Main Content

Macroeconomic Modeling and Financial Stability: Lessons from the Crisis

The dynamic stochastic general equilibrium model (DSGE) marked a major milestone by capturing the dynamic change of economic variables over time. However, many DSGE models were exposed as having omitted critical structural linkages relevant to the financial crisis. To address these deficiencies, existing DSGE models should be enhanced to better incorporate the role of the financial sector and financial markets. In addition, these models should reexamine key micro-foundations of the model and consider behavioral components.

By Andrew W. Lo

Since the Financial Crisis of 2007–2008, macroeconomic modeling has come under fire for the spectacular failure of macroprudential policies to anticipate the impact of crisis on the real economy. Guided by highly sophisticated mathematical models known as dynamic stochastic general equilibrium (DSGE) models, central bankers and regulators had no idea that the financial-market tremors that began as early as 2005 would exact such an enormous toll on real output and employment just a few years later.

Part of the reason was, of course, the fact that the DSGE models used by central bankers did not contain a financial sector. From a macroeconomist’s perspective, financial markets are a sideshow, always operating flawlessly and without constraints to facilitate production, real investment, and economic growth. With the benefit of hindsight, we now understand that financial constraints can become extraordinarily important when market dislocation strikes. Apparently, financial stability cannot be taken for granted.

The reaction against DSGE models has been swift, with the most severe critics arising from inside the economics profession. In his testimony before the U.S. House of Representatives Committee on Science and Technology, MIT economist and Nobel Laureate Robert Solow leveled the following critique against this literature:1

“Especially when it comes to matters as important as macroeconomics, a mainstream economist like me insists that every proposition must pass the smell test: does this really make sense? I do not think that the currently popular DSGE models pass the smell test. They take it for granted that the whole economy can be thought about as if it were a single, consistent person or dynasty carrying out a rationally designed, long-term plan, occasionally disturbed by unexpected shocks, but adapting to them in a rational, consistent way. I do not think that this picture passes the smell test. The protagonists of this idea make a claim to respectability by asserting that it is founded on what we know about microeconomic behavior, but I think that this claim is generally phony. The advocates no doubt believe what they say, but they seem to have stopped sniffing or to have lost their sense of smell altogether.”

In a surprisingly frank and courageous mea culpa, Narayana Kocherlakota, President of the Federal Reserve Bank of Minneapolis, acknowledged the limitations of macroeconomic theory and policy:2

“I believe that during the last financial crisis, macroeconomists (and I include myself among them) failed the country, and indeed the world. In September 2008, central bankers were in desperate need of a playbook that offered a systematic plan of attack to deal with fast-evolving circumstances. Macroeconomics should have been able to provide that playbook. It could not. Of course, from a longer view, macroeconomists let policymakers down much earlier, because they did not provide policymakers with rules to avoid the circumstances that led to the global financial meltdown.”

These devastating indictments of a large swath of modern macroeconomics and policy is hard to square with all the Nobel prizes awarded to the architects of the influential DSGE edifice: Lucas, Kyland, Prescott, Sargent, Sims, and Hansen. Could it be that so many people were so easily fooled for so long, or has the pendulum swung too far the other way?

In this article, I hope to shed some light on this conundrum by tracing the origins of DSGE models and asking what lessons we have learned about macroeconomic modeling from the Financial Crisis. The superficially damning criticism of the DSGE framework belies the importance of the notion of general equilibrium and the Lucas critique to macroeconomic policy, and may divert attention from the more urgent task of developing better alternatives for regulators and policymakers. By examining the historical roots of DSGE models and studying their strengths and weaknesses from a financial-markets perspective, we can see a clearer path for building the next generation of macroeconomic policy models.

A Brief History of Macro Policy Models

Macroeconomic modeling for policy purposes is a relatively new endeavor. The Dutch economist Jan Tinbergen published the first empirical macroeconomic model of an economy in 1936, intended to forecast the economic effects of Dutch policy responses to the Great Depression. However, Tinbergen’s model is more than a historical footnote. Although his model was primitive by modern standards, it was the first in a lineage of macroeconomic forecasting models that are still used today.3

Tinbergen’s original approach was ad hoc, without a strong theoretical basis. Structural relationships between economic variables were eyeballed from linear regressions4 on minimal econometric data. Because the modern system of national accounts was not developed until after World War II, important macroeconomic variables were omitted from Tinbergen’s model. Nevertheless, Tinbergen’s work was striking enough to form the basis for the first generation of postwar macroeconomic models.

