1Michel De Vroey offers his reader a very pleasant journey through the History of Macroeconomics, from Keynes’ original insights, through the Friedmanian and Lucasian revolutions, to the latest developments. De Vroey has a true vision of macroeconomics, he shares it with his reader and gives clear guidelines to understand the developments in the field. This leads to a nonlinear—not necessarily chronological—but always exciting journey through the many ideas that paved the way to today’s macroeconomics. Interestingly, the book adopts a very macroeconomic view of macroeconomics. First, like modern macroeconomics, the book is dynamic. This history of macroeconomics is not only backward looking, it is, like modern macroeconomics, forward looking. De Vroey shows how each contribution logically followed, answered or echoed the previous one but also announced the next one. He also suggests what the future of macroeconomics will or should be in his own view. Second, like in most macro models, technological progress is given a key role in the long maturing process that led to current macroeconomics. De Vroey shows how the evolution of ideas in the field was not simply the outcome of quarrels regarding competing visions of the world—e.g. the role of competition, equilibrium, expectations, dynamics... but also the outcome of constant progress in neighboring sciences. For instance, the development of optimal control, dynamic programming, Kalman filtering, econometrics among others permitted/facilitated the emergence of dynamic models and rational expectations. They also drastically changed the way we evaluate our models and enhanced their falsifiability. The development of computers permitted the development of simulation/estimation techniques and promoted the development of new algorithms to solve heterogeneous agent models or models featuring strong non-linearities. In that sense the book is as much a history of ideas in macroeconomics as a history of methodology. From the point of view of the applied macroeconomist I am, this book puts a lot of structure on our understanding of the evolution of macroeconomics over the last 80 years, therefore helping us to understand where we are actually standing and heading to.
2Throughout the book, De Vroey highlights the prominent role played by Robert E. Lucas in the developments of macroeconomics in the last 40 years, not only because he introduced the concept of rational expectations, but even more importantly for macroeconomists because he brought discipline into macroeconomics. In the sequel, I will try to give my (hopefully) honest point of view on this statement as an applied macroeconomist. In his “Remarks on the influence of Edward Prescott”, Lucas (2007) argued that
Macroeconomics is just a lot more interesting today than it was 40 years ago. The level of theoretical modeling is higher, and theory contributes more centrally to empirical discussion. A much wider range of evidence is brought to bear on quantitative questions in macroeconomics. (Lucas, 2007, 10)
3This is certainly true. Macroeconomics tackles a very large spectrum of questions ranging from the standard growth and business cycle theories, to fiscal and monetary economics, labor market regulations, health and pension systems, redistribution, regulation of financial systems, understanding trade and the determination of exchange rates… The tools and methodology improvements that macroeconomists have witnessed and developed over the last 40 years led to a tremendous change in the way questions are tackled. The level of scientific requirements has increased, with a particular focus on the theoretical foundations of the models. In particular, that models be micro-founded is now a clear and not to be discussed requirement. The time dimension is taken very seriously: models are dynamic (backward-forward looking) and expectations, which play a critical role in modern macroeconomics, receive particular attention.
4This process was accompanied by a greater mathematization of the field, which has borrowed many of the tools from engineering (Kalman filtering, optimal control, dynamic analysis, …) and statistics (Time series econometrics, machine learning, ...). Some may regret it as it may sometimes (wrongly) be viewed as a way to tighten the hands of the macroeconomists, and may even prevent them from tackling some problems. However, the mathematical language has had its own merits. First of all, it has forced macroeconomists to put more structure on their discourse. Second, because the mathematical language—while being extremely powerful—is somewhat limited, it has forced macroeconomists to narrow down, better focus and qualify the questions they address. Third, and most importantly, it led to impose much more discipline in model evaluation. As a consequence, macroeconomics has become, to a large extent, a quantitative discipline. Even though some areas of macroeconomics have remained more theoretical and mainly deliver qualitative insights on some broad questions, there is now a clear willingness to assess quantitatively the predictions of the models we develop. Models are commonly taken to the data, and their predictions are checked using various econometric and statistical tools. In that sense, macroeconomics attempts to satisfy the falsifiability principle that characterizes hard sciences. All these requirements—micro-foundations, modeling of expectations, dynamics, confrontation to the data—defines, in a nutshell, the methodology laid down in Lucas’ (1980) project
to propose fully articulated, artificial economic systems that can serve as laboratories in which policies that would be prohibitively expensive to experiment within actual economies can be tested out at much lower cost. (Lucas, 1980, 696)
5which underlines most of Business Cycle Theory that De Vroey explores. The first typical examples of models that fulfills these requirements are given by the Real Business Cycle (RBC) model proposed by Kydland and Prescott (1982) and King et al. (1988). Interestingly, 35 years later, the so-called Dynamic Stochastic General Equilibrium (DSGE) models, that are now commonly used in Central Banks to understand and predict the effects of monetary policy decisions, are still built in reference to the Lucas’ project. Notwithstanding the fundamentally different message that these models deliver, the internal structure of these models is very similar to that of the original RBC models. This is exemplified by the following stylized 3 equations-DSGE model
6where equations (1)-(3) denote, respectively, the IS curve, the New-Keynesian Phillips (NKP) curve and the monetary policy rule. Although simple, this stylized model captures the essence of most DSGE models and appears to contain all ingredients of the Lucasian project. The two fundamental behavioral equations, IS and NKP curves, are given explicit micro-foundations. The IS curve corresponds to the aggregation of individual optimal consumption saving arbitrage derived from the maximization of expected lifetime utility subject to the intertemporal budget constraint of the agent. Likewise, the NKP curve corresponds to the aggregation of optimal price setting behavior, as obtained from profit maximization when the firms only face a constant probability of resetting their price in a given period. Both equations are dynamic, featuring both a backward and a forward looking component. The model being solved under the rational expectations hypothesis, expectations are solved consistently with the model. Finally the model is subject to stochastic shocks that initiate the business cycle. The model therefore contains all the required ingredients that make a model achieve the Lucas’ project. Macroeconomics, in that sense, has reached a mature state. Is the picture all that rosy?
