over consumption and leisure. In particular, the preferences of households in country j are represented by E0 ∞ ∑ β t U (cjt , 1 − njt ), (3.4) t=0 where 0 < β < 1 is the discount factor, and U (.) is a concave sub-utility function. Period utility of the household in country j is given by [( ) function )1−µ ]γ µ( j j j j U (ct , 1 − nt ) = γ1 ct 1 − nt . Households choose consumption, cjt , and hours of work, njt ∈ [0, 1], to maximize their lifetime expected utility subject to a sequence of budget constraints, which depend on the ﬁnancial structure of the model economy. We consider three such structures. Under ﬁnancial autarky (fa), households can not trade any international ﬁnancial 86 3.2. Model Economies assets. Under bond economy (be), households can hold a single non-state contingent internationally traded bond. The third case we consider is that of complete markets (cm). Here households have access to a complete set of Arrow securities. We now describe the budget constraints facing households under each of these diﬀerent ﬁnancial structures. 3.2.2.1 Financial Autarky, FA In the ﬁnancial autarky, households do not have access to international ﬁnancial assets. As a result, households consume and invest out of their factor income. The period−t budget constraint of households in country j is ) ( j wtj njt + rtj ktj . cjt + xjt = qa,t Notice that fa rules out the possibility of international borrowing or lending, so neither country can have positive or negative trade balance. 3.2.2.2 Bond Economy, BE In the bond economy households only trade a single non-state-contingent international bond. We assume that bonds are denominated in the units of intermediate good a. Let Btj denote bond holdings of country j households and Qt be the price of the bonds. Then the period−t budget constraint of households in country j is ( ) j j j j cjt + xjt + qa,t Qt Btj = qa,t wtj njt + rtj ktj + qa,t Bt−1 . 3.2.2.3 Complete Markets, CM Following [47] we assume that households complete the markets by trading in a complete set of Arrow securities denominated in units of intermediate good a. Thus the households’ budget constraint can be written as ( ) ∑ j j j j cjt +xjt +qa,t Qt (st , st+1 )Btj (st , st+1 ) = qa,t wtj njt + rtj ktj +qa,t Bt−1 (st−1 , st ), st+1 where st = (s0 , s1 , s2 , ..., st ) denotes the entire state history of the economy till date t. 87 3.2. Model Economies 3.2.2.4 Equilibrium j j An equilibrium in this economy consists of a set of goods’ prices {qa,t , qb,t }, t and asset prices (i.e. {Qt } under be or {Qt (s , st+1 )} under cm) such that all markets clear when households optimally make their consumption, investment, and asset allocation decisions, taking goods and asset prices as given. Market clearing in the intermediate goods markets requires aht + aft = F (zth , kth , nht ), bht + bft = F (ztf , ktf , nft ). Market clearing in the ﬁnal goods markets requires cjt + xjt = G(ajt , bjt ), j = {h, f}. The market clearing conditions in ﬁnancial markets vary according to the ﬁnancial structure of the economy. Under be, the bond market clearing condition requires 0 = Bth + Btf . Under cm, a similar condition applies for every st+1 : 0 = Bth (st , st+1 ) + Btf (st , st+1 ). 3.2.3 Limited Asset Market Participation Next, we introduce LAMP in our model economy. This feature is used to capture the empirical observation that a large fraction of population does not hold any ﬁnancial assets. Thus, we assume that each country is populated by two types of households: non-participants and participants. Nonparticipants do not own any capital, do not have access to international markets, and only choose how much time to work and how much to consume. Participants hold all of the capital stock in the economy and can borrow and lend at the international markets (if the model speciﬁcation allows it). They also supply labor services to the intermediate goods producing ﬁrms and make all investment decisions. We assume that there is a fraction λ of such households in each country. The problem facing non-participants (n) is max E0 ∞ ∑ β t U (cjn,t , 1 − njn,t ), t=0 88 3.2. Model Economies subject to j cjn,t = qa,t wtj njn,t , where subscript n is used to denote the variables pertinent to non-participants. Note that the non-participants’ problem remains the same independent of the assumed asset market structure. The problem facing participants (p) is the same as in the economy with a representative agent: max E0 ∞ ∑ β t U (cjp,t , 1 − njp,t ), t=0 subject to a budget constraint. Here subscript p is used to denote the variables speciﬁc to asset market participants. For participants the exact form of the budget constraint varies with the ﬁnancial structure of the economy. For instance, the budget constraint of participants in the ﬁnancial autarky is ( ) j cjp,t + xjt = qa,t wtj njp,t + rtj ktj In the bond economy the budget constraint becomes ( ) j j j j cjp,t + xjt + qa,t Qt Btj = qa,t wtj njp,t + rtj ktj + qa,t Bt−1 , while under complete markets, it is ( ) ∑ j j j j cjp,t +xjt +qa,t Qt (st , st+1 )Btj (st , st+1 ) = qa,t wtj njp,t + rtj ktj +qa,t Bt−1 (st−1 , st ). st+1 Note that in this case, the asset markets are complete internationally for participants only. The optimization problems solved by i-ﬁrms and f -ﬁrms remain unchanged. Aggregate labor input in the economy consists of labor inputs of participants and non-participants and is deﬁned as: [ njt = λ ( njp,t ) υ−1 υ ( ) υ−1 υ j + (1 − λ) nn,t ] υ υ−1 , where υ is the elasticity of substitution between the two types of labor. The market clearing conditions in the goods markets remain the same, while the market clearing conditions in the asset markets apply to participants only. 89 3.2. Model Economies 3.2.4 Investment-specific Technology (IST) Shocks Several recent papers have emphasized the role played by investment-speciﬁc technology (IST) shocks in the international business cycles (IBC). In a framework similar to ours, [77] shows that IST shocks can help account for a number of puzzles in the business cycles literature. He emphasizes the Backus-Smith puzzle – the fact that consumption and real exchange rate tend to be negatively correlated in the data, while a standard IBC framework predicts the opposite; and the “price” puzzle – the fact that models generate far lower volatility of international relative prices relative to the data. At the same time, [69] show that an IBC model with IST shocks estimated from the data fails to reproduce the moments emphasized in [77]. Our interest in IST shocks is motivated by their potentially important interactions with LAMP. When only a segment of population has access to capital and asset markets IST shocks will have diﬀerential eﬀects on the participants and nonparticipants, leading to important distributional eﬀects between them. We investigate the role of IST shocks by incorporating them in our models as in [42] and [77], but using the properties of these shocks as estimated in [69]. In what follows we highlight the new model features introduced by IST shocks. The problem facing non-participants does not change when IST shocks are introduced. Objective functions of participants and their budget constraints also remain unchanged. In the presence of IST shocks, capital accumulation equation becomes j j kt+1 = (1 − δ)ktj + evt xjt , j where ev is the IST shock in country j. As shown in [42], in a competitive j equilibrium, e−v is interpreted as the relative price of capital goods in terms of consumption goods. We assume that IST shocks, vt ≡ [vth , vtf ]′ follow an AR(1) process: vt = αv vt−1 + ζt , (3.5) where ζt is a (2 × 1) vector of independently normally distributed, mean zero shocks with covariance Ωζ . All other model equations remain unchanged. 3.2.5 Definitions There are several variables of interest that we deﬁne here. Gross domestic product in country j expressed in terms of ﬁnal consumption goods is given j h af − q h bh . Imports by ytj = qa,t F (ztj , ktj , njt ). Net exports are nxht = qa,t t b,t t ratio for home country is deﬁned following [47], as the ratio of imports to 90 3.3. Calibration and Model Solution domestically consumed intermediate goods, both measured at the steady state prices which are symmetric under the benchmark calibration, giving irth = bht /aht . Terms of trade in h country are deﬁned as the price of imports h /q h , while the real exchange rate divided by the price of exports, pht = qb,t a,t is deﬁned as the relative price of foreign consumption goods to domestic h /q f . consumption goods, giving rerth = qa,t a,t 3.3 Calibration and Model Solution In calibrating the model we assign some parameters their values commonly used in the literature, while we estimate other parameters from the data. Such an approach has become standard in the literature. In our application it also allows us to illustrate our testing procedure in the most general case when some parameters are ﬁxed while other parameters are estimated.44 In the calibration we consider the world economy as consisting of two countries: country 1 matching the properties of the US economy in quarterly data, and country 2 as the rest of the world. Most of the parameter values are borrowed from [47]. We summarize them in Table 3.1. We set discount factor to 0.99, which implies annual real interest rate of 4 percent. Risk aversion coeﬃcient is set at 2. As in [47], we ﬁx consumption share parameter at µ = 0.34. We assume that capital income share, θ is 0.34; and depreciation rate δ of 2.5 percent. Parameter ω, which controls the consumption home bias in household’s preferences is set to match the observed import share in the U.S. equal to 15 percent of GDP. We set the elasticity of substitution between domestic and imported intermediate goods at 0.9, which is the value estimated in [47]. This is above the value of this parameter used in [77] and [69], but more along the lines of the values used in the IBC literature.45 In the model with LAMP a new parameter, λ, is introduced. It captures the share of nonparticipants, which we calibrate to match the share of US households who did not hold any equity as reported in the 2007 Survey of Consumer Finance equal to 51.1%. Therefore, we set λ = 0.5. The only remaining parameter is υ which equals the elasticity of substitution between labor input of participants and non-participants in the model. For simplicity and given the lack of estimates of this parameter in the literature, we assume 44 As we discuss in Section 3.4.1, when some model parameters are estimated, we must take into account the uncertainty due to this estimation when computing standard errors of our test statistics. 45 For instance, [7, 8] use a value of 1.5. [63] uses traded elasticity values as low as 0.6; [23] and [35] use 1.5. 91 3.3. Calibration and Model Solution that the two types of labor are perfectly substitutable. In what follows we check the robustness of our results with respect to this parameter. Table 3.1: Benchmark Parameter Values without Estimation Step Panel A: Preferences discount factor risk-aversion consumption share Panel B: Technology capital income share depreciation rate import share elasticity of subst, b/n goods a and b share of p households Panel C: IST shocks β 1−γ µ 0.99 2 0.34 θ δ is (ω) σ λ 0.36 0.025 0.15 0.9 0.5 [ ρ11 ρ21 transition matrix αI = std. dev. of innovations corr. of innovations σeI1 = σeI2 σeI1 e2 ρ12 ρ22 ] [ 0.975 0.024 0.024 0.975 0.0066 0.1955 ] TFP shocks are assumed to be persistent, but temporary. We estimate the process for TFP shocks as in [47]. Namely, we compute productivity sequences for the US and the rest of the world during 1973:1-2007:4 period, where the rest of the world is identiﬁed with the aggregate of 21 major trade partners for the U.S..46 In our estimation, we impose the symmetry restrictions ρ11 = ρ22 and ρ12 = ρ21 . Our estimation results for productivity process are presented in Table 3.2 and they are very similar to the estimates in [47]. Namely, our estimates of productivity persistence ρ11 and spill-over ρ12 are almost the same, while the standard deviation of productivity innovations σe1 and the correlation between domestic and foreign productivity innovations σe1 e2 are somewhat smaller than their values.47 In calibrating IST shocks, we follow the ﬁndings of [69] who show that IST processes for the U.S. and the rest of the world are very persistent and exhibit no spill-overs across countries. Importantly, [69] show that the variance of these shocks is of the same magnitude as the variance of TFP shocks. Motivated by these results, and to facilitate the comparison of the models with and without IST shocks, we assume that IST shocks are fully symmetric to TFP shocks, with no spillovers across the two types of shocks. Each model is solved by linearizing the sequence of equilibrium conditions and solving the resulting system of linear diﬀerence equations. We 46 47 Details on sample construction and data sources are provided in the Appendix C.1. When simulating the models we use σe1 = σe2 = 0.0066. 92 3.4. Econometric Methodology Table 3.2: Estimated Productivity Process Table 3.2 reports coeﬃcient estimates and their standard errors of the estimated [ produc] z1,t tivity process. Following [47], we estimate productivity shock process using: = z2,t [ ][ ] [ ] ρ11 ρ21 z1,t−1 ε1,t + with the symmetry restriction imposed, ρ11 = ρ22 and ρ12 ρ22 z2,t−1 ε2,t ρ12 = ρ21 . [ productivity transition matrix std. dev. of productivity innovations corr. of productivity innovations α= ρ11 ρ12 ρ21 ρ22 ] σe1 σe2 σe1 e2 0.975 0.024 (0.009) (0.009) 0.024 0.975 (0.009) (0.009) 0.0066 0.0039 0.1955 derive the second moments of model’s variables by simulating the model over 100 periods. The statistics based on which the model comparison is conducted are derived from 10000 simulations. All series, except net exports, are logged and Hodrick-Prescott (HP) ﬁltered with a smoothing parameter of 1600. 