publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made.The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Cover illustration: Lyroky / Alamy Stock Photo Printed on acid-free paper This Palgrave Macmillan imprint is published by Springer Nature The registered company is Springer International Publishing AG
This book is dedicated to Susan who has given me so much
Preface I left academic life in 1972, after getting my Ph.D. At that time large-scale econometric modeling of the economy was the rage; everyone thought it would be just a matter of time before we had “done enough science” to allow economists to discuss economics in the classroom, not in terms of the alphas and betas of theoretical models, but in terms of the real-world coefficients they represent. Economics would become the next branch of engineering, or so many thought. Much to my surprise, when I returned to academic life 25 years later things had not much progressed. Most economists were still using alphas and betas to describe how one variable affects another in economics. For lack of vigorous, concerted effort over those 25 years to pursue the hard numbers underlying the theories, and their statistical significance, economists were still just discussing theories with the best “numbers” we had – the abstract alphas and betas of pure theoretical discourse. Because we hadn’t disciplined our presentation of theories to those scientifically proven to work, even more theories abounded than was the case in 1972. Worse, the overriding emphasis in economic theory was not on “what works?”, but on “what’s new?”. My engineering students knew the difference. When I tried to describe macroeconomics as real science , and then described the coefficients that connect one variable to another in alphas and betas, instead of real numbers, they just snickered. “Yes, but what is the real relationship?” they would ask, meaning what are the real numbers? “And if you don’t have them, why do you call this science?” they would ask. Certainly in their engineering courses, where every equation describes what actually works, they were getting real numbers. This book attempts to meet that very standard by focusing on what works. It attempts to move forward the empirical efforts of Tinbergen, Goldberger, Klein, Eckstein, and Fair the past 80 years to determine what works. That is, the effort to convert economics from just theory to hard (by which I mean reliable) science. Doing so requires three things.
First, it requires that the postulates we test have some economic meaning, and not be just some collection of variables we are “running up the flagpole,” to see what happens. Second, it requires that the theory-based postulates we test are structured loosely enough so that the data determine what is real, i.e., the exact shape and content of the theory being tested. It is not for us to say a priori by how we structure the model we test, whether Keynes’ consumption function, whose principal determinant is current income, is correct, or whether Freidman’s, whose principal determinant is average income (permanent income) is correct. Third, it is not for us to claim some empirical result proves some theory is correct, simply because it explains some variation in the economy, in some time period, in some economic model. To be correct, it should explain most variance, in most or all time periods, in most or all models. This book tries to adhere to these three rules, we think successfully. To meet the first condition, its model is built around the theory that we found most consistent with the data. To meet the second, the shape (and inclusion) of each equation in the model is data-determined, e.g., there are no predetermined assumptions about what drives consumer or investment spending. Third, a large-scale econometric model is needed to capture all the sources of economic variation, and that’s what is used. Extensive robustness testing was used to prove that any initial statistical finding was real and not just some spurious artifact of the time period or particular model tested. I hope the reader will agree that the models developed in this book adhere to these rules for good engineering science.
SUNY, AlbanyJohn J. Heim
Acknowledgements Most of all, I am indebted to Nobel Laureate Robert Solow for providing review comments and suggestions on an earlier draft, as did David Colander and Ray Fair. They were a source of inspiration and without their involvement and support, especially Robert Solow’s, this book probably would not have been finished. I am also indebted to distinguished econometrician, Kajal Lahiri, for bringing me to SUNY Albany and providing a place where I could work on this book with a minimum of other distractions. He has provided a very supportive and intellectually stimulating atmosphere within which to work, and provided guidance on econometric issues through his careful review of an earlier draft. I would also be remiss if I did not mention the long line of earlier economists who toiled long and hard as both macroeconomists and econometricians to turn macroeconomics from philosophy into science. These economists include Jan Tinbergen, Lawrence Klein, Frank deLeeuw, Arthur Goldberger, and, more recently, Ray Fair. Fair has had the doubly difficult job of keeping the strongly scientific Cowles tradition alive during recent decades, when many economists turned to different, less scientific approaches. We owe him much. For similar reasons, we owe Greg Mankiw much. His 2006 article in the Journal of Economic Perspectives convinced many that the detour in the 1980s away from Cowles modeling and toward DSGE has proven unproductive, and helped resurrect interest in Cowles modeling again. Solow’s (2010) testimony to Congress reached the same conclusion about DSGE and helped in the same way. Nor could the book have been written without the strong support of my wife Sue. This book required 2 years full-time work, and before that, considerable part-time work. The problems to be resolved required endless long hours at work, and endlessly preoccupied my mind, even at home. Sue was always willing to make the sacrifices necessary to cope with all that. Finally, I must acknowledge the secretarial assistance provided by Annemarie Hebert. She has helped pull together, duplicate, and send out endless drafts of this work.