7Before answering this question it is worth reminding one of the reasons for the Lucas’ quest for discipline: the so-called Lucas’ (1976) critique which states that the estimated coefficients of the behavioral equations in macroeconomic models in the Keynesian tradition are not invariant to changes in economic policy. In other words, these parameters are not deep, and using these models for policy evaluation purposes can lead to spurious conclusions. Micro-founded models do not suffer this critique as the behavioral equations then depend on deep parameters which are invariant to economic policy, while the dependence of the reduced form coefficients to both deep and policy parameters is well understood. So what about our three equations model? Let us focus on the NKP curve. Seen through the lenses of the previous statement, it satisfies the requirement. The price setting behavior is derived from an explicit profit maximization problem at the firm level, and the so obtained price setting equation depends, in a non-linear way, on the psychological discount factor, β, the probability of resetting the price, α, and the degree of indexation of current inflation to past inflation, γ. All these parameters being specified at the micro level, they are deep parameters and the model is robust to the Lucas critique. The problem with this statement lies in the exact status of the “deep” parameters, more precisely α and γ. There is indeed no reason, a priori, to think that the probability that a firm resets its price schedule is (i) constant and (ii) invariant to monetary policy. Likewise, the degree of price indexation is unlikely to remain invariant to monetary policy. In other words the model is not robust to the Lucas critique. What about rationality? The mere presence of price indexation (the fact that firms that do not reset their price optimally index the current price on past inflation) goes against rationality in this model: Where is the rationality in this decision? Why should firms adopt this rule of thumb? More generally, this last observation raises a more general question: why should firms accept nominal price contracts? Interestingly, the same question arose in the late 1970’s. The absence of a satisfactory answer then led to the emergence of the New classical macroeconomics, and the resurgence of equilibrium/flexible price models. While the modern DSGE models derive the Phillips curve from an explicit optimization problem, it is not clear that they are doing any better than models à la Gray (1976) or Taylor (1980) in terms of micro-foundations of nominal rigidities. We are then back to the seventies.
8Does it mean that not much progress has been made? The answer is negative. Many improvements have been (and still are) introduced over the recent years, addressing some of the concerns and critics that were raised against DSGE models. A first common critique addressed to these models is the use of the representative agent assumption (see for example Kirman, 2010). At first glance, the representative agent assumption appears to be inconsistent with a serious treatment of micro-foundations. It however alleviates a very tricky technical difficulty. Indeed, in a dynamic model, agents heterogeneity introduces a complication: one has to keep track of the dynamics of the wealth distribution across time. The difficulty is that this distribution is an infinite dimension object which a computer cannot deal with; this is the so-called curse of dimensionality problem. By making the representative agent assumption, one obviously bypasses this problem. Initially, the RBC literature therefore relied on this assumption to keep models tractable and be able to simulate them on the computers available at that time. Since, much progress has been made. Krusell and Smith (1998) proposed a model featuring agents that can either be employed or not on the labor market. It is however assumed that these agents cannot fully insured against the idiosyncratic employment risk which therefore preserves all heterogeneity and breaks away from the representative agent assumption. Krusell and Smith proposed a method to approximate numerically the solution of such models and broke the curse of dimensionality. Interestingly they showed that allowing for heterogeneity does not affect the business cycle properties of aggregate variables. In other words, as long as the focus is put on the macro-dimension, bringing heterogeneity in the standard model is not critical. Once the focus is shifted towards redistributive policies or models where heterogeneity affect aggregate level (typically models featuring financial frictions), heterogeneity shall be kept. Krusell and Smith paved the way to a branch of business cycle theory featuring explicit heterogeneity. This development was mainly permitted by technological progress: increase in computation power and the development of new algorithms to approximate distributions (see the special issues of the Journal of Economic Dynamics and Control, 2010 and 2011). This approach was recently applied by Gornemann et al. (2014) to build a New-Keynesian DSGE model featuring heterogeneous agents, where households differ in terms of wealth, income and employment status, to analyze the redistributive effects of monetary policy. While this has proven fruitful, this approach is also very computationally expensive. Recently, Challe and Ragot (2011) proposed a modeling framework that preserves a limited heterogeneity and allows for tractability. Using a similar approach, Challe et al. (2015) estimate a full-fledged New Keynesian DSGE model by Bayesian maximum likelihood. Departing from the representative agent hypothesis is therefore not a problem anymore.