3.4 Econometric Methodology For a formal statistical comparison of the considered models, we rely on a Vuong-type [86] test for potentially misspeciﬁed calibrated models proposed in [50, 51]. Relative to this work we develop two key extensions. First, we adjust the procedure to account for simulation uncertainty. This becomes important when the model moments can not be computed exactly and instead simulations must be used.48 Second, we introduce a class-based test that allows us to compare classes of models with several models in each class. This becomes important when one is interested in evaluating the model’s performance with diﬀerent features, for diﬀerent parameter values, or with diﬀerent shocks. In this case, one needs a way to aggregate model ﬁts across the diﬀerent scenarios, which is what our proposed class-based test does. 48 Using simulations to obtain model implied moments is a common practice in the business cycles literature. 93 3.4. Econometric Methodology 3.4.1 Pairwise Comparison We begin by assuming that data can be summarized using two mutually exclusive vectors of characteristics denoted by h1 and h2 , where the ﬁrst vector is used for estimation of unknown structural parameters, while the second vector is used to compare structural models. This reﬂects a standard practice in applied macroeconomics, when parameters are calibrated to one group of data characteristics, while models are evaluated on another. We assume that h1 and h2 can be estimated from data without employing a structural model. For example, in our case, h1 consists of the estimated productivity shocks, while h2 consists of volatilities and correlations between the variables of interest as described in Tables 3.3-3.5 below. Suppose that there are two structural models denoted f (θ) and g(β), where θ and β are the corresponding structural parameters describing consumer’s preferences, technology, etc. Here, f (θ) and g(β) denote the value of h2 predicted by models f and g, respectively. Naturally, vectors h2 , f (θ) and g(β) must be of the same dimension; we assume that they are m-vectors. We allow for the competing models to be misspeciﬁed, i.e. it is possible that for all permitted values of θ and β, h2 ̸= f (θ) and h2 ̸= g(β). The models are allowed to share some of the parameters. Note, however, that θ and β contain only the parameters that must be estimated from data. We allow that some of the parameters may be assigned ﬁxed values, for example, values that are commonly used in the literature. Such parameters are excluded from θ and β and absorbed into f and g.49 We are interested in testing a hypothesis that models f and g have equivalent ﬁt to the data as described by h2 . For an m × m symmetric and positive deﬁnite weight matrix Wh2 , the null hypothesis of the models’ equivalence is H0 : (h2 − g(β))′ Wh2 (h2 − g(β)) − (h2 − f (θ))′ Wh2 (h2 − f (θ)) = 0. The notation indicates that the weight matrix Wh2 can depend on h2 . A simple choice for a weight matrix is to use the identity matrix. In that case, the weight matrix is independent of h2 , and the models are compared in terms of their squared prediction errors. Another example for Wh2 is a diagonal matrix with the reciprocals of the elements of h2 on the main diagonal. With such a choice of the weight matrix, the models are compared in terms of the squares of their percentage prediction errors. In our application, we use a combination of the two. That is to evaluate the models, for some 49 In our application θ and β are the same and describe the productivity process. 94 3.4. Econometric Methodology parameters, such as correlations, we use prediction errors, while for others, such as volatilities, we use percentage prediction errors. The alternative hypotheses are Hf : (h2 − g(β))′ Wh2 (h2 − g(β)) − (h2 − f (θ))′ Wh2 (h2 − f (θ)) > 0, Hg : (h2 − g(β))′ Wh2 (h2 − g(β)) − (h2 − f (θ))′ Wh2 (h2 − f (θ)) < 0, where f has a better ﬁt according to Hf , and g has a better ﬁt according to Hg . ˆ 1 and h ˆ 2 denote the estimators of h1 and h2 , respectively. We asLet h ˆ ˆ 2 do not require the knowledge of the true structural sume that h1 and h model, are consistent and asymptotically normal as described in the following assumption: ( ) ( ( )) ˆ − h1 √ h Λ11 Λ12 n ˆ1 →d N 0, , (3.6) Λ′12 Λ22 h2 − h2 where n denotes the sample size used in estimation of h1 and h2 , Λ11 and ˆ 1 and h ˆ 2 respecΛ22 denote the asymptotic variance-covariance matrices of h ˆ ˆ 2 . Let tively, and Λ12 denotes the asymptotic covariance between h1 and h ˆ 11 , Λ ˆ 22 and Λ ˆ 12 denote consistent estimators of the corresponding elements Λ in the above asymptotic variance-covariance matrix. In a typical time-series application, Λ11 , Λ22 and Λ12 are long-run variances and covariances and, therefore, require HAC-type estimators, see [72] and [3]. Let θˆ and βˆ denote the estimators of θ and β respectively. We assume that the estimators are asymptotically linear in h1 : ) ) √ ( √ ( ˆ 1 − h1 + op (1), n θˆ − θ = A n h (3.7) ) ) √ ( √ ( ˆ 1 − h1 + op (1), n βˆ − β = B n h (3.8) where matrices A and B may depend on the elements of h1 . This speciﬁcation is satisﬁed by most estimators used in practice. Appendix C.2 contains the derivations of equation (3.7) for our estimators.50 We assume that A ˆ to denote their and B can be consistently estimated, and use Aˆ and B estimators. When functions f (θ) and g(β) are too complicated for analytical or even exact numerical calculations, we assume that they can be estimated by simulations. For example, as in our case, one can draw random shocks and solve the models as described in Section 3.2 using θˆ for model f and βˆ for 50 In our application, because β and θ are the same, we do not use equation (3.8). 95 3.4. Econometric Methodology model g, and obtain a set of random equilibrium values for the variables of interest. By repeating this process R times, one obtains a sample of R observations for the variables of interest, which can be used to estimate f ˆ and gˆ(β) ˆ denote such and g by averaging across the simulations. Let fˆ(θ) estimators. We assume that, at the true values θ and β, estimators fˆ(θ) and gˆ(β) ˆ 1 and h ˆ 2 , and satisfy the following assumption: are independent of h ( ) ( ( )) √ Λf f Λf g fˆ(θ) − f (θ) R →d N 0, . (3.9) Λ′f g Λgg gˆ(β) − g(β) ˆff , Λ ˆ gg and Λ ˆ f g to denote consistent estimators of the asymptotic We use Λ variances and covariance in (3.9). Our test is based on the diﬀerence between the estimated ﬁts of the two models: ( ) ( ) ( ) ( ) ˆ 2 − gˆ(β) ˆ ′ Wˆ h ˆ 2 − gˆ(β) ˆ − h ˆ 2 − fˆ(θ) ˆ ′ Wˆ h ˆ 2 − fˆ(θ) ˆ . S= h h2 h2 Under the assumptions in (3.6)-(3.9), S is asymptotically normal, and its √ standard error can be computed as σ ˆ / n, where51 σ ˆ 2 = 4ˆ σ12 + 4ˆ σ22 , (3.10) ′ ( )′ ( )′ ( ( ) ) ˆ ˆ ˆ ( ) ( ) ) ( ˆ ˆ ˆ ˆ ˆ ˆ′ ∂ f θ ˆ ′ ∂ˆg β ˆ 11 Λ ˆ 12 Λ A ∂θ Whˆ 2 h2 − f θ − B ∂β Whˆ 2 h2 − gˆ β 2 σ ˆ1 = ( ( ( ) ˆ 2 )′ ( ) ( )) ( )) ˆ′ ˆ ∂w(h Λ 12 Λ22 ˆ fˆ θˆ , gˆ βˆ Whˆ 2 fˆ θˆ − gˆ βˆ + 0.5 ∂h2 J ′ K h, ( )′ ( )′ ( ( ( )) ( )) ∂ fˆ θˆ ∂ˆ g βˆ ′ ′ ˆ ˆ ˆ ˆ ˆ ˆ ˆ A ∂θ Whˆ 2 h2 − f θ − B ∂β Whˆ 2 h2 − gˆ β × ( ) , ( ( ) ˆ 2 )′ ( ) ( ) ( )) ∂w h ( ˆ fˆ θˆ , gˆ βˆ Whˆ 2 fˆ θˆ − gˆ βˆ + 0.5 ∂h2 J ′ K h, ( ( )) ′ ( ˆ ˆ ˆff W h − f θˆ 2 ˆ2 n Λ h ( σ ˆ22 = ( )) ˆ′ ˆ 2 − gˆ βˆ Λ R −Whˆ 2 h fg ˆfg Λ ˆ gg Λ (3.11) ( ) ( )) ˆ 2 − fˆ θˆ Whˆ 2 h ( ( )) . ˆ 2 − gˆ βˆ −Whˆ 2 h (3.12) In the expression for σ ˆ12 , K(h, f, g) = ((h − g) ⊗ (h − g)) − ((h − f ) ⊗ (h − f )) , 51 (3.13) The asymptotic variance formula is explained in Appendix C.3 96 3.4. Econometric Methodology vector w(h2 ) collects the element of Wh2 without duplicates, and J denotes a known m2 × m selection matrix of zeros and ones such that vec(Wh2 ) = Jw(h2 ). (3.14) For example, when Wh2 is a diagonal matrix with the reciprocals of the elements of h2 on the main diagonal, we have that wi (h) = 1/hi , i = 1, . . . , m, and J1 J = ... , Jm where, for i = 1, . . . , m, Ji is an m × m matrix with 1 in position (i, i) and zeros everywhere else. In (3.10), the ﬁrst term, σ ˆ12 , reﬂects the uncertainty due to estimation of θ, β, and h2 . For example, when comparing the models at some known ﬁxed ¯ matrices Aˆ and B ˆ should be replaced by zeros. parameter values θ¯ and β, Similarly, when comparing the models using a known ﬁxed weight matrix ˆ ′ /∂h)J ′ K(h, ˆ fˆ, gˆ) in (3.11) should (independent of h2 ), the terms 0.5(∂w(h) be replaced with zeros. The second term in (3.10), σ ˆ22 , is due to the simulations uncertainty in ˆ and gˆ(β). ˆ This term is zero when f and g can be computation of fˆ(θ) evaluated numerically (without resorting to simulations). Uncertainty due to simulations can be ignored if one can select a large number of simulations R so that the ratio n/R is suﬃciently small. Our asymptotic test with signiﬁcance level α is: √ Reject H0 in favor of Hf when nS/ˆ σ > z1−α/2 , √ σ < −z1−α/2 , Reject H0 in favor of Hg when nS/ˆ where z1−α/2 denotes a standard normal critical value. 3.4.2 Comparison of Model Classes When there are general classes of models with each class containing several sub-models, the researcher may be interested in overall comparison of classes instead of pairwise comparison of each sub-model. We discuss such a procedure in this section. Suppose that we have two classes of models with k models in each class: F = {f1 (θ), . . . , fk (θ)} and G = {g1 (β), . . . , gk (β)}. We are interested in comparing the overall performances of F and G. More speciﬁcally, we are 97 3.4. Econometric Methodology testing whether F and G have the same distance from the moments vector h2 . Here we adopt the von Mises-type (or average) distance between a set F and a point h2 : DM (F, h2 ) = k ∑ d(fj (θ), h2 ; Wh2 ), j=1 where d(fj (θ), h2 ; Wh2 ) denotes the previously used weighted Euclidean distance between vectors fj (θ) and h2 : d(fj (θ), h2 ; Wh2 ) = (h2 − fj (θ))′ Wh2 (h2 − fj (θ)). Note that, alternatively, one could use a Kolmogorov-type distance between F and h2 : Dmin (F, h2 ) = minj=1,...,k d(fj (θ), h2 ; Wh2 ) or Dmax (F, h2 ) = maxj=1,...,k d(fj (θ), h2 ; Wh2 ). While with a Kolmogorov-type distance each class is represented by its best (or worst) performer, the von Mises-type distance measures the average performance of a class of models, and we ﬁnd it more appropriate when the object of interest is the overall performance of a class. Thus, our null hypothesis of interest can now be stated as H0 : DM (F, h2 ) = DM (G, h2 ), (3.15) and a test can be based on the diﬀerence of sample analogues of DM (F, h2 ) and DM (G, h2 ):52 SM = k ( ) ∑ ˆ h ˆ 2 ; Wˆ ) − d(fˆj (θ), ˆ h ˆ 2 ; Wˆ ) . d(ˆ gj (β), h2 h2 j=1 Let σ ˆM denote the standard error of S M . As before, the null hypothesis in √ (3.15) should be rejected when the studentized statistic nS M /ˆ σM exceeds standard normal critical values. The standard error can be computed as follows.53 Deﬁne ( ) ∂fj (θ)′ ∂gj (β)′ A′ ∂θ Wh2 (h2 − fj (θ)) − B ′ ∂β Wh2 (h2 − gj (β)) Qj = , ′ 2) ′ K(h , f (θ), g (β)) J Wh2 (fj (θ) − gj (β)) + 0.5 ∂w(h 2 j j ∂h 52 We assume here, as in our case, that the same estimator of structural parameters is used inside each class of models. A generalization allowing for model-speciﬁc estimators inside each class is straightforward. 53 The details of the derivation are provided in Appendix C.4. 98 3.5. Empirical Results ˆ j denote a consistent estimator of Qj . Ignoring the simulation and let Q uncertainty, the standard error of S M is given by the square-root of ′ ( ) ∑ k k ∑ ˆ ˆ Λ11 Λ12 2 ˆj ˆj . σ ˆM = 4 Q Q (3.16) ˆ′ ˆ 22 Λ Λ j=1 12 j=1 The expression in (3.16) can be easily adjusted to account for simulation uncertainty. Note that the formula will depend on whether each model is simulated independently or if the same simulated structural shocks used in all models. In our case, the number of simulations is suﬃciently large for the simulation uncertainty to be ignored. 3.5 Empirical Results In this section we present the ﬁndings from the numerical solutions of our models and model comparisons. We conduct model comparisons based on two sets of moments: volatilities of endogenous variables and correlations, which include co-movements of key macroeconomic aggregates with output and cross-country correlations. To perform the comparison, we estimate the corresponding moments in the U.S. quarterly data over the period of 1973:1-2007:4. Details on data sources and calculations are provided in the Appendix C.1. We begin by presenting the results for the BKK and LAMP economies under the benchmark calibration. Then we conduct several extensions. In particular, (i) we consider model scenario in which labor input of participants and non-participants are imperfect substitutes by varying the elasticity of substitution between them, υ; (ii) we allow for investment-speciﬁc technology shocks. 