Summary The book has two parts: Part I contains 45 equations describing in detail the “product side” of the National Income and Product Accounts (NIPA). It contains tested models of the GDP and its major components, and the determinants of their level of production (Chapters 4 – 19 ). Part II provides 11 additional equations describing how the value of the product generated producing the GDP is distributed among the factors of production. For each factor of production there are two equations. The first describes the variables that were found to determine each factor’s percentage share of national income. The second describes the variables found to determine the total amount (the level ) of each factor’s total income. These models describe the variables whose own changes cause the distribution of income among factors to shift from one factor to another over time ( Chapter 20 ). Chapter 19 provides a summary of the substantive findings as to the determinants of GDP and its components. Chapter 20 , Section 20.5 , summarizes the determinants of factor shares and levels of income.
The Production Side Model Production is treated as a response to aggregate demand (AD). Hence the key determinants of GDP production are expressed as determinants of AD. Supply shortages can also affect the level of production, but the empirical evidence indicates that demand is far more commonly the driving factor. Fully 85–95% of the variation of GDP over the 50-year period 1960–2010 appears to stem from variation in AD. Demand-driven models are commonly thought of as Keynesian models, and to that extent this is a Keynesian model. However, when a variable to measure “crowd out” is added to standard Keynesian consumption and investment equations, this model’s conclusions about the effectiveness of fiscal policy in stimulating the economy are just the opposite of Keynes’. Its conclusions about monetary policy conclusions are also not the same. The model indicates the stimulus effects of changes in the money supply to be modest at best. The 45-equation first part (the production side) includes 30 behavioral equations and 15 identities. The identities connect the behavioral equations into a comprehensive model of the real U.S. economy. The behavioral equations were generally estimated applying strong instrument 2SLS to 1960–2010 data. The model includes eight consumption and nine investment equations, including three for personal, corporate, and depreciation allowance savings. Two interest rate determination models based on the Taylor rule or the Keynesian LM curve are included. Also included are two unemployment determination models, a Phillips curve model, one export function, and two “IS” curve functions determining GDP. Other behavioral models are provided for taxes and government spending, recognizing that part of these variables levels is endogenously determined by the state of the economy. Two functions describe the determinants of M1 and M2 velocity. These are included to show mathematically how fiscal policy can shift the AD curve. Extensive efforts were made to ensure that all identification issues were resolved by replacing Hausman-endogenous variables with Waldstrong instruments which were Sargan-tested to ensure they also were not endogenously determined. There are 75 variables (or different lags of the same variables) in the 45 equations. Robustness testing, a non-negotiable requirement of good science, was exhaustive. All models were tested in four different time periods to ensure estimated effects were consistent over time, i.e., immune to Lucas critique. All coefficients were also tested for robustness to changes in the model being tested, i.e., to
see how additions and subtractions of variables from the model affected the remaining variables estimated effects. Because of the pervasiveness of the multicollinearity problem, this type of robustness testing is also a non-negotiable requirement of good science. Finally, almost all were tested using OLS as well as 2SLS techniques to allow comparisons with literature of an earlier day, which sometimes used OLS. DSGE and VAR methodologies are currently more popular methodologies for macroeconomic modeling. Therefore, a lengthy section is included in Chapter 2 discussing the advantages of the older Cowles methodology and why it is used here. Chapter 2 is literally a paper within a paper. It deals with what may be the most pressing unresolved methodological issue facing macroeconomic modelers today: how to successfully model the macroeconomy the way it actually works , so that models can be reliably used by policy makers to predict consequences of decision-making. Early models designed to do this were referred to as Cowles Commission models and were very good at explaining the data, though not always 100% successful. Cowles models dominated model building from the advent of the econometric revolution up to the mid-1980s. However, in the last 30 years, many economists have turned away from Cowles types of modeling in favor of DSGE and VAR. Which of these three methods for discerning economic reality is to be preferred? To shed some light on this question, the statistical performance of several VAR and DSGE models are compared with Cowles-type structural models. Comparisons are made, or reported from other studies, and include comparisons with a Sims (1980) VAR model, the Smets-Wouters model, FRB/US, and a simplified version of the FRB/NY model. These tests overwhelmingly indicate the more Keynesian (Cowles) structural models outperform the others in accurately modeling the actual year-to-year fluctuations of the economy. Therefore, they should become the models of choice in future macroeconomic studies analyzing the consequences of changes in economic variables. Nobel Laureate economist Robert Solow (2016) concurs; he has said Cowles models far better explain the data than DSGE or VAR models: after reviewing this paper’s analysis of the three methods, Solow wrote … Your arguments in favor of Cowles-type models as against VAR and DSGE models have real weight … I think that you get across that whatever can be said for DSGE models … they are inferior at explaining the facts … You do the same for general VAR models After Keynes himself, Solow is arguably the greatest economist of the twentieth century.