9A second critique which is often addressed to this literature is related to the “implausible” cognitive abilities that the rational expectations assumption assumes for the agents. But, agents actually do not need to be fully informed to take decisions. This was, for example, already the case in the Lucas’ (1972) seminal islands model, where agents were already facing a static signal extraction problem. The implicit information problem was further extended to a full dynamic setting by Collard et al. (2009), Collard and Dellas (2010) among others. One would object that, given the signal extraction problems are solved relying on the Kalman filter, this still assumes a lot of cognitive ability on the part of the agents. This problem was addressed in the Rational Inattention literature developed by Sims (2010), which explicitly takes into account the limited cognitive abilities of agents through an entropy constraint (see Œconomia, 2015). Interestingly, this leads to an endogenous determination of the information set used by the agents to formulate their expectations. Mackowiak and Wiederholt (2009) apply this approach to a price setting model and show that this can explain the presence of price rigidities. While the preceding literature essentially focused on the observability of the state of the economy, it is as likely possible to take into account that fact that the agents are uncertain about the model. This is studied by Hansen and Sargent (2007) who adapt robust control techniques to study situations where decision makers acknowledge misspecification in economic modeling. Their approach proved useful to explain some observed cautious behavior in policy design or solve macro-finance problems (equity premium among others). More recently, Angeletos and La’O. (2013) offer a new theory of fluctuations that can accommodate the Keynesian notion of animal spirits in unique equilibrium, rational expectations models. This is the result of limited communication across agents, obtained by allowing trading to be random and decentralized. In that setting, the business cycle may be driven by a certain type of extrinsic shocks, the sentiment shock. Angeletos et al. (2015) consider a tractable version of the previous paper and show, in an estimated version of the model, that the sentiment shocks account for up to 70% of the volatility of the business cycle. Benhabib et al. (2015) and Nimark (2014) develop alternative models in which shocks capturing “animal spirits” play a non trivial role. They reach similar conclusions regarding the importance of these shocks. It therefore appears that the literature has gone a long way from the prototypical rational expectations model.
10During the last recession, rather severe critiques were raised against DSGEs, some even talked about a failure of macroeconomics. The purpose here is not to go back to this debate, but rather to show how macroeconomics reacted to these attacks. One commonly heard argument against the macroeconomic models were their inability to talk about the financial aspects of the business cycle. In that respect, the literature reacted quite fast and the number of contributions studying the impact of financial frictions on the business cycle sky-rocketted (see Brunnermeier et al. (2012), or Sneessens (2016) for surveys). The introduction of financial frictions proved fruitful to generate persistence and, when combined with illiquidity problems, non-linear amplification mechanisms. Models featuring endogenous risk and liquidity spirals explain the observed financial instability, and can explain why downturns can be deep. Even though such progresses were made, some still argued that macroeconomic models could not predict the emergence of a crisis. One reason for this failure is again technological. Most DSGE models are solved relying on local perturbation methods, which assumes differentiability of the solution, and cannot therefore generate crises. Indeed, most financial crises involve a credit crunch, which imply a discrete jump, and hence a non-differentiability in the evolution of credit that perturbation techniques cannot accommodate. However, the recent availability of global methods (see Judd, 1998) makes it now possible. For example, Boissay et al. (2016) develop a model in which banking crises result from the procyclicality of bank balance sheets that emanates from interbank market funding. During booms, bank market funding and credit supply increase, which pushes down the rates of return on corporate and interbank loans. The lower rates aggravate agency problems in the interbank market, which leads to a reduction in market funding and further pushes down the interest rate. When this process lasts long enough, the interbank market freezes, triggering a credit crunch and a deep financial crisis. What is key in the model is that the crisis emerges endogenously, and tracking credit helps predicting the next crisis. In particular, the larger the credit boom relative to the possibilities for productive use of loans, the larger the fall in interest rates, and the higher the probability of disastrous freeze of the interbank market.
11Just to repeat it, “Macroeconomics is just a lot more interesting today than it was 40 years ago” and Michel de Vroey’s book helps us better appreciate it and realize how, even though macroeconomic theory is making a lot of progress, it is standing on the shoulders of giants like Keynes, Leijonhufvud, Lucas, Sargent ... But even more importantly this makes us realize how much we still need to achieve. From that perspective, this book should actually be taught in any PhD course in macroeconomics, to have students realize how and why macroeconomics reached the particular state of development we are facing, and why, from a methodological point of view, we practice macroeconomics the way we do.