3.5.1 Benchmark Case In this section we present the results from our simulations of BKK and LAMP models under the benchmark parameterization. Table 3.3 presents the volatilities of various macroeconomic aggregates in the data and in different versions of our models. Thus, panel (a) reports the statistics from the original BKK model speciﬁcation. Panel (b) reports the corresponding statistics in the model with LAMP under perfect substitutability in labor inputs of participants and non-participants. As in [47], ﬁnancial autarky model generates signiﬁcantly higher volatilities of exports, imports and especially relative prices, in comparison with 99 3.5. Empirical Results Table 3.3: Volatilities: Benchmark Calibration Table 3.3presents actual and simulated percent standard deviations for the U.S. economy. The data statistics are for the period of 1973:1-2007:4. Details on the data are available in the Appendix C.1. Model-based statistics are obtained from 10000 simulations, 100 periods long, each. All series, except net exports (nx), are logged and HP-filtered. The following models are considered: (a) original BKK; (b) BKK with LAMP. FA, BE and CM refer, respectively, to financial autarky, bond economy and complete markets economy. % std dev U.S. Data (a) BKK FA BE CM (b) LAMP FA BE CM y 1.49 % std dev % std dev of y c x n 0.62 2.92 0.68 % std dev ex 3.93 im 4.98 nx 0.50 ir 3.84 p 2.64 rx 3.55 0.98 1.01 1.01 0.54 0.54 0.55 1.86 2.71 2.73 0.25 0.29 0.29 1.07 0.81 0.82 1.07 0.81 0.82 0.00 0.20 0.21 1.37 0.75 0.70 1.53 0.83 0.78 1.00 0.54 0.51 0.95 0.97 0.97 0.59 0.62 0.63 1.67 2.47 2.49 0.20 0.23 0.22 1.03 0.79 0.80 1.03 0.79 0.80 0.00 0.19 0.20 1.31 0.67 0.62 1.46 0.75 0.69 0.95 0.49 0.45 the complete markets and bond economies; but implies lower volatilities of output, consumption, investment and employment relative to bond economy and complete markets economy. These results are driven by the inability of agents in the environment of ﬁnancial autarky to run trade imbalances. In such a framework, following productivity shocks, it is impossible to shift ﬁnal goods production to the country that has comparative advantage in doing so. As a result, a larger adjustment in relative prices, such as terms of trade, is needed to clear the markets. Such larger movements in the terms of trade under ﬁnancial autarky partially oﬀset the productivity changes (as in 29), thus reducing the incentives to work and invest. Consequently, employment, investment, output and consumption all become less volatile when no access to ﬁnancial assets is available. When agents become heterogeneous in terms of their access to ﬁnancial instruments, there are two key changes in the volatility characteristics of our economies. First, volatility of consumption increases across all ﬁnancial regimes; second, the volatility of all other variables declines across all ﬁnancial regimes. In our setup, introducing LAMP implies that asset markets become incomplete within a country. Namely, the non-participants can not trade any assets (neither ﬁnancial, nor real, like capital) and only consume their labor income. Their consumption, as a result becomes more volatile, thus raising the volatility of aggregate consumption in the country. On the other hand, employment is the only source of income for non-participants, as a result, their labor supply is inelastic. This implies that aggregate employment, output and investment, all become less volatile relative to the economy with no LAMP. Next, we evaluate the performance of our model in terms of co-movements with output. The results are summarized in Table 3.4. As before, the top 100 3.5. Empirical Results row of the table reports the co-movements in the data, while panels (a) and (b) report them, respectively, in the original BKK model and in the economy with LAMP. Table 3.4: Correlations with Output: Benchmark Calibration Table 3.4 presents actual and simulated correlations of macroeconomics aggregates with output for the U.S. economy. The data statistics are for the period of 1973:1-2007:4. Details on the data are available in the Appendix C.1. Model-based statistics are obtained from 10000 simulations, 100 periods long, each. All series, except net exports (nx), are logged and HP-filtered. The following models are considered: (a) original BKK; (b) BKK with LAMP. FA, BE and CM refer, respectively, to financial autarky, bond economy and complete markets economy. U.S. Data (a) BKK FA BE CM (b) LAMP FA BE CM correlation between im, y nx, y 0.82 -0.37 c, y 0.82 x, y 0.94 n, y 0.85 ex, y 0.42 0.89 0.93 0.94 0.99 0.95 0.95 0.98 0.96 0.96 1.00 0.55 0.50 0.07 0.80 0.84 0.93 0.97 0.97 0.99 0.95 0.95 0.99 0.97 0.97 1.00 0.54 0.48 0.08 0.83 0.87 p, y -0.16 rx, y 0.16 rx, c1 − c2 -0.17 0.01 -0.65 -0.65 0.64 0.64 0.64 0.64 0.64 0.64 0.96 0.99 0.99 0.01 -0.64 -0.64 0.64 0.63 0.62 0.64 0.63 0.62 0.99 0.97 0.97 As was the case for volatilities, the ﬁnancial autarky economy is the most distinct among our three ﬁnancial regimes. The fact that all trades in this economy must be quid pro quo implies that net exports are acyclical. Financial autarky also generates more procyclical exports and less procyclical imports relative to the bond and complete markets economies. In terms of these co-movements ﬁnancial autarky economy departs from the data relative to the other two ﬁnancial regimes. When LAMP is introduced, the comovement properties of the model do not change much. The only exception is the comovement of consumption with output, which increases when LAMP is introduced. The main reason is again the behavior of nonparticipants, whose work hours are inelastic, which in turn makes their wage income and thus consumption more sensitive to productivity changes. Consumption of non-participants, therefore, is more strongly procyclical than consumption of participants. This makes aggregate consumption move more closely with output relative to the original BKK framework. Both BKK and LAMP economies fail to replicate the negative correlation between real exchange rate and relative consumption of domestic to foreign economies that is observed in the data. This mismatch of theory and data is a well-known Backus-Smith puzzle due to [9] and [62]. Adding LAMP reduces this correlation, but only marginally. Lastly, we summarize the model performance based on cross-country co-movements of various macroeconomic aggregates. Table 3.5 reports our results. The top row reports the estimates in the data, the second panel summarizes them in the BKK economies, and the bottom panel - in the economies with LAMP. There are several puzzles associated with the cross101 3.5. Empirical Results country correlations, and they can be seen clearly from Table 3.5. First, is the fact that consumption is less correlated than output across countries in the data, while models predict the opposite (“quantity” puzzle). Second, in the data the correlations of investment and employment across countries are positive, while complete markets and bond economy models predict negative correlations (“international comovement” puzzle). Financial autarky, on the other hand, generates investment and employment across countries that are positively correlated, consistent with the data. So, as was the case with volatilities, ﬁnancial autarky model seems to provide a better match to the data even when it comes to the cross-country co-movements. Table 3.5: Cross-country Correlations: Benchmark Calibration Table 3.5presents actual and simulated cross-country correlations for the U.S. economy and the rest of the world. The data statistics are for the period of 1973-2007. Details on the data are available in the Appendix C.1. Modelbased statistics are obtained from 10000 simulations, 100 periods long, each. All series are logged and HP-filtered. The following models are considered: (a) original BKK; (b) BKK with LAMP. FA, BE and CM refer, respectively, to financial autarky, bond economy and complete markets economy. U.S. Data (a) BKK FA BE CM (b) LAMP FA BE CM y1 , y2 0.58 c1 , c2 0.43 correlation between x1 , x2 0.41 n1 , n2 0.45 0.16 0.08 0.09 0.86 0.68 0.64 0.29 -0.42 -0.43 0.01 -0.32 -0.29 0.17 0.12 0.13 0.79 0.57 0.53 0.37 -0.39 -0.40 0.10 -0.22 -0.18 Adding agents’ heterogeneity in asset market access works towards resolving these puzzles. In particular, LAMP reduces the cross-country correlation of consumption, while simultaneously increasing it for output; and does so for all three ﬁnancial regimes considered. It also signiﬁcantly increases the cross-country correlation in investment and employment. To understand these results, consider what happens to employment, investment, consumption and output in the economy with a representative households following a positive productivity shock. The country experiencing a productivity improvement (say, home country) sees its real wages rise, leading to an increase in labor supply, output and investment. At the same time, following the shock, the terms of trade depreciate in the home country, thus making foreign households relatively wealthier.54 As a result, 54 There are several channels through which wealth eﬀect in the foreign country arises following productivity improvement in the home country. First is the fact that productivity shocks spill over across countries. Second, is the terms of trade eﬀect mentioned in the text. Third eﬀect works through the world interest rate (whenever any assets are traded across countries). In particular, interest rate in the country experiencing a productivity improvement rises, creating an additional positive wealth eﬀect for foreign households, 102 3.5. Empirical Results they reduce their labor supply, lowering real output. For consumption in the foreign country to go up, investment must fall. When markets are complete or a single non-contingent bond is available these adjustments imply a negative correlation of employment and investment between home and foreign economies. In the ﬁnancial autarky, where shifting production across countries is not an option, terms of trade must adjust to eliminate the incentives to do so. These terms of trade movements are larger than in the bond or complete market economies as was argued before. By oﬀsetting some of the productivity improvement in the home country, terms of trade adjustment implies that output, consumption, investment and employment in this country increase by less under ﬁnancial autarky than under bond or complete market regimes. Correspondingly, in the foreign country, these macroeconomic aggregates increase by more as foreign households take advantage of larger favorable terms of trade movements. These adjustments imply positive cross-country correlations under ﬁnancial autarky. Adding LAMP changes these dynamics. With LAMP non-participants allocate all their time endowment to work. Thus, only households participating in the asset and capital markets adjust their labor supply following the shock. Consequently, aggregate labor supply in both countries responds to shocks less relative to the economy with a representative agent. This results in larger cross-country correlation of hours and output in the LAMP economy.55 In the bond and complete market economies this reduces the incentives to shift production across countries following the shocks and increases the cross-country correlation in investment. With investment responding less, so does consumption, thus lowering consumption correlation across countries. This result highlights how the absence of risk-sharing within a country spills into lower international risk-sharing. The results above show that diﬀerent versions of our model perform better in matching diﬀerent data characteristics. Financial autarky economy does best in matching volatilities of macroeconomic aggregates, but can not account for the cyclical properties of trade variables. Complete markets and bond economies do better in accounting for the cyclical properties of the data, but under-perform in terms of volatilities and cross-country correlations. Adding LAMP has three key eﬀects in the model: (i) it raises the who want to lend following the shock. 55 While adding LAMP moderates the dynamics of hours, it is quantitatively distinct from lowering labor supply elasticity in the representative agent BKK economy. In fact, our experiments show that reducing labor supply elasticity in the BKK economy leads to a larger negative correlation of labor inputs and investment across countries; as well as larger positive cross-country correlation of consumption. This deepens the puzzles. 