The Income Shares Model Part II of this book ( Chapter 20 ) describes how the income generated producing the GDP is distributed. Four equations describe the variables found to determine the level of income received as labor, profit rent, and interest income. An additional four equations describe the variables found to affect the percentage share of national income received by each of these factors, that causes factor shares to vary from decade to decade. A summary of findings is presented at the beginning of Chapter 20 . The econometric methodology used, including exhaustive robustness testing, was the same as used in Part I of the book.
Good science requires replicability of results. This chapter’s goal was to provide, to the best extent possible, models whose results meet the replicability standard. Largely, this goal appears to be achieved, though in some areas more remains to be done. Hopefully, future generations of researchers will find it worthwhile to take up where this study leaves off. In particular, in some equations we were not able to fully resolve the “left out” variables and multicollinearity problems that affects the credibility of parameter estimates in any economic model. In most models 85–95% of the variance is explained. However, in some models, there are definitely some “left out” explanatory variables remaining to be found. Less of the total variance in the model than we would like is explained by the variables. Models with this problem are identified in the text. In addition, the problem of multicollinearity needs to be better resolved. It is perhaps the most serious impediment to doing good science in economics today. To mitigate the problem in this study, we use first differencing, and careful selection of combinations of explanatory variables used. In addition, we do extensive robustness testing, by adding and subtracting explanatory variables to a model, to ensure (reasonable) model changes do not cause marked changes in other parameter estimates. For most of our parameter estimates we are able to show these techniques achieved the desired level of stability, but not for all. For some models, parameter estimates are still sensitive to exactly what other variables are included in the model (these models are identified in the text). Economists needs to develop better scientific methods for dealing with this problem.
Contents 1 Introduction 1.1 Modern Macroeconomics: Moving from the Methods of Economic Philosophy to Those of Economic Science 1.2 Summary of Ways in Which This Large-Scale Econometric Model Improves on Past Work 1.3 The 56-Equation Model: 30 Behavioral Equations, 15 Identities (Product Side of National Income and Product Accounts (NIPA)), and 8 Behavioral Equations, 3 Identities (Income Side of NIPA) 1.4 The 38 Behavioral Equations: Coefficients, Significance, R 2 , and Durbin Watson Tests: (Summary of Results: Detailed Explanations of Findings Presented in Chapters 4-20) Part I Production of the GDP 2 Methodology 2.1 General Methodological Issues 2.2 Choosing Between VAR, DSGE, and Cowles Commission Models 3 Literature Review 3.1 Lawrence Klein and Michael Evans (1968): The Wharton Econometric Forecasting Model 3.2 Otto Eckstein’s (1983) The DRI Model of the U.S. Economy 3.3 Ray Fair’s Estimating How the Macroeconomy Works (2004) 3.4 Federal Reserve Board/U.S. Model (1996) 3.5 Literature Review Summary 4 The Consumption Models 4.1 Total Consumer Spending on Both Domestically Produced and Imported Consumer Goods 4.2 Spending on Imported Consumer Goods – OLS Estimates