Following the war, the late Lawrence Klein, then a young economist with the Cowles Commission, combined Tinbergen’s framework with elements of John Maynard Keynes’ macroeconomic theory to model the prewar American economy.5 Klein’s approach had two great early successes. His first model successfully predicted that the end of the war would result in a boom, not a return to the Depression-era conditions feared by many policymakers. On the other hand, an updated model, which Klein developed with Arthur Goldberger, predicted that the Korean conflict would end in a recession, which also came to pass.6

These predictive successes, made against prevailing economic opinion, cemented Klein’s approach in the minds of economists and policymakers alike. Klein’s models were theoretically and statistically much more sophisticated than Tinbergen’s early efforts, and correspondingly more influential. In the late 1960s, Klein’s ambitious Brookings model inspired the Federal Reserve Board to develop its own macroeconomic model to use in forecasting and policy analysis.7 Franco Modigliani of MIT, Albert Ando of the University of Pennsylvania, and Frank de Leeuw of the Federal Reserve Board led the project to develop the MPS model (an acronym for MIT/Pennsylvania/SSRC, which funded Ando), which the Fed used from 1970 to 1995.

A hallmark of the Tinbergen-Klein school of modeling was the use of a multitude of empirically derived behavioral equations to specify relationships between macroeconomic variables. For example, the original MPS model had sixty behavioral equations, a third of them having to do with the housing and mortgage market. Unexpected macroeconomic events, such as the oil shocks of the 1970s, induced modelers to add highly specific equations to model previously unforeseen linkages, such as ones for domestic coal consumption, or dummy variables for automobile and dock strikes.8

However, one macroeconomic event of the 1970s struck at the heart of the modelers’ basic assumptions. Most of these models included the inverse relationship between unemployment and inflation known as the Phillips curve.9 Although this was an empirical result, it fit the Keynesian macroeconomic framework very well, and the Phillips curve quickly became part of the standard toolkit of the postwar generation of macroeconomists.

Unfortunately, the American economy during the 1970s demonstrated that Phillips curve was not an immutable natural law at all, but an apparent statistical fluke. The 1970s were a time of stagflation, stagnant economic growth and high inflation. Rising inflation did nothing to alleviate high unemployment in the 1970s. Central bankers who relied too much on expanding the money supply to stimulate the economy found themselves stimulating inflation instead.

Enter the Lucas Critique

The collapse of the Phillips curve, and the apparent failure of Keynesian macroeconomics to predict it, became an important battleground in the rational expectations revolution of the next decade. Robert Lucas was foremost among these intellectual combatants. His key insight, first published in 1976, was that economic modelers had no privileged insights into the functioning of the economy that the market did not already share. In fact, according to Lucas, the more successful an economic model was in the short run, the more likely it would fail in the long run. This stochastic drift in the quality of long-term prediction was the consequence of economic agents taking into account the changes suggested by the models. Standard econometric models failed in the long run precisely because people adapted to new economic conditions.10

This paradox became known as the Lucas critique, and it represented an existential threat to the entire postwar program of economic modeling. Any model that narrowly used historical macroeconomic data to evaluate policy was, according to Lucas, flawed at its foundation. Lucas was well aware that his conclusion was “destructive,” almost nihilistic, but he offered economists a way out from this dilemma. A macroeconomic model, to escape the Lucas critique, should consist of economic agents responding to policy changes or outside shocks according to their economic preferences. These agents would then adapt to change by following their economic self-interest. This type of model, Lucas believed, with strong microfoundations based on the preferences of the agents themselves, rather than on empirical relationships between macroeconomic variables, would be able to avoid the problems of his critique.