103 3.5. Empirical Results sensitivity of aggregate consumption to income shocks; (ii) it reduces the sensitivity of aggregate labor supply to productivity shocks; (iii) it makes investment less responsive to productivity shocks. These eﬀects improve models performance primarily in matching cross-country correlations of consumption, output, investment and employment, but worsen their performance in matching volatilities. Given these results, a formal statistical test is necessary to aggregate various characteristics and pick a winner among our model variants. We turn to this next. 3.5.2 Comparison Results To determine which version of the model described above provides the best ﬁt to the data, we apply our test described in Section 3.4.1 to all possible pair-wise model comparisons. Our null hypothesis is that any two models considered provide an equivalent ﬁt to the data. We evaluate each model’s performance based on three criteria: (i) its ability to match volatilities; (ii) its ability to match co-movements with output and cross-country correlations; and (iii) on its overall performance which aggregates all of the aforementioned characteristics. Aggregation of model characteristics is equivalent to choosing the weight matrix Wh2 deﬁned above and deserves a special note. The simplest approach would be to assign equal weights to all model characteristics, that is to use an identity weighting matrix. Such an approach, however, may not be very informative if diﬀerent data characteristics have signiﬁcantly diﬀerent scales. For instance, in our case, variances can take any non-negative values, while correlations are restricted by [−1, 1] interval. Thus, if we use an identity weighting matrix to aggregate across such variances and correlations, the overall model performance will be heavily inﬂuenced by its performance for the moments that are larger - variances in our case. To account for the diﬀerences in scale we utilize a data-dependent weighting scheme in which various characteristics of interest are brought to a common base, thus facilitating their aggregation. According to our weighting scheme, prediction errors are assigned weights that are inversely related to the values of corresponding moments in the data. This way, instead of measuring absolute distance between the model and data to construct the test statistic as with the identity weighting matrix, we measure the percentage error made by the model relative to the data. In other words, we compute the ratio of the distance between model and data over the data moment, (h − f )/h where f is the model moment and h is its data counterpart, and 104 3.5. Empirical Results use it to construct a scale-free measure of the test statistic.56 We use such percentage errors in the case of volatilities. Since correlations are unit-free, we aggregate them using simple prediction errors.57 The comparison results for BKK and LAMP models with various ﬁnancial structures are presented in Table 3.6. Three panels in the table identify the set of characteristics based on which we conduct the comparisons: variances (panel (a)), co-movements with output and cross-country correlations (panel (b)), and overall performance (panel (c)). The test statistic is computed as the diﬀerence between the loss function of the model in the row (model g) and the loss function of the model in the column (model f ). Therefore, a positive sign of the test statistic implies that the model in the row does worse in matching data moments as compared with the model in the column. In addition, the larger the test statistic, the worse the model in the row performs. We report p-values in parenthesis below the test statistics. First, consider the original BKK models. Among the three ﬁnancial regimes our test picks ﬁnancial autarky as the winning speciﬁcation based on volatilities, correlations and the overall performance. This result is in accord with the informal ﬁndings in the literature. Extending comparisons to include LAMP economies, our test results show that ﬁnancial autarky with no LAMP outperforms all other models based on volatilities. Its superior performance for volatilities is statistically signiﬁcant in all ﬁve pair-wise comparisons. Based on co-movements with output and cross-country correlations, our test picks complete market economy with LAMP as the model that matches data best. This result is also statistically signiﬁcant in three pair-wise comparisons out of ﬁve possible. Finally, based on the overall performance, complete markets economy with LAMP outperforms all other models, with most comparisons being highly statistically signiﬁcant. The only exception is autarky economies, but the diﬀerence in their overall performance is not statistically signiﬁcant. The pair-wise model comparisons discussed above are informative in iso56 To bring the variances to a (0,1) base that is more comparable with the correlations we apply a logistic re-scaling to them. 57 Another possibility is to use a weight matrix that is inversely related to the asymptotic variance-covariance matrix of data moments. Such an approach gives a scale-free measure of ﬁt which is reminiscent of the GMM approach, i.e. moments that are more precisely estimated are assigned greater weights. Note, however, that in the time-series context where the asymptotic variance-covariance matrix is estimated by HAC methods, the uncertainty in the estimation of the weight matrix will dominate the uncertainty in estimation of parameters and moments. This is because HAC estimators converge at a slower than square root-n rate (see, for example, [45] for details. As a result, such a test may have poor power in ﬁnite samples. 105 3.5. Empirical Results Table 3.6: Test Results from Benchmark Models Comparisons Table 3.6 reports the test statistics for comparison of the model in the row (model g) against the model in the column (model f ). Positive numbers for the test statistic indicate that, compared with the model in the column, the model in the row provides a worse fit to the data moments. P-values are in the parentheses. * p-value≤0.10, ** p-value≤0.05, *** p-value≤0.01. Model f Model g (a) Volatilities BKK, FA FA 0 BKK BE 0.19*** 0 (0.00) 0.20*** 0.02*** (0.00) (0.00) LAMP, FA 0.03*** -0.16*** (0.00) (0.00) LAMP, BE 0.23*** 0.04*** (0.00) (0.00) LAMP, CM 0.25*** 0.06*** (0.00) (0.00) (b) Correlations (with output and cross-country) BKK, FA 0 CM FA LAMP BE CM BKK, BE BKK, CM BKK, BE BKK, CM LAMP, FA LAMP, BE LAMP, CM (c) Overall BKK, FA BKK, BE BKK, CM LAMP, FA LAMP, BE LAMP, CM LAMP - BKK class comparison 0.13 (0.77) 0.05 (0.91) -0.12 (0.40) -0.22 (0.57) -0.31 (0.42) FA 0 0.32 (0.47) 0.25 (0.57) -0.09 (0.52) 0.01 (0.97) -0.06 (0.87) 0 -0.17*** (0.00) 0.03*** (0.00) 0.05*** (0.00) 0 0.20*** (0.00) 0.22*** (0.00) 0 0.02*** (0.00) 0 0 -0.08*** (0.00) -0.25 (0.65) -0.35*** (0.00) -0.45*** (0.00) BE 0 -0.17 (0.76) -0.27** (0.03) -0.36*** (0.01) CM 0 -0.10 (0.84) -0.19 (0.68) FA 0 -0.09*** (0.00) BE 0 CM 0 -0.07*** (0.00) -0.41 (0.46) -0.31*** (0.01) -0.38*** (0.01) 0 -0.34 0 (0.54) -0.24** 0.10 (0.05) (0.83) -0.31** 0.03 (0.02) (0.95) -0.71* (0.07) 0 -0.08*** (0.00) 0 lating the combinations of model features that produce the closest ﬁt to the data (e.g. LAMP and complete markets, as above). But does the asset participation margin improve model performance independent of the assumed international ﬁnancial regime? To provide such an evaluation we apply our class-based test from Section 3.4.2. The test evaluates the overall performance of LAMP class of models relative to a representative agent BKK class of models by aggregating the ﬁts across the three ﬁnancial regimes within each class. We test whether these two model classes have the same distance from the data moments. We ﬁnd that LAMP speciﬁcation provides a better ﬁt to the data relative to a representative agent benchmark, and the diﬀerence is statistically signiﬁcant. More precisely, the resulting test statistic, reported at the bottom of Table 3.6, is -0.71 in favor of LAMP with the standard error of 0.39 and the resulting p-value of 0.07. 106 3.5. Empirical Results 3.5.3 Extensions Next we consider the robustness of our results with respect to parameter υ and to the presence of IST shocks. The results under alternative calibrations are presented in Table 3.7 for volatilities, Table 3.8 for correlations with output, and Table 3.9 for cross-country correlations. Panel (a) in each table presents the results for the case of imperfect substitutability in labor inputs of participants and non-participants, where we set the elasticity parameter υ to 0.5. Panels (b) and (c) report, respectively, second moments from the speciﬁcations of BKK and LAMP models with IST shocks. In each exercise we keep all remaining parameters unchanged. Table 3.7: Volatilities: Robustness Table 3.7 presents actual and simulated volatilities for the U.S. economy. All data statistics are for the period of 1973:1-2007:4. Details on the data are available in the Appendix C.1. Model-based statistics are obtained from 10000 simulations, 100 periods long, each. All series, except net exports (nx), are logged and HP-filtered. The following models are considered: (a) LAMP with imperfectly substitutable labor input of participants and non-participants; (b) BKK with IST shocks; (c) LAMP with IST shocks. FA, BE and CM refer, respectively, to financial autarky, bond economy and complete markets economy. % std dev U.S. Data (a) LAMP, υ = 0.5 FA BE CM (b) BKK with IST FA BE CM (c) LAMP with IST FA BE CM % std dev % std dev of y x n 2.92 0.68 y 1.49 c 0.62 0.90 0.92 0.91 0.65 0.69 0.70 1.49 2.25 2.27 1.00 1.04 1.04 0.61 0.57 0.57 0.96 0.99 0.98 0.63 0.62 0.63 % std dev ex 3.93 im 4.98 nx 0.50 ir 3.84 p 2.64 rx 3.55 0.13 0.15 0.15 0.98 0.76 0.78 0.98 0.76 0.78 0.00 0.18 0.19 1.24 0.60 0.55 1.38 0.67 0.61 0.90 0.44 0.40 2.27 3.86 3.90 0.38 0.45 0.45 1.09 1.42 1.48 1.09 1.43 1.49 0.00 0.33 0.35 1.41 1.03 0.89 1.57 1.15 1.15 1.02 0.75 0.75 1.97 3.48 3.52 0.30 0.36 0.35 1.04 1.44 1.50 1.04 1.44 1.50 0.00 0.33 0.34 1.34 1.04 0.68 1.48 1.16 1.17 0.97 0.76 0.77 Consider ﬁrst the scenario where labor inputs of participants and nonparticipants are imperfectly substitutable with elasticity υ = 0.5. In this case, the distinction between the original BKK and LAMP models becomes quantitatively sharper. In particular, relative to the case of perfect substitutability between two labor types reported in panel (b) of Tables 3.3, 3.4, and 3.5, volatility of consumption rises further, while volatilities of all other aggregates fall. Reducing elasticity of substitution in labor has the largest eﬀect on cross-country correlations. In particular, it signiﬁcantly lowers cross-correlation of consumption, and raises the cross-correlation of employment and investment. These changes are reﬂected in the formal model comparison. We ﬁnd that while qualitatively, our test results remain unchanged, quantitatively they become stronger and more signiﬁcant.58 58 Given that test results do not change qualitatively in this case, we do not report them in the paper. These results are available from the authors upon request. 107 3.5. Empirical Results Table 3.8: Correlations with Output: Robustness Table 3.8 presents actual and simulated correlations of macroeconomics aggregates with output for the U.S. economy. The data statistics are for the period of 1973:1-2007:4. Details on the data are available in the Appendix C.1. Model-based statistics are obtained from 10000 simulations, 100 periods long, each. All series, except net exports (nx), are logged and HP-filtered. The following models are considered: (a) LAMP with imperfectly substitutable labor input of participants and non-participants; (b) BKK with IST shocks; (c) LAMP with IST shocks. FA, BE and CM refer, respectively, to financial autarky, bond economy and complete markets economy. U.S. Data (a) LAMP, υ = 0.5 FA BE CM (b) BKK with IST FA BE CM (c) LAMP with IST FA BE CM correlation between im, y nx, y 0.82 -0.37 p, y -0.16 rx, y 0.16 rx, c1 − c2 -0.17 0.01 -0.63 -0.63 0.63 0.61 0.61 0.63 0.61 0.61 1.00 0.95 0.95 0.06 0.59 0.60 0.02 -0.52 -0.52 0.65 0.35 0.30 0.65 0.35 0.30 0.26 0.98 0.95 0.08 0.56 0.57 0.02 -0.48 -0.48 0.64 0.31 0.26 0.64 0.31 0.26 0.55 0.78 0.67 c, y 0.82 x, y 0.94 n, y 0.85 ex, y 0.42 0.95 0.98 0.99 0.99 0.95 0.95 0.99 0.97 0.98 1.00 0.52 0.46 0.10 0.86 0.89 0.69 0.78 0.81 0.89 0.78 0.78 0.75 0.74 0.74 1.00 0.17 0.13 0.80 0.90 0.92 0.90 0.77 0.76 0.74 0.72 0.72 1.00 0.18 0.15 Next, we turn to IST shocks. The simulated moments for the original BKK models with IST shocks are shown in panel (b) of Tables 3.7, 3.8, 3.9; while those for the LAMP model with IST shocks are in panel (c) of the same three tables. Not surprisingly, when IST shocks are introduced, all volatilities go up, especially for investment, international trade variables and relative prices. This increase is particularly pronounced in the bond economy and complete markets economy. Correlations with output, on the other hand, decline. Cross-country correlations of output and consumption also fall, while those of investment and employment turn more negative. These changes are characteristic of both BKK and LAMP economies. Table 3.9: Cross-country Correlations: Robustness Table 3.9 presents actual and simulated cross-country correlations for the U.S. economy and the rest of the world. The data statistics are for the period of 1973:1-2007:4. Details on the data are available in the Appendix C.1. Model-based statistics are obtained from 10000 simulations, 100 periods long, each. All series are logged and HP-filtered. The following models are considered: (a) LAMP with imperfectly substitutable labor input of participants and non-participants; (b) BKK with IST shocks; (c) LAMP with IST shocks. FA, BE and CM refer, respectively, to financial autarky, bond economy and complete markets economy. U.S. Data (a) LAMP, υ = 0.5 FA BE CM (b) BKK with IST FA BE CM (c) LAMP with IST FA BE CM correlation between x1 , x 2 0.41 y 1 , y2 0.58 c1 , c 2 0.43 n1 , n2 0.45 0.18 0.15 0.15 0.73 0.48 0.44 0.44 -0.38 -0.38 0.18 -0.12 -0.06 0.14 0.05 0.06 0.54 0.59 0.57 0.11 -0.64 -0.64 -0.17 -0.48 -0.45 0.17 0.09 0.10 0.63 0.56 0.53 0.21 -0.62 -0.62 -0.08 -0.39 -0.36 What is behind these results? As in [77] and [69], IST shocks in our setup 108 3.5. Empirical Results act as demand shocks. For instance, consider a positive IST shock in the domestic economy. Following this shock, domestic investment demand goes up, appreciating home terms of trade, on impact. To accommodate higher investment demand, domestic households must reduce their consumption. In bond and complete market economies, imports from abroad also rise to ﬁnance domestic investment boom, leading to trade deﬁcit. Domestic households also increase their labor supply in