4.3 Spending on Imported Consumer Goods – 2SLS Estimates 4.4 Consumer Spending on Domestically Produced Consumer Goods (OLS) 4.5 Determinants of Consumer Borrowing – OLS Estimates 4.6 Determinants of Consumer Borrowing – 2SLS Estimates 4.7 Modeling the Major Components of Total Consumption 4.8 Determinants of Spending on Consumer Durables (OLS) 4.9 Determinants of Spending on Consumer Durables (2SLS) 4.10 Determinants of Spending on Consumer Nondurables (OLS) 4.11 Determinants of Spending on Consumer Nondurables (2SLS) 4.12 Determinants of Spending on Consumer Services (OLS) 4.13 Determinants of Spending on Consumer Services (2SLS) 5 Models Identifying the Determinants of Investment Spending and Borrowing 5.1 OLS Estimates of the Determinants of Total Investment Spending 5.2 2SLS Estimates of the Determinants of Total Investment 5.3 OLS Estimates of the Determinants of Domestically Produced Investment Goods 5.4 2SLS Estimates of the Determinants of Domestically Produced Investment Goods 5.5 OLS Estimates of the Determinants of Imported Investment Goods 5.6 2SLS Estimates of the Determinants of Imported Investment Goods 5.7 An Alternative Method of Calculating Coefficients in the Investment Imports Model 5.8 OLS Estimates of the Determinants of Investment Borrowing 5.9 Determinants of Spending on Fixed Plant and Equipment Investment (OLS) 5.10 Determinants of Spending on Fixed Plant and Equipment Investment (2SLS) 5.11 Determinants of Spending on Residential Investment (OLS) 5.12 Determinants of Spending on Residential Investment (2SLS)
5.13 Determinants of Spending on Inventory Investment (OLS) 6 The Exports Demand Equation 6.1 OLS Model of Export Demand 7 Statistically Estimated Real GDP Determination Functions (#x201C;IS” Curves) 7.1 The GDP as a Function of the Determinants of Domestically Produced Consumer and Investment Goods and Services, Government Spending and Exports (GDP = C D + I D + G + X) 7.2 The GDP as a Function of the Determinants of Total Consumer and Investment Goods and Services, Government Spending, and Exports Minus Imports (GDP = C T + I T + G + X – M) 8 Real GDP Determination Function (#x201C;IS#x201D; Curve) Coefficients Aggregated from Parameter Estimates Obtained by Statistically Estimating the Subcomponent Functions Comprising the GDP 8.1 Using the GDP Determination Model (GDP = C D + I D + G + X) 8.2 Using the GDP Determination Model (GDP = C T + I T + G + X – M) 9 Determinants of the Prime Interest Rate: Taylor Rule Method 9.1 OLS Estimates 10 Determinants of the Prime Interest Rate – LM Curve Method 10.1 OLS Models of the LM Curve 11 Determinants of Inflation – The Phillips Curve Model 11.1 Reconciling the Money Supply Variable in the Taylor Rule and LM Equation Interest Rate Models with the Money Supply Variable in the Inflation (Phillips Curve) Equation 12 Determinants of Unemployment 12.1 A Simple OLS Model Based on Okun’s Law 12.2 The 2SLS Okun Model 12.3 The OLS Technological Change Model 12.4 The 2SLS Technological Change Model
13 The Savings Functions 13.1 The Corporate Savings Function 13.2 The Depreciation Allowances Savings Function 13.3 Personal Savings 14 Determinants of Government Receipts 14.1 Contributions to Explained Variance 14.2 Robustness Over Time 14.3 Robustness to Model Specification Changes (1960–2010 Data Set) 15 Endogeneity of Government Spending Levels 15.1 The Model for Total Government Spending for All Purposes: Goods, Services, and Transfers 15.2 The Model for Government Spending on Goods and Services Only 16 Capacity of the Model to Explain Behavior of the Macroeconomy Beyond the Period Used to Estimate the Model 16.1 Model #1 Treating All Determinants of C, I, and X as Exogenous 16.2 Model 2: Treating C, I, and X Model Determinants for Which We Have Explanatory Functions as Endogenous 17 Converting the Older Keynesian IS-LM Model to the More Modern AS-AD Interpretation of the Keynesian Model 17.1 Short– and Long-Run Aggregate Supply Curves 17.2 The Aggregate Demand Curve and the Role of Velocity In Aggregate Demand 17.3 OLS Tests of M1 Velocity’s Determinants 17.5 OLS Tests of M2 Velocity’s Determinants 17.6 Which Determinants of GDP Are Also Determinants of Velocity 17.7 Stationarity Issues
17.8 Alternative Method: Calculating Impact of Determinants of GDP on Velocity Using Regression Coefficients Obtained Estimating Consumption, Investment, and Export Functions 18 Dynamics 18.1 Introduction 19 Summary and Conclusions (Production Side of the NIPA Accounts) 19.1 Other Major Findings Part II Income Side of the NIPA Accounts 20 Determinants of Factor Shares 20.1 Introduction, Theory of Factor Shares, and Summary of Findings 20.2 Literature on Factor Shares 20.3 Methodology 20.4 Determinants of Labor, Profits, Rent, and Interest Factor Shares and Income Levels 20.5 Summary and Conclusions (Income Side of the NIPA Accounts) Bibliography Index