The Birth of DSGE

The DSGE model satisfied Lucas’s criteria. Within a forward-looking, rational expectations framework, a DSGE model was populated by agents that optimized their economic choices according to their preferences, in order to find a full equilibrium solution to the general (albeit simplified) economy. Rather than focusing on static snapshots of the economy, a DSGE model allowed economic variables to change dynamically over time in a logically consistent manner. A DSGE model could also respond to random stochastic shocks, such as fluctuations in prices like the oil shocks of the 1970s, changes in policymaking, or the adoption of new technologies by economic agents. In theory, this flexibility allowed a DSGE model to be used in previously unencountered economic conditions, without having to specify new relationships in the manner of earlier macroeconomic forecasting models.

DSGE models were originally associated with real business cycle (RBC) theory, which held that expansions and recessions in an economy were principally the result of outside shocks rather than changes in money supply or aggregate demand. In 1982, Finn Kydland and Edward C. Prescott found that a simple growth model, when fitted into a DSGE framework, could fit the business cycles of the postwar United States rather well. According to Kydland and Prescott’s model, technology shocks in the postwar period accounted for half the ‘action’ in U.S. business cycles. 11 For many macroeconomists, the success of Kydland-Prescott and its follow-ups validated RBC theory. Older schools of macroeconomic thought, such as classic Keynesianism or monetarism, were seen as fundamentally flawed since they could not be formulated in terms of microfoundations with optimizing agents.

These early DSGE models left little room for the traditional role of fiscal and monetary policy to guide an economy. In part, this represented the modelers’ tendency to shave with Occam’s razor. If the inclusion of money or a financial sector in a model was not necessary to describe the data, they shaved it off. For many, however, this modeling decision was due to skepticism about the role of a central bank or a government in a macroeconomy. If one believes the business cycle is caused by technological shocks, as in RBC theory, one will downplay the role of central bank intervention in mediating the business cycle. Nevertheless, an opposing school, the New Keynesians, began using the DSGE microfoundational framework while incorporating Keynesian concepts such as wage and price stickiness to model the macroeconomy, allowing greater room for the role of government stabilization in these models.

Academic macroeconomic modeling diverged from the models used by policymakers for a generation. For example, the Federal Reserve Board’s current model of the U.S. economy, called FRB/US, was adopted in 1995, but it is still of the highly parameterized Tinbergen-Klein lineage. Its developers conceded the importance of the Lucas critique, but they did not consider it foundational. As a result, FRB/US allows the modeler to use rational expectations, but as a secondary option in forecasting. One can direct an agent to use model-consistent future expectations, in the manner of a rational expectations model, or to limit and lag what knowledge is available to the agent.12

DSGE Goes Abroad

However, the continued success and increasing sophistication of DSGE models in academia caused policymakers to take a second look at their usefulness. In 2003, the European Central Bank (ECB) released its influential DSGE model of the eurozone, devised by Frank Smets of the ECB and Raf Wouters of the National Bank of Belgium.13 In its original form, Smets-Wouters operated at a very high level of abstraction. It did not model individual countries or regions or markets or industries within the eurozone. As in many earlier DSGE models, Smets-Wouters also did not incorporate an explicit role for money and financial markets. Rather, it included a “Taylor rule” to simulate the behavior of a central bank. Meanwhile, the role of the government in Smets-Wouters was quite simple: it had the power to tax and the power to adjust interest rates.

Despite these deficiencies, the Smets-Wouters model was considered a success. Smets-Wouters fit the historical macroeconomic data of the eurozone well, including gross domestic product, consumption, investment, wages, employment, and short-term interest rates. Moreover, it outperformed vector autoregression models on the same data. With a very small number of variables, Smets-Wouters managed to describe the macroeconomic behavior of one of the most complicated economic regions in the world—and it did so better than its competition.