response to the shock. As home output goes up and investment demand subsides (with temporary IST shocks), domestic terms of trade begin to depreciate. So does the real exchange rate. The impact appreciation of the terms of trade and real exchange rate, followed by depreciation some quarters later helps understand the higher volatility of these variables in the economy with IST shocks. Foreign economy, on the other hand, being relatively less productive, cuts down its investment and employment. Released resources are used for temporarily higher consumption by foreign households. These dynamics imply low (for consumption) or negative (for output, employment and investment) cross-country correlations after IST shocks. Overall, adding temporary IST shocks to our benchmark economies helps improve their performance on some dimensions, such as volatilities and some correlations. However, the models ﬁt also worsens in some other dimensions, such as cross-country co-movements of investment and employment. As a result, a formal statistical method of model comparison is again warranted. Our results from comparison of models with IST shocks are presented in Table 3.10, where as before, to measure overall performance we aggregate variances and covariances using data-dependent weighting matrix. In the presence of IST shocks our test picks BKK bond economy as the preferred model speciﬁcation among all pair-wise comparisons when the objective is to match volatilities. The results are statistically signiﬁcant in all but one pair. If the objective is to match correlations, out test implies that BKK autarky with IST shocks comes out at the top. When the overall performance (variances and correlations) is considered, BKK autarky economy with IST shocks is chosen as the winner, although this superior performance is statistically signiﬁcant in only two out of ﬁve possible pairs. Turning to the comparison between BKK and LAMP models classes, we ﬁnd that LAMP with IST shocks outperforms the original BKK representative agent model with IST shocks and that this superior performance is highly statistically signiﬁcant. More precisely, the test statistic for the overall test between LAMP and BKK model classes is -1.34 in favor of LAMP, with the standard error of 0.29 and implied p-value of 0.00 (see the bottom of Table 3.10). These results imply that also in the presence of IST shocks, 109 3.6. Conclusion Table 3.10: Comparison Results of Models with IST Shocks Table 3.10 reports the test statistics for comparison of the model in the row (model g) against the model in the column (model f ). Positive numbers for the test statistic indicate that, compared with the model in the column, the model in the row provides a worse fit to the data moments. P-values are in the parentheses. * p-value≤0.10, ** p-value≤0.05, *** p-value≤0.01. Model f Model g (a) Volatilities BKK, FA FA 0 BKK BE -0.01 0 (0.65) -0.00 0.01** (0.89) (0.03) LAMP, FA 0.03*** 0.05* (0.00) (0.09) LAMP, BE -0.01 0.01* (0.83) (0.10) LAMP, CM 0.03 0.04*** (0.29) (0.00) (b) Correlations (with output and cross-country) BKK, FA 0 CM FA LAMP BE CM BKK, BE BKK, CM BKK, BE BKK, CM LAMP, FA LAMP, BE LAMP, CM (c) Overall BKK, FA BKK, BE BKK, CM LAMP, FA LAMP, BE LAMP, CM 1.19*** (0.00) 1.02** (0.02) 0.12 (0.54) 0.48 (0.20) 0.21 (0.60) 0 0.04 (0.14) -0.00 (0.66) 0.03*** (0.00) 0 -0.04 (0.18) -0.01 (0.86) 0 0.03*** (0.00) 0 0 -0.17*** (0.00) -1.08** (0.03) -0.71*** (0.00) -0.99*** (0.00) 0 -0.90* (0.07) -0.54*** (0.00) -0.81*** (0.00) 0 0.37 (0.42) 0.09 (0.85) 0 -0.28*** (0.00) 0 0 1.18*** (0.00) 1.02*** (0.01) 0.15 (0.43) 0.48 (0.20) 0.23 (0.54) LAMP - BKK class comparison 0 -0.17*** (0.00) -1.03** (0.03) -0.70*** (0.00) -0.95*** (0.00) 0 -0.87* (0.08) -0.54*** (0.00) -0.78*** (0.00) 0 0.33 (0.47) 0.08 (0.86) -1.34*** (0.00) 0 -0.24*** (0.00) 0 LAMP delivers a better match to the data. 3.6 Conclusion In this paper we propose a novel statistical test to conduct evaluation and formal comparison of DSGE models. Our procedure explicitly accounts for the possibility that a DSGE model might be misspeciﬁed. It also accounts for simulation uncertainty, the fact that some model parameters are estimated rather than calibrated, and allows for both pair-wise comparison of models and comparison of model classes. We apply our test to a standard international business cycles model with three speciﬁcations for asset markets structure: ﬁnancial autarky, single risk-free bond economy, and an economy 110 3.6. Conclusion with complete asset markets. We ﬁnd that ﬁnancial autarky economy indeed ﬁts the data best, in line with the informal ﬁndings in the literature. We then allow for domestic asset market incompleteness by introducing handto-mouth consumers that do not participate in the domestic or foreign ﬁnancial markets. With limited asset market participation (LAMP), the models’ performance is improved in matching cross-country correlations, but worsened in matching volatilities. Formal statistical comparison ﬁnds that the improvements brought out by LAMP are statistically signiﬁcant, allowing economies with LAMP to outperform the representative agent benchmark economies. The superior performance of LAMP is robust to lower substitutability in labor inputs of participants and non-participants, and to the presence of the investment-speciﬁc productivity shocks. 111 Bibliography [1] Almeida, H., M. Campello, and M. S. Weisbach (2004, August). The Cash Flow Sensitivity of Cash. The Journal of Finance 59 (4), 1777–1804. [2] Alvarez, F., A. Atkeson, and P. J. Kehoe (2002, February). Money, interest rates, and exchange rates with endogenously segmented markets. Journal of Political Economy 110 (1), 73–112. [3] Andrews, D. W. K. (1991, May). Heteroskedasticity and autocorrelation consistent covariance matrix estimation. Econometrica 59, 817–858. [4] Arellano, C., Y. Bai, and P. Kehoe (2011). Financial Markets and Fluctuations in Uncertainty. Working Paper . [5] Armenter, R. and V. Hnatkovska (2011). The Macroeconomics of Firms’ Savings. Working Paper . [6] Arvanitis, A. V. and A. Mikkola (1996). Asset-market structure and international trade dynamics. American Economic Review 86 (2), 67–70. [7] Backus, D., P. J. Kehoe, and F. E. Kydland (1995). International business cycles: Theory and evidence. In T. F. Cooley (Ed.), Frontiers of Business Cycle Research, pp. 331–356. Princeton: Princeton University Press. [8] Backus, D. K., P. J. Kehoe, and F. E. Kydland (1994). Dynamics of the trade balance and the terms of trade: The J-curve? American Economic Review 84 (1), 84–103. [9] Backus, D. K. and G. W. Smith (1993). Consumption and real exchange rates in dynamic economies with non-traded goods. Journal of International Economics 35 (3-4), 297–316. [10] Balakrishnan, R., T. Linsmeier, and M. Venkatachalam (1996). Financial Beneﬁts from JIT Adoption: Eﬀects of Customer Concertration and Cost Structure. The Accounting Review 71 (2), 183–205. 112 Bibliography [11] Bates, T. W., K. M. Kahle, and R. M. Stulz (2009, October). Why Do U.S. Firms Hold So Much More Cash than They Used To? The Journal of Finance 64 (5), 1985–2021. [12] Baumol, W. (1952). The Transactions Demand for Cash : An Inventory Theoretic Approach. The Quarterly Journal of Economics 66 (4), 545– 556. [13] Baxter, M. (1995). International trade and business cycles. In G. M. Grossman and K. Rogoﬀ (Eds.), Handbook of International Economics (1 ed.), Volume 3, Chapter 35, pp. 1801–1864. Elsevier. [14] Baxter, M. and M. J. Crucini (1995). Business cycles and the asset structure of foreign trade. International Economic Review 36 (4), 821– 854. [15] Belo, F. and X. Lin (2012). The Inventory Growth Spread. Review of Financial Studies 25 (1), 278–313. [16] Bilbiie, F. O. (2008, May). Limited asset markets participation, monetary policy and (inverted) aggregate demand logic. Journal of Economic Theory 140 (1), 162–196. [17] Blinder, A. (1986). Can the Production Smoothing Model of Inventory Behavior be Saved? The Quarterly Journal of Economics 101 (3), 431– 454. [18] Blinder, A. S. and L. J. Maccini (1991, December). The Resurgence of Inventory Research: What Have We Learned? Journal of Economic Surveys 5 (4), 291–328. [19] Boileau, M. and N. Moyen (2010). Corporate Cash Savings: Precaution versus Liquidity. Working Paper . [20] Brown, J. R. and B. C. Petersen (2011, June). Cash Holdings and R&D Smoothing. Journal of Corporate Finance 17 (3), 694–709. [21] Callen, J. L., C. Fader, and I. Krinsky (2000). Just-in-Time : A Cross Sectional Plant Analysis. International Journal of Production Economics 63, 277–301. [22] Campbell, J. Y., M. Lettau, and B. G. Malkiel (2001). Have Individual Stocks Become More Volatile ? An Empirical Exploration of Idiosyncratic Risk. Journal of Finance 56 (1), 1–43. 113 Bibliography [23] Chari, V. V., P. J. Kehoe, and E. R. McGrattan (2002). Can sticky price models generate volatile and persistent real exchange rates? Review of Economic Studies 69 (3), 533–63. [24] Chatterjee, S. and D. Corbae (1992). Endogenous market participation and the general equilibrium value of money. Journal of Political Economy 100 (3), 615–646. [25] Chen, H., M. Frank, and O. Wu (2005). What Actually Happened to the Inventories of American Companies between 1981 and 2000? Management Science 51 (7), 1015–1031. [26] Chien, Y.-L., H. L. Cole, and L. Hanno (2011). A multiplier approach to understanding the macro implications of household ﬁnance? Review of Economic Studies 78 (1), 199–234. [27] Chien, Y.-L., H. L. Cole, and L. Hanno (2012, October). Is the volatility of the market price of risk due to intermittent portfolio re-balancing? American Economic Review 102 (6), 2859–2896. [28] Christiano, L. (1988). Why Does Inventory Investment Fluctuate So Much? Journal of Monetary Economics 21, 247–280. [29] Cole, H. L. and M. Obstfeld (1991, August). Commodity trade and international risk sharing : How much do ﬁnancial markets matter? Journal of Monetary Economics 28 (1), 3–24. [30] Comin, D. and S. Mulani (2006). Diverging Trends in Aggregate and Firm Volatility. The Review of Economics and Statistics 88 (2), 374–383. [31] Comin, D. and T. Philippon (2005). The Rise in Firm-Level Volatility : Causes and Consequences. NBER Working Paper n.11388 . [32] Cooper, R. W. and J. C. Haltiwanger (2006, July). On the Nature of Capital Adjustment Costs. Review of Economic Studies 73 (3), 611–633. [33] Corsetti, G., L. Dedola, and S. Leduc (2008). International risk sharing and the transmission of productivity shocks. Review of Economic Studies 75 (2), 443–473. [34] Dittmar, A. and R. Duchin (2011). The Dynamics of Cash. Working Paper . 114 Bibliography [35] Engel, C. and A. Matsumoto (2009, July). The international diversiﬁcation puzzle when goods prices are sticky: It’s really about exchange-rate hedging, not equity portfolios. American Economic Journal: Macroeconomics 1 (2), 155–88. [36] Fazzari, S. M. and B. C. Petersen (1993). Working Capital and Fixed Investment. The RAND Journal of Economics 24 (3), 328–342. [37] Foley, F., J. Hartzell, S. Titman, and G. Twite (2007, December). Why Do Firms Hold So Much Cash? A Tax-based Explanation. Journal of Financial Economics 86 (3), 579–607. [38] Fullerton, R. R., C. S. McWatters, and C. Fawson (2003, July). An Examination of the Relationships between JIT and Financial Performance. Journal of Operations Management 21 (4), 383–404. [39] Gilchrist, S. and J. Sim (2010). Uncertainty, Financial Frictions, and Investment Dynamics. Working paper . [40] Gomes, J. (2001). Financing Investment. The American Economic Review 91 (5), 1263–1285. [41] Gourio, F. (2008). Estimating Firm-Level Risk Preliminary. Working paper . [42] Greenwood, J., Z. Hercowitz, and P. Krusell (2000, January). The role of investment-speciﬁc technological change in the business cycle. European Economic Review 44 (1), 91–115. [43] Grossman, S. and L. Weiss (1983, December). A transactions-based model of the monetary transmission mechanism. American Economic Review 73 (5), 871–80. [44] Hadlock, C. J. and J. R. Pierce (2010). New Evidence on Measuring Financial Constraints : Moving Beyond the KZ Index. Review of Financial Studies 23 (5), 1909–1940. [45] Hall, A. R. and A. Inoue (2003). The large sample behavior of the generalized method of moments estimator in misspeciﬁed models. Journal of Econometrics 114 (2), 361–394. [46] Haushalter, D., S. Klasa, and W. Maxwell (2007, June). The Inﬂuence of Product Market Dynamics on a Firm’s Cash Holdings and Hedging Behavior. Journal of Financial Economics 84 (3), 797–825. 115 Bibliography [47] Heathcote, J. and F. Perri (2002). Financial autarky and international business cycles. Journal of Monetary Economics 49 (3), 601–627. [48] Hennessy, C. a. and T. M. Whited (2005, July). Debt Dynamics. The Journal of Finance 60 (3), 1129–1165. [49] Hennessy, C. a. and T. M. Whited (2007, August). How Costly Is External Financing? Evidence from a Structural Estimation. The Journal of Finance 62 (4), 1705–1745. [50] Hnatkovska, V., V. Marmer, and Y. Tang (2011a). Comparison of misspeciﬁed calibrated models: The minimum distance approach. Journal of Econometrics, forthcoming. [51] Hnatkovska, V., V. Marmer, and Y. Tang (2011b). Supplement to comparison of misspeciﬁed calibrated models. Working Paper, University of British Columbia. [52] Holt, C. C., F. Modigliani, J. Muth, and H. Simon (1960). Planning Production, Inventories and Work Force. Prentice Hall. [53] Hugonnier, J., S. Malamud, and E. Morellec (2010). Capital Supply Uncertainty, Cash Holdings, and Investment. New York . [54] Huson, M. and D. Nanda (1995, June). The Impact of Just-in-Time Manufacturing on Firm Performance in the US. Journal of Operations Management 12 (3-4), 297–310. [55] Imrohoroglu, A. and S. Tuzel (2011). Firm Level Productivity, Risk, and Return. Working paper . [56] Irvine, P. and J. Pontiﬀ (2009). Idiosyncratic Return Volatility , Cash Flows , and Product Market Competition. Review of Financial Studies 22 (3). [57] Kahn, J. (1987, February). Inventories and the Volatility of Production. The American Economic Review 77 (4), 667–679. [58] Kahn, J. (1992). Why is production more volatile than sales? Theory and evidence on the stockout-avoidance motive for inventory-holding. The Quarterly Journal of Economics 107 (2), 481–510. [59] Kan, R. and C. Robotti (2008). Model comparison using the HansenJagannathan distance. Review of Financial Studies forthcoming. 