List of Figures Fig 4.1.1 Actual consumption compared to levels calculated from Model 4.1.T 1960–2010
Graph. 6.1.1 Equation 6.1 Graphed
Graph. 12.2.1 The augmented Okun model (Eq. 12.4) model for explaining variation in unemployment 1960–2010
Graph. 12.4.1 Technological change model of determinants of unemployment (Eq.12.4.1)
Graph. 13.1.1 Fifty years annual variation in corporate saving (calculated from Eq. 13.1.1, then compared to actual)
Graph. 13.2.1 Explained and actual depreciation allowance savings the past 50 years
Graph. 13.3.1 The explanatory power of the Eq. 13.3.1 model
Graph. 17.4.1 Actual and fitted V1 values 1960–2010 (taken from Eq. 17.4.1.TR)
Graph. 17.5.1 Actual and fitted V2 values 1960–2010 (taken from Eq. 17.5.2.TR)
Graph. 18.104.22.168 MPK and MPL curves – constant slopes
Graph. 22.214.171.124 MPK and MPL curves – varying slopes
Graph. 126.96.36.199 MPK and MPL curves – non – market wages
Graph. 188.8.131.52 Model of only variables robust in at least three of four sample periods (Eq. 184.108.40.206.TR)
Graph. 220.127.116.11 Graph of the initial profit's share model (Eq. 18.104.22.168)
List of Tables Table.1.4.1 Determinants of consumption
Table.1.4.2 Determinants of investment
Table.1.4.3 Determinants of GDP (Cptr.8; arithmetically calculated from IS curve components)
Table.1.4.4 Is the prime interest rate determined by the Taylor rule?
Table.1.4.5 Is the prime interest rate determined by traditional Keynesian “LM” theory?
Table.1.4.6 Determinants of savings
Table.1.4.7 Determinants of government receipts and spending
Table.1.4.8 Determinants of unemployment and inflation
Table.1.4.9 Determinants of export demand
Table.1.4.10 Determinants of velocity robust models only (where V 1or2 = Y(P/M 1or2 )
Table.1.4.11 Determinants of labor's total income and percentage share of NI
Table.1.4.12 Determinants of profits' total income and percentage share of NI
Table.1.4.13 Determinants of rent's total income and percentage share of NI
Table.1.4.14 Determinants of interest total income and percentage share of NI
Table.22.214.171.124.1 DSGE model inflation forecast accuracy
Table.126.96.36.199.2 DSGE model GDP growth forecast accuracy
Table 188.8.131.52.1 (1) Current and four future year annual changes in income (real GDP) (Billions of 2005 Dollars)
Table 184.108.40.206.2 (1) Yearly variation in consumer spending 1960–2010. Explained by yearly variation in TFP compared to other determinants of consumption
Table 220.127.116.11.3 (1) Robustness over time: (2SLS detrended model; subsamples of 1960–2010 data set)
Table 18.104.22.168.3 (2) Robustness over time: (2SLS model 5.2, 1960–2010 data)
Table 22.214.171.124.4 (1) Forecasts of observable variables
Table 126.96.36.199.5 (1) Error of fit of a model similar to FRB/US'S nondurables and nonhousing services consumption model compared to Cowles model (yearly change in ND&S consumption
Table.188.8.131.52.1 Comparison of % error of GDP estimates of VAR with structural models for the 10 years after their 1960–2000 estimation period (absolute value of error % used)
Table.184.108.40.206.1 Time period robustness of SVAR model results
Table.220.127.116.11.2 Out–of–sample fit comparisons: Structural models vs. SVARs
Table.4.0.1 Determinants of consumption assumed endogenous when applying endogeneity tests
Table.4.0.2 Determinants of consumption or investment initially assumed exogenous or lagged, and
used as regressors in the first–stage regression in Hausman of endogeneity tests (subscripts denote lags)
Table.4.1.1 Explained variance – total consumption
Table.4.1.2 Robustness over time – (2SLS detrended model, Eq. 4.1.T)
Table.4.2.1 Explained variance – consumer imports
Table.4.2.2 Robustness over time – consumer imports
Table.4.4.1 Explained variance – domestically produced consumer goods
Table.4.4.2 Robustness over time – domestically produced consumer goods