The Smets-Wouters model would become a workhorse for the ECB. Smets-Wouters was simpler and apparently more powerful than the highly specified, almost baroque models of the Tinbergen-Klein tradition. It used the fundamental economic principle that agents would always try to optimize their behavior, subject to their constraints and preferences, to overcome the Lucas critique. Other central banks and international financial institutions quickly followed the ECB in adopting DSGE models for policy use: the International Monetary Fund and its Global Economic Model (GEM)14; the Bank of England Quarterly Model (BEQM)15; the Bank of Canada and its Terms of Trade Economic Model (ToTEM)16; and the Riksbank Aggregate Macromodel for Studies of the Economy of Sweden (RAMSES).17 Even the Federal Reserve Board of the United States, with its fantastically detailed, institutionally tested FRB/US model, commissioned its SIGMA model for use in policy analysis in 2006.18 By 2008, the macroeconomist Michael Woodford could tell his colleagues at the January annual meeting of the American Economics Association, “It is now accepted that macroeconomic models should be general equilibrium models.”19

DSGE Models and the Financial Crisis

Macroeconomic crises have consistently inspired economists to create new predictive models. The Great Depression motivated Jan Tinbergen to invent a new form of macroeconomic model from scratch, while World War II impelled Lawrence Klein to synthesize Tinbergen and Keynes to create a new school of macroeconomic modeling. The macroeconomic crises of the 1970s made Robert Lucas’s critique of those models urgent and compelling, while Lucas’s insights into rational expectations made the new paradigm of the DSGE model possible. Perhaps necessity is indeed the mother of invention.

Today we live in the aftermath of the worst global financial crisis since the Great Depression. No economic model predicted the crisis or its extent beforehand. However, DSGE models came under especially harsh criticism, since many models then in use were too weakly specified to be useful for policymakers in a financial crisis. The financial crisis was transmitted through structural linkages of a sort omitted as unnecessary by DSGE modelers. Some DSGE models omitted a financial sector entirely. While other DSGE models did incorporate financial frictions, the complex range and rich variety of financial instruments were reduced to one or two assets trading in a single market. The many possible roles of a monetary authority were usually simulated by a simple Taylor rule on the interest rate. The crisis had widely different effects among populations with different financial characteristics, such as wealth and income; however, some DSGE models avoided this heterogeneity as too difficult to calculate, in favor of unitary agents representing, e.g., all the households in the economy. 20

Another critique of DSGE models that emerged after the crisis regarded their lack of behavioral realism. Important economic phenomena, such as unemployment and asset bubbles, were harder to understand within a DSGE framework, which assumed optimizing agents with rational expectations. For example, unemployment can be included in a DSGE model, but the premises of the DSGE framework imply that the optimizing agent somehow prefers being unemployed. DSGE models rely on exogenous shocks to drive significant changes in the macroeconomy, but these shocks—to the technological frontier, to the depreciation rate, to the labor market’s willingness to work—made little contextual sense. To quote Narayana Kocherlakota, “Why should everyone want to work less in the fourth quarter of 2009? What exactly caused a widespread decline in technological efficiency in the 1930s?”21 However, in the aftermath of the financial crisis and its illustration of “the madness of crowds,” it was the impossibility of irrational behavior within DSGE models that struck many critics as a fatal flaw.

The Future of DSGE

Just as the Depression and its aftermath inspired Tinbergen and Klein, and the crises of the 1970s inspired Lucas, Kydland, and Prescott, the current macroeconomic moment now appears ripe for a shift in how we think about modeling the economy. One approach, which appears to have been adopted by the Federal Reserve Board, is to downplay the importance of the Lucas critique in macroeconomic modeling, as witnessed by its aggressive use of the FRB/US model in policy analysis. However, most macroeconomists believe that the Lucas critique still holds, and that strong microfoundations are a necessity for a successful macroeconomic model. Kocherlakota (2010) presents an eloquent and inspiring overview of the current state of macroeconomics and several promising directions for the future. Let me add to his vision by suggesting two incremental shifts, and one radical shift in how we might approach this challenge.

The first shift is to take risk seriously in macroeconomic models and incorporate individual, institutional, and regulatory responses to changing risks—both real and perceived—in their decisions. We measure many aspects of our macroeconomy such as inflation, output, and unemployment but we currently have no measure of aggregate risk in the economy. Macroeconomic risks must be modeled more accurately and monitored more regularly, and traditional macroeconomic forecasting models may simply lack the necessary resolution to capture modern financial crises in the making. A broader set of measures, including those that focused on patterns in financial linkages, may be more appropriate to act as early warning indicators.22 These risk models would complement standard macroeconomic forecasts the way the National Weather Service provides real-time monitoring and active alerts—hurricanes, tornadoes, and floods cannot be legislated away but our early warning system has greatly reduced property damage and the number of lives lost.