116 Bibliography [60] Kinney, M. R. and W. F. Wempe (2002, January). Further Evidence on the Extent and Origins of JIT’s Proﬁtability Eﬀects. The Accounting Review 77 (1), 203–225. [61] Klapper, L., L. Laeven, and R. Rajan (2010). Trade Credit Contracts. Working paper . [62] Kollmann, R. (1996). Incomplete asset markets and the cross-country consumption correlation puzzle. Journal of Economic Dynamics and Control 20 (5), 945–961. [63] Kollmann, R. (2006). International portfolio equilibrium and the current account: A dynamic general equilibrium perspective. Mimeo, Free University of Brussels. [64] Krane, S. and S. Braun (1991, January). Production smoothing evidence from physical-product data. Journal of Political Economy 99 (3), 558–581. [65] Lins, K. V., H. Servaes, and P. Tufano (2010, October). What Drives Corporate Liquidity? An International Survey of Cash Holdings and Lines of Credit. Journal of Financial Economics 98 (1), 160–176. [66] Maccini, L. and R. Rossana (1984). Joint Production, Quasi-ﬁxed Factors of Production, and Investment in Finished Goods Inventories. Journal of Money, Credit and Banking 16 (2), 218–236. [67] Magnus, J. R. and H. Neudecker (1999). Matrix Diﬀerential Calculus with Applications in Statistics and Econometrics (Revised ed.). Chichester: Wiley. [68] Maiga, A. S. and F. A. Jacobs (2009). JIT Performance Eﬀects : A Research Note. Advances in Accounting 25 (2), 183–189. [69] Mandelman, F., P. Rabanal, J. F. Rubio-Ramirez, and D. Vilan (2011, January). Investment speciﬁc technology shocks and international business cycles: An empirical assessment. Review of Economic Dynamics 14 (1), 136–155. [70] Mankiw, N. G. and S. P. Zeldes (1991). The consumption of stockholders and nonstockholders. Journal of Financial Economics 29 (1), 97–112. [71] Morellec, E. and B. Nikolov (2009). Cash Holdings and Competition. Unpublished working paper. University of Rochester . 117 Bibliography [72] Newey, W. K. and K. D. West (1987, May). A simple, positive semideﬁnite, heteroskedasticity and autocorrelation consistent covariane matrix. Econometrica 55, 703–708. [73] Nikolov, B. and T. Whited (2010). Agency Conﬂicts and Cash: Estimates from a Structural Model. Unpublished working paper. University of Rochester . [74] Opler, T.; Pinkowitz, L.; Stulz, R.;Williamson, R. (1999, April). The determinants and implications of corporate cash holdings. Journal of Financial Economics 52 (1), 3–46. [75] Pinkowitz, L. and R. Wilianson (2001). Bank Power and Cash Holdings : Evidence from Japan. The Review of Financial Studies 14 (4), 1059– 1082. [76] Polkovnichenko, V. (2004). Limited stock market participation and the equity premium. Finance Research Letters 1 (1), 24 – 34. [77] Raﬀo, A. (2010). Technology shocks: novel implications for international business cycles. Technical report. [78] Richardson, H. (1995). Control Your Costs then Cut Them. Transportation 36 (12), 94–94. [79] Riddick, L. and T. Whited (2009). The Corporate Propensity to Save. The Journal of Finance 64 (4), 1729–1766. [80] Rossana, R. (2009). Cost Shocks and Inventories. Working paper . [81] Schorfheide, F. (2000). Loss function-based evaluation of dsge models. Journal of Applied Econometrics 15, 645–670. [82] Seta, M. D. (2011). Cash and Competition. Working Paper , 1–47. [83] Suﬁ, A. (2009). Bank Lines of Credit in Corporate Finance : An Empirical Analysis. Review of Financial Studies 22 (3), 1057–1088. [84] van Wincoop, E. (1996, June). A multi-country real business cycle model with heterogeneous agents. Scandinavian Journal of Economics 98 (2), 233–51. [85] Vissing-Jorgensen, A. (2002, August). Limited asset market participation and the elasticity of intertemporal substitution. Journal of Political Economy 110 (4), 825–853. 118 [86] Vuong, Q. H. (1989). Likelihood ratio tests for model selection and non-Nested hypotheses. Econometrica 57 (2), 307–333. [87] Walleigh, R. (1986). What’s Your Exercuse for Not Using JIT? Harvard Business Review 101 (3), 38–54. [88] White, R. E., J. N. Pearson, and J. R. Wilson (1999). JIT Manufacturing: A Survey of Implementations in Small and Large U.S. Manufacturers. Management Science 45 (1), 1–15. 119 Appendix A A.1 Variable Definitions Following [11], I construct the sample from Compustat and deﬁne variables used in the cash and inventory regressions as follows: Cash is deﬁned as the ratio of cash and short-term investments over total asset; Inventory is the ratio of total inventories over total asset; Firm size is the natural logarithm of total asset; Risk is computed as the standard deviation of annual operating cash ﬂow in the past ﬁve years, with operating cash ﬂow deﬁned as earnings after interest, dividends and tax but before depreciation divided by total asset; Market-to-book ratio is the sum of market value and debt over total asset; Net working capital is equal to working capital net of cash and inventory over total asset; Capital investment is the ratio of capital expenditure over total asset; Leverage is the sum of long-term debt and debt in current liabilities normalized by total asset; R&D investment is research and development expenses to total asset ratio; Dividend is a dummy variable taking value of one if dividend payout (common) is non-zero; Acquisition is the ratio of acquisition over total asset. A.2 Corporate Cash Holdings in Wholesale, Retail and Services As a robustness check, I also examine the correlation between cash and inventory in wholesale, retail and service industries. I re-estimate regression equation (1.1), and present results below in Tables A.1 and A.2. A negative, statistically signiﬁcant correlation between cash and inventory is also found in wholesale and retail trade, although the magnitude becomes smaller compared to manufacturing. With cross-sectional variations, the estimated magnitude of the correlation lies within the range 120 A.2. Corporate Cash Holdings in Wholesale, Retail and Services [0.329, 0.372]. Using within-ﬁrm over-time variations, the magnitude goes up to 0.524. Table A.1: Cash Holdings in Wholesale and Retail Table A.1 reports the estimation results of Regression (1.1) with a sample of ﬁrms operating in wholesale and retail industries from Compustat during 1980-2006. The heteroskedasticity-consistent standard errors reported in parenthesis account for possible correlation within a ﬁrm cluster. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1%, respectively. Variables Inventory Size Market-to-book Risk Cash ﬂow Net working capital Capital investment Leverage R&D Dividend Acquisition Industry FE (2-digit) Industry FE (3-digit) Industry FE (4-digit) Firm FE Year FE Cohort dummy Observations R-squared (1) Pooled OLS (2) Pooled OLS (3) Pooled OLS (4) Fixed Eﬀect -0.3289*** (0.0258) -0.0098*** (0.0022) 0.0113*** (0.0027) 0.0050 (0.0253) 0.0819*** (0.0152) -0.0857*** (0.0141) -0.3825*** (0.0481) -0.2086*** (0.0183) 0.1350 (0.1115) -0.0062 (0.0072) -0.1977*** (0.0309) -0.3666*** (0.0293) -0.0107*** (0.0021) 0.0113*** (0.0026) 0.0149 (0.0227) 0.0775*** (0.0152) -0.0784*** (0.0138) -0.3961*** (0.0485) -0.1976*** (0.0180) 0.0987 (0.1028) 0.0015 (0.0073) -0.1799*** (0.0312) -0.3718*** (0.0316) -0.0111*** (0.0026) 0.0116*** (0.0026) 0.0164 (0.0235) 0.0765*** (0.0160) -0.0737*** (0.0143) -0.3946*** (0.0483) -0.1939*** (0.0184) 0.0992 (0.1032) 0.0011 (0.0074) -0.1784*** (0.0314) -0.5241*** (0.0254) -0.0145*** (0.0036) 0.0089*** (0.0022) -0.0134 (0.0201) 0.0394*** (0.0138) -0.0521*** (0.0124) -0.2818*** (0.0262) -0.1484*** (0.0114) -0.1273* (0.0727) 0.0077*** (0.0025) -0.2571*** (0.0124) Yes Yes Yes Yes Yes Yes Yes Yes Yes 6,858 0.359 6,858 0.388 6,858 0.397 Yes Yes 6,858 0.722 As shown in Table A.2, similar results are found for service industries. There is a signiﬁcant negative correlation between cash and inventory. According to column (4), a 10 percentage-point decrease in inventory will be correlated with a 5.8 percentage-point increase in cash holdings. In addition, I show the dynamics of average cash and inventory ratios by industries in Figure A.1. The constant average cash and inventory ratio prevails in most industries, except for agriculture and wholesale. 121 0 .1 .1 .15 ratios .2 ratios .2 .25 .3 .3 .4 .35 A.2. Corporate Cash Holdings in Wholesale, Retail and Services 1970 1980 1990 year 2000 2010 1980 1990 2000 2010 year average inventory ratio averge cash ratio average cash and inventory ratio average inventory ratio ratios .3 0 .1 .05 .2 ratios .1 .15 .4 .2 .5 average cash ratio average cash and inventory ratio 1980 1990 2000 2010 1980 1990 year 2000 2010 year average inventory ratio average cash ratio average cash and inventory ratio average inventory ratio .1 .1 .2 .2 ratios .3 ratios .3 .4 .4 .5 average cash ratio average cash and inventory ratio 1970 1980 1990 year 2000 2010 1980 1990 2000 2010 year average inventory ratio average cash ratio average cash and inventory ratio average inventory ratio 0 .1 .1 .2 ratios ratios .2 .3 .3 .4 average cash ratio average cash and inventory ratio 1980 1990 2000 2010 year average cash ratio average cash and inventory ratio 1980 1990 2000 2010 year average inventory ratio average cash ratio average cash and inventory ratio average inventory ratio Figure A.1: Average Cash and Inventory Ratios by Industries. The ﬁgure summarizes the average cash-to-asset ratio, average inventory-to-asset ratio and sum of those two ratios over time in (1) All industries; (2) Agriculture; (3) Mining; (4) Construction; (5) Manufacturing; (6) Wholesale; (7) Retail; (8) Services. 122 A.3. Description of the JIT-adopter Sample Table A.2: Cash Holdings in Services Table A.2 reports the estimation results of Regression (1.1) with a sample of ﬁrms operating in services from Compustat during 1980-2006. The heteroskedasticityconsistent standard errors reported in parenthesis account for possible correlation within a ﬁrm cluster. Signiﬁcance levels are indicated by *, **, and *** for 10%, 5%, and 1%, respectively. Variables Inventory Size Market-to-book Risk Cash ﬂow Net working capital Capital investment Leverage R&D Dividend Acquisition Industry FE (2-digit) Industry FE (3-digit) Industry FE (4-digit) Firm FE Year FE Cohort dummy Observations R-squared A.3 (1) Pooled OLS (2) Pooled OLS (3) Pooled OLS (4) Fixed Eﬀect -0.3953*** (0.0336) 0.0032 (0.0024) 0.0104*** (0.0015) 0.0421*** (0.0140) 0.0238*** (0.0086) -0.0204** (0.0089) -0.4162*** (0.0524) -0.3200*** (0.0193) 0.1111*** (0.0286) -0.0264** (0.0129) -0.4507*** (0.0343) -0.3928*** (0.0343) 0.0040* (0.0025) 0.0100*** (0.0014) 0.0405*** (0.0138) 0.0234*** (0.0085) -0.0181** (0.0090) -0.3832*** (0.0533) -0.3045*** (0.0193) 0.0936*** (0.0292) -0.0317*** (0.0129) -0.4545*** (0.0339) -0.3820*** (0.0359) 0.0026 (0.0025) 0.0096*** (0.0014) 0.0389*** (0.0140) 0.0227*** (0.0086) -0.0157* (0.0089) -0.3339*** (0.0530) -0.2948*** (0.0197) 0.0730** (0.0298) -0.0219* (0.0123) -0.4536*** (0.0336) -0.5806*** (0.0436) -0.0087** (0.0043) 0.0066*** (0.0010) 0.0156 (0.0144) 0.0104 (0.0081) -0.0165* (0.0093) -0.3576*** (0.0390) -0.1861*** (0.0154) -0.0471* (0.0283) 0.0047 (0.0088) -0.3770*** (0.0299) Yes Yes Yes Yes Yes Yes Yes Yes Yes 7,713 0.356 7,713 0.371 7,713 0.380 Yes Yes 7,713 0.702 Description of the JIT-adopter Sample Table A.3 provides the distribution of the JIT adoption year for the sample of 169 JIT adopters. About 11% of the ﬁrms in the sample adopted JIT in the ﬁrst half of 1980s (1980-1984), with the earliest in 1982. Over 50% of the sample ﬁrms implemented JIT in the second half of 1980s. The number of adopters in the sample reached the peak in 1990, and declined since then. Table A.4 reports the distribution of adopters by two-digit SIC industry. A large portion (approximately 70%) of adopters operate in four industries. In order by number, these industries are: electronic equipment (SIC 36, 23.7%), industrial equipment (SIC 35, 21.9%), instrumentation (SIC 38, 13%), and motor vehicles (SIC 37, 10.7%). The rest of adopters in the sample are relatively evenly distributed in other industries. 123 A.4. Work-in-process Inventory and JIT manufacturing Table A.3: Descriptive Statistics for 169 JIT Adopters Distribution of JIT Adoption Years A.4 Year Number of Firms Distribution 1982 3 1.77% 1983 5 2.96% 1984 11 6.51% 1985 13 7.69% 1986 14 8.28% 1987 17 10.1% 1988 21 12.4% 1989 22 13.0% 1990 23 13.6% 1991 18 10.7% 1992 12 7.10% 1993 10 5.92% Total 169 100% Work-in-process Inventory and JIT manufacturing In the main body of the paper, I assume that ﬁrms purchase both material and work-in-process inventory from suppliers and deﬁning JIT as JITpurchasing. In the background, I have in mind that ﬁrms also use JITmanufacturing. This section shows that as ﬁrms adopt JIT-manufacturing to improve production eﬃciency, ﬁrms shift resources from work-in-process inventory to material inventory. As such, once ﬁrms adopt JIT (both purchasing and manufacturing), all input inventory — material inventory and work-in-process inventory — will be converted into cash. JIT-manufacturing is deﬁned as eﬃcient production. Prior to its implementation, production takes time. Work-in-process inventory is generated in the process of transforming materials into ﬁnished goods due to waiting. 124 A.4. Work-in-process Inventory and JIT manufacturing Table A.4: Descriptive Statistics for 169 JIT Adopters (continued ) Distribution of Two-Digit Industry Classiﬁcations 2-Digit SIC Code Industry 20 22 23 24 25 26 27 28 30 31 33 34 35 36 37 38 39 Food Textile mill product Apparel Lumber Furniture Paper Printing, publishing Chemicals Rubber and plastics Leather Primary metals Fabricated metals Industrial equipment Electronic equipment Motor vehicles Instrumentation Other manufacturing Total Number of Firms Distribution 1 2 1 1 7 4 4 4 4 2 9 8 37 40 18 22 5 0.59% 1.18% 0.59% 0.59% 4.14% 2.37% 2.37% 2.37% 2.37% 1.18% 5.32% 4.73% 21.9% 23.7% 10.7% 13.0% 2.96% 169 100% That is, with JIC-manufacturing, ﬁrms transform materials into work-inprocess goods which however are not ready for being processed into ﬁnal products in current period. As ﬁrms adopt JIT, they operate in such an eﬃcient way that they can produce products with raw materials in time to satisfy customers’ needs. I model JIC-manufacturing and JIT-manufacturing as follows. Firms use linear technology, G1 (N ) = N , to transform materials into work-inprocess products which are then used to produce ﬁnal goods with technology G2 (N ) = N α with 0 < α < 1. Production process is ineﬃcient in JIC environment. Newly-generated work-in-process products from materials cannot be converted into ﬁnal goods. To smooth operation, ﬁrms hold both material inventory s1 and work-in-process inventory s2 as working capital. Both depreciate at the same rate δs . JIT-manufacturing shortens production time and makes newly-generated work-in-process inventory available for currentperiod ﬁnal-good production. Similar to the model presented in Section 4, here I assume away uncertainty and ﬁxed inventory adjustment costs. To further simplify the model, I assume that there are no ﬁnancial frictions, capital or cash, and that ﬁrms use JIC-purchasing in both manufacturing environments. Without inven125 A.4. Work-in-process Inventory and JIT manufacturing tory adjustment costs, a ﬁrm adjusts its material inventory holdings each period. The ﬁrm’s problem is to maximize the expected value of the discounted future dividend stream by choosing how many materials to purchase is , and how many materials N1 and work-in-process products N2 to use in production, given the beginning of period material and work-in-process inventory stocks, s1 and s2 . A.4.1 Steady State under JIC-manufacturing In the JIC-manufacturing environment, production ineﬃciency causes the unavailability of the newly-generated half-ﬁnished goods for producing current period ﬁnished goods. I write the ﬁrm’s problem in the JIC environment as follows, V (s1 , s2 ) = max is >0,N1 ,N2 {N2α − is + βV (s′1 , s′2 )}, where s1 ≥ N1 , s2 ≥ N2 , s′1 = (1 − δs )(s1 − N1 + is ), s′2 = (1 − δs )(s2 − N2 + N1 ). Rewriting the problem with multipliers, the ﬁrst order conditions of material use, work-in-process goods use, new purchases, material inventory holdings and work-in-process inventory holdings are given by: V (s1 , s2 ) = max is >0,N1 ,N2 {N2α −is +βV ((1−δs )(s1 −N1 +is ), (1−δs )(s2 −N2 +N1 )) + µ1 (s1 − N1 ) + µ2 (s2 − N2 )}, ′ ′ N1 : β(1 − δs )β ∂V + µ2 = β(1 − δs ) ∂V + µ1 , ∂s′ ∂s′ 2 1 ′ N2 : αN2α−1 = β(1 − δs ) ∂V + µ2 , ∂s′ 2 ′ is : 1 = β(1 − δs ) ∂V , ∂s′ 1 s1 : ∂V ∂s1 ′ = β(1 − δs ) ∂V + µ1 , ∂s′ 1 126 A.4. Work-in-process Inventory and JIT manufacturing s2 : ∂V ∂s2 ′ = β(1 − δs ) ∂V + µ2 . ∂s′ 2 Solving the system of equations above at the steady state gives the steady state material inventory level and work-in-process inventory level: 1 [β 2 (1−δs )2 α] 1−α 1−δs s∗1 = , 1 s∗2 = [β 2 (1 − δs )2 α] 1−α . A.4.2 Steady State under JIT-manufacturing In the JIT-manufacturing environment, eﬃcient internal operations make direct conversion from materials into ﬁnished goods feasible. The ﬁrm’s problem in the JIT environment is therefore written as V (s1 , s2 ) = max is >0,N1 ,N2 where {N2α − is + βV (s′1 , s′2 )}, s′1 ≥ N1 , s′2 + N1 ≥ N2 , s′1 = (1 − δs )(s1 − N1 + is ), s′2 = (1 − δs )(s2 − N2 + N1 ). The Bellman equation with multipliers can be formulated as V (s1 , s2 ) = max is >0,N1 ,N2 {N2α −is +βV ((1−δs )(s1 −N1 +is ), (1−δs )(s2 −N2 +N1 )) + µ1 (s1 − N1 ) + µ2 (s2 + N1 − N2 )}, ′ ′ 2 1 = β(1 − δs ) ∂V + µ1 , N1 : β(1 − δs )β ∂V ∂s′ ∂s′ ′ N2 : αN2α−1 = β(1 − δs ) ∂V + µ2 , ∂s′ 2 ′ is : 1 = β(1 − δs ) ∂V , ∂s′ 1 ′ s1 : ∂V ∂s1 = β(1 − δs ) ∂V + µ1 , ∂s′ s2 : ∂V ∂s2 = β(1 − δs ) ∂V + µ2 . ∂s′ 1 ′ 2 127 A.4. Work-in-process Inventory and JIT manufacturing Again, I can derive the equilibrium material inventory and work-in-process 1 inventory: s∗1 = [β(1 − δs )α] 1−α and s∗2 = 0. A.4.3 Discussion I ﬁrst show analytically that the optimal level of material inventory holdings under JIT is greater than the optimal level under JIC, illustrated in Proposition 8. I then parameterize the model to quantitatively measure the magnitude of the rise in material inventory as ﬁrms implement JITmanufacturing. Proposition 8 As ﬁrms switch from JIC-manufacturing to JIT manufacturing, they increase their material inventory holdings, s∗1,JIT > s∗1,JIC . Proof. s∗1,JIT − s∗1,JIC = [β(1 − δs )α] 1−α − β 1−α (1 − δs ) 1−α −1 α 1−α 1 1 2 2 1 α 1 = [β(1 − δs )α] 1−α [1 − β 1−α (1 − δs ) 1−α ] 1 1+α 1+α > [β(1 − δs )α] 1−α [1 − max{β 1−α , (1 − δs ) 1−α }] > 0, The last equality follows from the non-negativity of returns to scale α > 0 1+α and therefore max{β, 1 − δs } 1−α < 1. Intuitively, this result arises from production eﬃciency. JIT manufacturing reduces production costs by eliminating work-in-process inventory. The improved productivity leads ﬁrm to invest more resources in production by using more materials. In the absence of JIT-purchasing, ﬁrms hoard materials to smooth production. Figure A.2 plots the ratio of optimal material inventory holdings under JIT over the sum of optimal material and work-in-process inventory holdings under JIC as a function of inventory depreciation rate δs , given the parameter values used in the main body of the paper β = 0.96 and α = 0.55. As the inventory depreciation rate rises from 20% to 40%, the ratio increases from 80% to 127%. When δs is 0.32, the value calibrated in the paper, the ratio equals to 1.04. That is, as ﬁrms switch from JIC-manufacturing to JITmanufacturing, ﬁrms allocate approximately the same amount of resources from work-in-process inventory to material inventory. 128 A.4. Work-in-process Inventory and JIT manufacturing 1.4 s∗1,J I T ,β = 0.96, α = 0.55 s∗1,J I C + s∗2,J I C Relative Material Inventory Holdings 1.3 1.2 1.1 1 0.9 0.8 0.7 0.2 0.25 0.3 δs 0.35 0.4 Figure A.2: Relative Material Inventory Holdings after the Adoption of JIT manufacturing. This ﬁgure plots the model-implied ratio of the steady-state material inventory holdings under JIT-manufacturing to the total input inventory holdings (both material and work-in-progress) under JIC-manufacturing. It illustrates that ﬁrms shift resources from work-in-progress inventory to material inventory when switching to JIT. With reasonable parameter values on inventory holding costs δs , the steady-state material inventory holdings under JIT range from 80% to 127% of the steady-state pre-adoption input inventory holdings. 129 Appendix B B.1 Proof of Proposition 5 ˆ < 0, θ¯ > 1 Proof. (1) and (2): Under the condition x1 < ln(1 − δ) − ln R x1 ∗ ¯ exists and equals to ˆ . It follows that if θ ≥ θ, s drops below zero, ln(1−δ)−ln R which contradicts with the non-negativity condition of inventory holdings. ¯ s∗ = 0. If θ < θ, ¯ s∗ > 0 exists. Thus if θ ≥ θ, (3): Under the condition θ < θ¯ when an interior solution exists, from equation (8) we can evaluate the derivative of optimal inventory holdings s∗ with respective to θ: ∂s∗ 1 1 −2 −2 a −2 ex1 1 − δ θ 1 − δ = [( + ) a + (1 + ) ]( ) ln < 0. ˆ ˆ ∂θ a 1−δ 1−δ 1−δ R R This is negative, because all terms on the right side except for the last one are positive. Taking derivatives with respect to inventory holding costs δ and the ˆ , we have return on cash R ∂s∗ 1 1 −2 −2 1 − δ θ−1 = −( + ) [a θ( ) + (1 − δ)−2 ] ˆ ∂δ a 1−δ R − ex1 (1 + a −2 1 ) (1 − δ)θ−2 (θ − 1) < 0, ˆθ 1−δ R and ∂s∗ 1 1 −2 −2 (1 − δ)θ a −2 1 − δ θ−1 = −( + − ex1 θ(1 + ) < 0. ) a θ ) ( θ−1 ˆ ˆ ˆ a 1 − δ 1 − δ ∂R R R Under the condition ex1 + s∗ < 1 when an interior solution c∗ exists, we ˆ From equation can prove that optimal cash holdings increase in θ, δ and R. (2.7), 1 1 ∂s∗ ∂ ln(1 − c∗ ) ∂c∗ −2 x1 ∗ = θ ln(e + s ) + (1 − ) < 0, ∂c∗ ∂θ θ ex1 + s∗ ∂θ ∗ which follows ln(ex1 + s∗ ) < 0 and ∂s ∂θ < 0. Because the ﬁrm will never save cash more than the amount it needs in the future, ∂ ln(1 − c∗ ) 1 =− < 0, ∂c∗ 1 − c∗ 130 B.2. Proof of Proposition 6 which in turn implies that ∂c∗ > 0. ∂θ Similarly, we can derive ∂ ln(1 − c∗ ) ∂c∗ 1 1 ∂s∗ = (1 − ) < 0, ∂c∗ ∂δ θ ex1 + s∗ ∂δ and ∂s∗ ∂ ln(1 − c∗ ) ∂c∗ 1 1 = (1 − ) < 0, ˆ ˆ ∂c∗ θ ex1 + s∗ ∂ R ∂R which imply that ∂c∗ ∂c∗ > 0, > 0. ˆ ∂δ ∂R ˆ< (4): Also, given the expression of θ¯ and the condition x1 < ln(1 − δ) − ln R 0, we have ∂ θ¯ x1 ˆ −2 < 0, = [ln(1 − δ) − ln R] ∂δ 1−δ ∂ θ¯ x1 ˆ −2 < 0 = [ln(1 − δ) − ln R] ˆ ˆ ∂R R and B.2 ∂ θ¯ ˆ −1 < 0. = [ln(1 − δ) − ln R] ∂x1 Proof of Proposition 6 Proof. To assess the impact of risk on optimal cash holdings, I take the total diﬀerential of equation (2.8) with respect to c∗ and σ and derive the following equation 0= ˆ ˆ ln b − µ ∂b ∗ β Rλ β Rλ ln b − µ Φ′ ( ) ∗ dc − 2 Φ′ ( )(ln b − µ)dσ, σb σ ∂c σ σ where θ b = (1 − c∗ ) θ−1 − s∗ , ˆ 1 ∂b θR ˆ θ−1 < 0. = − (1 − c∗ R) ∗ ∂c θ−1 It follows that dc∗ (ln b − µ)b = . ∂b dσ σ ∂c ∗ 131 B.3. Proof of Proposition 7 The strict concavity of the ﬁrm’s problem, due to the costly external borrowing, ensures that the ﬁrm is risk averse and its certainty equivalent is smaller than the expected value, ln b < µ. Therefore, dc∗ > 0. dσ B.3 Proof of Proposition 7 Proof. Assuming the existence of an interior solution, I rewrite equation (2.3) as s∗ − 1 (1 − (B.1) ) θ = β(1 − δ)(G1 + G2 ), 1−δ where ∫ +∞ G1 = ∫ G2 = (ex + s∗ )− θ f (x)dx, 1 −∞ ln b λ(ex + s∗ )− θ f (x)dx, 1 −∞ θ b = (1 − c∗ ) θ−1 − s∗ , and f (x) is the probability density function of the random variable x which has a normal distribution, x ∼ N (µ, σ 2 ). I then take the total diﬀerential of the equation (B.1) with respect to s∗ and σ, [ s∗ − 1 −1 ∂G1 ∂G2 1 ∂G1 ∂G2 (1− ) θ −β(1−δ)( ∗ + ∗ )]ds∗ = [β(1−δ)( + )]dσ. θ(1 − δ) 1−δ ∂s ∂s ∂σ ∂σ The expressions for the derivatives of G1 and G2 with respect to s∗ and σ are ∫ +∞ 1 ∂G1 1 − (ex + s∗ )− θ −1 f (x)dx, = ∗ ∂s θ −∞ ∫ ln b 1 1 x ∂G2 ∂ ln b ln b ∗ − θ1 = λ[ (e + s ) f (ln b) − (e + s∗ )− θ −1 f (x)dx], ∗ ∗ ∂s ∂s −∞ θ ∫ +∞ 2 1 1 (x − µ) ∂G1 1 =− { [(ex + s∗ )− θ − (ex + s∗ )− θ ]f (x)dx}, ∂σ σ −∞ σ2 ∫ ln b 2 1 (x − µ) 1 ∂G2 λ =− { [(ex + s∗ )− θ − (ex + s∗ )− θ ]f (x)dx}. ∂σ σ −∞ σ2 132 B.3. Proof of Proposition 7 The derivative ∂G1 ∂s∗ is negative, because it is an integral over a negative θ ∂ ln b ∗ θ−1 − 2 function. The negativity of the derivative ∂G ∂s∗ follows ∂s∗ = −[(1−c ) ∂G1 1 s∗ ]−1 < 0. I next show that the derivatives ∂G ∂σ and ∂σ are both positive. Using the series representation of (u + v)w , w (u + v) = ∞ ∑ v n Cnw n=0 u 1+wn n , I approximate (ex + s∗ )− θ by 1 (ex + s∗ )− θ ≈ (ex )− θ + g(s∗ ). 1 1 Here, Cnw is the binomial coeﬃcient and g(s∗ ) > 0 is a function of s∗ . ∂G1 Deﬁne y = x−µ σ , the expression of ∂σ can be written as ∫ +∞ 1 1 2 1 1 ∂G1 [(eσy+µ )− θ + g(s∗ )] √ e− 2 y dy =− { ∂σ σ −∞ 2π ∫ +∞ 1 1 2 1 − [(eσy+µ )− θ + g(s∗ )]y 2 √ e− 2 y dy} 2π −∞ ∫ +∞ µ σ 1 2 1 1 =− [ e− θ y− θ (1 − y 2 ) √ e− 2 y dy σ −∞ 2π ∫ +∞ 1 2 1 (1 − y 2 ) √ e− 2 y dy] + g(s∗ ) 2π −∞ ∫ +∞ µ σ 1 2 1 1 =− [ e− θ y− θ (1 − y 2 ) √ e− 2 y dy] σ −∞ 2π √ σ 2 σ2 1 −µ 1 = − e θ √ [− 2π 2 e 2θ2 ] σ θ 2π 2 µ σ σ = 2 e 2θ2 − θ , θ which is positive, ∂G1 ∂σ > 0. 133 B.4. Variable Definitions Similarly, the derivative ∂G2 λ =− { ∂σ σ ∫ ln b−µ σ ∂G2 ∂σ σy+µ − θ1 [(e ) −∞ can be expressed as 1 2 1 + g(s )] √ e− 2 y dy − 2π ∗ ∫ ln b−µ σ [(eσy+µ )− θ 1 −∞ 1 2 1 + g(s∗ )]y 2 √ e− 2 y dy} 2π ∫ ln b−µ σ µ σ 1 2 λ 1 =− [ e− θ y− θ (1 − y 2 ) √ e− 2 y dy σ −∞ 2π ∫ ln b−µ σ 1 2 1 + g(s∗ ) (1 − y 2 ) √ e− 2 y dy] 2π −∞ ∫ ln b−µ 2 σ µ σ 1 2 λ 1 ln b − µ − (ln b−µ) 2σ 2 =− [ e ] e− θ y− θ (1 − y 2 ) √ e− 2 y dy + g(s∗ ) σ −∞ σ 2π ∫ ln b−µ σ µ σ 1 2 λ 1 >− [ e− θ y− θ (1 − y 2 ) √ e− 2 y dy], use g(s∗ ) > 0, ln b − µ < 0 σ −∞ 2π ln b−µ √ σ 2 σ2 σy y2 − σ + y ln b−µ λ −µ 1 σ σ σ = − e θ √ [− 2π 2 e 2θ2 erf ( θ√ )|−∞ − e− θ − 2 (− − y)|−∞ ], σ θ θ 2π 2 √ σ 2 σ2 −σ + y λ µ 1 = − e− θ √ {− 2π 2 e 2θ2 [erf ( θ√ )|y= ln b−µ − erf (−∞)] σ σ θ 2π 2 σy y2 σ − e− θ − 2 (− − y)|y= ln b−µ }, σ θ where erf is the Gauss error function. This derivative − σθ +y √ < 0 and erf ( following that y = ln b−µ σ ∂G2 ∂G1 1 Given that ∂s∗ < 0, ∂s∗ < 0, ∂G ∂σ > ds∗ = dσ B.4 2 1 θ(1−δ) (1 − s∗ − θ1 −1 1−δ ) is positive, )|y= ln b−µ − erf (−∞) > 0. 