The second shift is to incorporate the financial sector more fully into existing DSGE models. It may be that the financial crisis simply came too soon in the development of DSGE modeling for it to have reached its full utility. One promising approach, following Simon Gilchrist, has been to introduce a financial accelerator to amplify the effect of shocks to the credit market on the general economy.23 A calibrated DSGE model that includes the financial accelerator can successfully account for the overall drop in economic activity during the crisis period, as well as the dramatic increase in credit spreads. A richer, more robust description of financial markets is needed.

The third shift, and the most radical, is to reexamine the microfoundations of the DSGE model. Instead of agents that optimize their behavior according to rational expectations, we should consider agents with predictable irrationalities in their behavior—investor psychology, for example. Just as the optimizing behavior and rational expectations of agents in a standard DSGE model are analogous to the Efficient Markets Hypothesis, the adaptive behavior of agents in this variation have their parallels in the Adaptive Markets Hypothesis.24 These agents would fulfill the spirit of the Lucas critique by adapting to economic circumstance, not necessarily in an optimal way, but subject to the past environment of the agent.

Macroeconomic models need to take into account Keynes’ infamous animal spirits as well as economic rationality. The assumption that aggregate behavior in an economy is approximated by optimizing, forward-looking behavior should have been put to rest by the Financial Crisis, if not the housing bubble, or the dot-com bubble, or the Dutch tulip bubble in the mid-1600s. Optimization holds one great virtue for the modeler: it is mathematically precise. However, complexity and psychology are two sides of the same coin. Can the full complexity of the macroeconomy be captured in the behavior of precise optimizing agents, or will disasters like the recent crisis continue to emerge just beyond the grasp of our models? When Albert Einstein was criticized for the complexity of his theory of relativity, he responded that “A theory should be as simple as possible, but no simpler.” The same applies to theories of the macroeconomy. We may discover, as Keynes did over half a century ago, that from a policy perspective, being precisely wrong is not as helpful as being approximately right.25

1 Solow (2010, p. 2).

2 Kocherlakota (2010, p. 5).

3 Dhaene and Barten (1989); Tinbergen (1937); Tinbergen (1981).

4 In statistics, a regression analysis is an econometric tool prominently used for investigating the relationships between variables. Typically, the investigator seeks to ascertain the causal effect of one variable for another. This requires that the investigator assemble data on underlying variables in question and use regression to determine the approximate quantitative effect of the causal variables upon the variables they influence. Investigators generally assess the degree to which the regression instills confidence in the results by assessing “statistical significance.”

5 Klein (1950).

6 Klein and Goldberger (1955).

7 Brayton et al (1997).

8 Brayton and Mauskopf (1985).

9 Phillips (1958).

10 Lucas (1976).

11 Kydland and Prescott (1982).

12 Brayton, Laubach, and Reifschneider (2014).

13 Smets and Wouters (2003).

14 Bayoumi (2004).

15 Harrison et al (2005).

16 Murchison and Rennison (2006).

17 Adolfson et al (2007).

18 Erceg et al (2006).

19 Woodford (2009).

20 Kocherlakota (2010), pp. 12-16.

21 Kocherlakota (2010), p. 16

22 Billio, Getmansky, Lo, and Pelizzon (2012), Bisias, Flood, Lo, Valavanis (2012).

23 Gilchrist and Zakrajšek (2011).

24 Lo (2004, 2012).

25 Research Support from the MIT Laboratory for Financial Engineering is gratefully acknowledged. I thank Bob Chakravorti and Clark Peterson for helpful comments and Jayna Cummings and Carlos Yu for editorial and research assistance. The views and opinions expressed in this article are those of the authors only, and do not necessarily represent the views and opinions of any institution or agency, any of their affiliates or employees, or any of the individuals acknowledged above.