0 and 1 β(1 − δ)( ∂G ∂σ + ∂G2 ∂σ ∂G2 ∂σ σ > 0, ∂G2 ∂σ ) 1 − β(1 − δ)( ∂G ∂s∗ + ∂G2 ∂s∗ ) > 0. Variable Definitions Following [11], I construct the sample from Compustat and deﬁne variables used in the cash and inventory regressions as follows: Cash is deﬁned as the ratio of cash and short-term investments over total asset; 134 B.4. Variable Definitions Inventory is the ratio of total inventory over total asset; Markup is measured as sales over the sum of cost of goods sold and selling, general and administrative expenses; Market share (4-digit) is annual sales of the ﬁrm divided by totals sales in the 4-digit SIC code industry; Market share (3-digit) is annual sales of the ﬁrm divided by totals sales in the 3-digit SIC code industry; Sales growth is sales changes divided by average sales of this period and the previous period; Firm size is the natural logarithm of total asset; Risk is computed as the standard deviation of annual operating cash ﬂow in the past ﬁve years, with operating cash ﬂow deﬁned as earnings after interest, dividends and tax but before depreciation divided by total asset; Market-to-book ratio is the sum of market value and debt over total asset; Net working capital (cash) is equal to working capital net of cash over total asset; Net working capital (inventory) is working capital net of inventory over total asset; Capital investment is the ratio of capital expenditure over total asset; Leverage is the sum of long-term debt and debt in current liabilities normalized by total asset; R&D investment is research and development expenses to total asset ratio; Dividend is a dummy variable taking value of one if dividend payout (common) is non-zero; Acquisition is the ratio of acquisition over total asset. 135 Appendix C C.1 Data Sources and Calculations Following [47], we collect data from OECD Main Economic Indicator (MEI) and OECD Quarterly National Accounts (QNA) for the period 1973-2007 and construct variables using the deﬁnitions summarized in Table C.1. Table C.1: Data Sources and Calculations Variable The U.S. Output (y1 ) Consumption (c1 ) Investment (x1 ) Employment (n1 ) Real exchange rate (rx) Import price Export price Terms of trade (p) Net exports ratio (nx) Real imports ratio (ir) Rest of the World Output (y2 ) Consumption (c2 ) Investment (x2 ) Employment (n2 ) Definition Source Gross Domestic Product (at constant price 2000) Private plus Government Final Consumption Expenditure (at constant price 2000) Gross Fixed Capital Formation (at constant price 2000) Civilian Employment Index Price-adjusted Broad Dollar Index imports at current prices/imports at constant prices exports at current prices/exports at constant prices import price/export price (import-p*export)/y1 (all at current prices) import/(GDP-export) OECD MEI OECD MEI Aggregate of Canada, Japan and 19 European Counties (aggregate with PPP exchange rates in 2000) Aggregate of Canada, Japan and 19 European Counties (aggregate with PPP exchange rates in 2000) Aggregate of Canada, Japan and 19 European Counties (aggregate with PPP exchange rates in 2000) Aggregate of Canada, Japan and 8 European Counties (weighted with populations in 2000) OECD MEI OECD MEI Board of Governors OECD QNA OECD QNA OECD MEI OECD MEI OECD MEI OECD MEI However, since OECD no longer reports aggregate data series on GDP, consumption and investment for European 15 which [47] used to compute variables for the rest of the world and since consistent series for each of those 15 European counties are not available either, instead, we used 19 European countries, including Austria, Belgium,Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, Spain, Sweden and United Kingdom, Iceland, Luxembourg, Switzerland and Turkey. The employment series for the rest of the world, because of data unavailability, is computed as the weighted aggregate of Canada, Japan and 8 European countries (Austria, Finland, Germany, Italy, Norway, Spain, Sweden and UK). These diﬀerences in the sample may be contribute to the diﬀerences between the estimates of productivity shock process in [47] and 136 C.2. Estimation Details in this paper. C.2 Estimation Details In this section, we describe our estimation procedure, and show how it corresponds with the asymptotic linearization in (3.7) and (3.8). First, note that in our case, θ = β = (ρ11 , ρ12 , σe1 , σe1 e2 )′ . The parameters are estimated using the following estimating equations: ) ) ( )( ) ( ) ( ( ρ11 ρ12 z1,t−1 ε1,t µ1 z1,t + , + = (C.1) z2,t−1 ε2,t ρ12 ρ11 z2,t µ2 √ (C.2) σe1 = Eε21,t , Eε1,t ε2,t σe1 e2 = √ . Eε21,t Eε22,t (C.3) Deﬁne y1,t = z1,t and y2,t = z2,t for t = 2, . . . , n, and let Y1 and Y2 denote the corresponding (n − 1)-vectors of observations. Let Xt = (1, z1,t−1 , z2,t−1 )′ for t = 2, . . . , n, and let X denote the corresponding (n − 1) × 3 matrix of observations. Let ε1 and ε2 denote the (n − 1)-vectors of observations on the error terms. We have the following SUR system: Y∗ = (I2 ⊗ X) γ∗ + ε∗ , where Y∗ = (Y1′ , Y2′ )′ , ε∗ = (ε′1 , ε′2 )′ , and γ∗ = (µ1 , ρ11 , ρ12 , µ2 , ρ12 , ρ11 )′ , and note that γ∗ is restricted by Rγ∗ = 02×1 , where ( ) 0 1 0 0 0 −1 R= . 0 0 1 0 −1 0 Deﬁne Σ as the variance-covariance matrix of (ε1 , ε2 )′ : ( ) σe1 σ e1 e2 σe1 σe2 Σ= , σe1 e2 σe1 σe2 σe2 ˆ denote its consistent estimator. For example, Σ ˆ can be constructed and let Σ using the residuals obtained from OLS equation-by-equation estimation of (C.1). The restricted (FGLS) eﬃcient SUR estimator of γ∗ is given by: ( ) ( ( ) )−1 ˆ −1 ⊗ (X ′−1 R′ R Σ ˆ −1 ⊗ (X ′−1 R′ γˆ∗ = γ˜∗ − Σ R˜ γ∗ , 137 C.2. Estimation Details where γ˜∗ denotes the unrestricted OLS equation-by-equation estimator of γ∗ .59 Let σ ˆe1 and σ ˆe1 e2 denote the estimators of σe1 and σe1 e2 respectively constructed by replacing the expectations in (C.2) and (C.3) with sample averages and ε’s with ﬁtted residuals from the SUR system above. We need additional notation to describe the linearization of the estimator of β in (3.8). Deﬁne: ( H = I6 − Σ −1 ( ⊗ plimn→∞ X ′X n )−1 ) ( ( R ′ R Σ −1 ( ⊗ plimn→∞ X ′X n )−1 ) )−1 R √ and let H2;3 denote the second and third rows of H. In this case, n(βˆ − β), √ ˆ B, and n(h 1 − h1 ) in (3.8) are given by the corresponding terms in the following expression: ) ( ρˆ11 − ρ11 H2;3 02×1 02×1 02×1 n √ ρˆ12 − ρ12 1 1 ∑ = 01×6 0 0 √ n 2σe1 σe1 e2 σe1 e2 σ ˆ e1 − σe1 n t=2 1 01×6 − 2σ − 2σ 2 2 σ σ e1 e2 e e 1 2 σ ˆ e1 e2 − σe1 e2 ( ) ε1,t ⊗ Xt ε2,t 2 2 + op (1), ε − σ e 1,t 1 2 2 ε2,t − σe2 ε1,t ε2,t − σe1 e2 σe1 σe2 where ρˆ11 and ρˆ12 denote the second and third elements of the eﬃcient SUR estimator γˆ∗ . To estimate Λ11 and B, one should replace the population parameters in the above expression with their sample counterparts and ε’s with ﬁtted residuals from SUR estimation. To estimate Λ22 and Λ12 , one can use a ˆ 2. linearization (similar to that of σ ˆe1 and σ ˆe1 e2 above) for h 59 Since the two equations have the same set of regressors, the unrestricted eﬃcient SUR estimator is the equation-by-equation OLS estimator. 138 ′ R, C.3. Derivation of the Asymptotic Variances Formulas in (3.10)-(3.12) C.3 Derivation of the Asymptotic Variances Formulas in (3.10)-(3.12) When H0 is true, S can be written as [( ] )′ ( ) ′ ˆ ˆ ˆ ˆ S = h2 − gˆ(β) Whˆ 2 h2 − gˆ(β) − (h2 − g(β)) Wh2 (h2 − g(β)) [( ] )′ ( ) ˆ 2 − fˆ(θ) ˆ Wˆ h ˆ 2 − fˆ(θ) ˆ − (h2 − f (θ))′ Wh (h2 − f (θ)) . − h h2 2 (C.4) Next, ( )′ ( ) ˆ 2 − gˆ(β) Wˆ h ˆ 2 − gˆ(β) − (h2 − g(β))′ Wh (h2 − g(β)) h 2 h2 ( )′ ( )( ) ˆ 2 − gˆ(β) ˆ 2 − gˆ(β) = h Whˆ 2 − Wh2 h ( ) ( )′ ˆ 2 − h2 ˆ 2 − gˆ(β) + h2 − g(β) Wh h + h 2 ( )′ ˆ 2 − gˆ(β) + h2 − g(β) Wh (ˆ − h 2 g (β) − g(β)) ( ) ˆ 2 ) − w(h2 ) = ((h2 − g(β)) ⊗ (h2 − g(β)))′ J w(h ( ) ˆ 2 − h2 + 2 (h2 − g(β))′ Wh2 h √ − 2 (h2 − g(β))′ Wh2 (ˆ g (β) − g(β)) + op (1/ n), (C.5) where the last equality holds by vec(ABC) = (C ′ ⊗ A) vec(B) (see [67], equation (5) on page 30), (3.9), and (3.14). With a similar expression for ˆ 2 ), we the second term in (C.4) and a ﬁrst-order Taylor expansion for w(h √ obtain that (C.5) multiplied by n is equal to ) ′ √ (ˆ ′ 2) ′ n h − h2 2Wh2 (f (θ) − g(β)) + ∂w(h ∂h J K(h2 , f (θ), g(β)) ) √n√ ( 2 R fˆ(θ) − f (θ)) 2Wh2 (h2 − f (θ)) √ nR√ ( −2Wh2 (h2 − g(β)) ˆ(β) − f (β) R R g (C.6) +op (1). Next, by a ﬁrst-order Taylor expansion, ( ( ( ( )) ′ ( )) ( ( ))′ ( )) ˆ 2 − gˆ βˆ Wˆ h ˆ 2 − gˆ βˆ = h ˆ 2 − gˆ β ˆ 2 − gˆ β h W h ˆ h2 h2 ( ( ))′ ) ∂g(β) ( ˆ ˆ 2 − gˆ β −2 h Whˆ 2 β − β ∂β ′ √ + op (1/ n). 139 C.4. Derivation of the Standard Error Formula in (3.16) By combining this result (and a similar expansion for model f ) with the results in (C.6) and (C.4), and using (3.7)-(3.8), we obtain: √ ′ ′ (θ) A′ ∂f∂θ Wh2 (h2 − f (θ)) − B ′ ∂g(β) ∂β Wh2 (h2 − g(β)) ∂w(h2 )′ ′ nS = 2 Wh2 (f (θ) − g(β)) + 0.5 ∂h J K(h2 , f (θ), g(β)) Wh2 (h2 − f (θ)) −Wh2 (h2 − g(β)) ) √ (ˆ n h1 − h1 ) √ (ˆ n h − h2 ) × √ n √ ( ˆ2 + op (1). R f (θ) − f (θ) R ) √n√ ( ˆ(β) − f (β) R R g ′ (C.7) The results in (3.10)-(3.12) now follow by (3.6) and (3.9). C.4 Derivation of the Standard Error Formula in (3.16) When H0 is true, one can write SM = k ( ∑ ˆ h ˆ 2 ; Wˆ ) − d(fˆj (θ), ˆ h ˆ 2 ; Wˆ ) − d(gj (β), h2 ; Wh ) d(ˆ gj (β), 2 h2 h2 j=1 +d(fj (θ), h2 ; Wh2 ). Assuming that the contribution of simulation uncertainty is negligible, it follows from (C.7) that √ nS M =2 k ∑ j=1 √ Qj n ( ˆ 1 − h1 h ˆ 2 − h2 h ) + op (1), where note that Qj is the same as the ﬁrst two row-blocks of the multiplication matrix appearing in (C.7). The result in (3.16) follows. 140
- Library Home /
- Search Collections /
- Open Collections /
- Browse Collections /
- UBC Theses and Dissertations /
- Essays on macroeconomics
Open Collections
UBC Theses and Dissertations
Featured Collection
UBC Theses and Dissertations
Essays on macroeconomics Gao, Xiaodan 2013
pdf
Page Metadata
Item Metadata
Title | Essays on macroeconomics |
Creator |
Gao, Xiaodan |
Publisher | University of British Columbia |
Date Issued | 2013 |
Description | This dissertation studies two important topics in macroeconomics. The first topic is on the corporate cash hoarding. The first two chapters analyze the cash-inventory tradeoff from two different but complementary perspectives and shed light on the causes of cash hoarding. The second topic is international business cycles. A new feature of capital market is introduced into a standard international business cycle model to account for the disconnect between theory and data. The first chapter proposes an explanation for the joint dynamics of cash and inventory -- the adoption of the Just-in-Time (JIT) system. I start by demonstrating the importance of JIT in shaping corporate cash. I then develop a dynamic stochastic model to analyze the mechanisms and quantify their impacts. In the model, both cash and inventory can serve as working capital. As firms switch over from the traditional operating system (Just-in-Case, JIC) to JIT, they allocate the resources freed up from inventory to cash, in order to ensure smooth transactions with suppliers. On average, this switchover accounts for 45% and 69% of the observed cash increase and inventory decline respectively. The second chapter provides a complementary explanation for the cash-inventory joint dynamics. It models inventory as a reversible store of liquidity and studies the tradeoff between cash and inventory when a firm manages its liquidity needs. I argue that two key determinants of a firm's resource allocation decision are its market power and its exposure to risk. In the model, firms with lower market power and firms operating in riskier environments rely more heavily on cash rather than inventory. Model implications are supported by data. The third chapter studies the role of limited asset market participation (LAMP) in explaining international business cycles. We show that when LAMP is introduced into an otherwise standard model of international business cycles, the performance of the model improves significantly, especially in matching cross-country correlations. To perform formal evaluation of the models we develop a novel statistical procedure that adapts the statistical framework of Vuong (1989) to DSGE models. Using this methodology, we show that the improvements brought out by LAMP are statistically significant. |
Genre |
Thesis/Dissertation |
Type |
Text |
Language | eng |
Date Available | 2013-06-27 |
Provider | Vancouver : University of British Columbia Library |
Rights | Attribution-NonCommercial-NoDerivatives 4.0 International |
DOI | 10.14288/1.0073923 |
URI | http://hdl.handle.net/2429/44610 |
Degree |
Doctor of Philosophy - PhD |
Program |
Economics |
Affiliation |
Arts, Faculty of Vancouver School of Economics |
Degree Grantor | University of British Columbia |
Graduation Date | 2013-11 |
Campus |
UBCV |
Scholarly Level | Graduate |
Rights URI | http://creativecommons.org/licenses/by-nc-nd/4.0/ |
Aggregated Source Repository | DSpace |
Download
- Media
- 24-ubc_2013_fall_gao_xiaodan.pdf [ 714.35kB ]
- Metadata
- JSON: 24-1.0073923.json
- JSON-LD: 24-1.0073923-ld.json
- RDF/XML (Pretty): 24-1.0073923-rdf.xml
- RDF/JSON: 24-1.0073923-rdf.json
- Turtle: 24-1.0073923-turtle.txt
- N-Triples: 24-1.0073923-rdf-ntriples.txt
- Original Record: 24-1.0073923-source.json
- Full Text
- 24-1.0073923-fulltext.txt
- Citation
- 24-1.0073923.ris
Full Text
Cite
Citation Scheme:
Usage Statistics
Share
Embed
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"
src="{[{embed.src}]}"
data-item="{[{embed.item}]}"
data-collection="{[{embed.collection}]}"
data-metadata="{[{embed.showMetadata}]}"
data-width="{[{embed.width}]}"
async >
</script>
</div>
Our image viewer uses the IIIF 2.0 standard.
To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0073923/manifest