A key distinction between economies and physical systems, like the weather, is that economies consist of individuals who form expectations about the future. These beliefs not only influence the outcome of the system, but they also depend on it. Expectations about future income and interest rates enter into individual decisions about how much to work, invest and save, which in turn influence today’s interest rates and GDP. At the same time, interest rates and GDP provide feedback that shapes our thinking about their future values.
This feedback loop between expectations and outcomes is sometimes referred to as ‘reflexivity’. Here, I refer to it simply as the expectations channel (see Figure 1). Expectations can be very powerful. An expectation that the economy will enter an upturn, for example, may generate enough economic activity to make the expectation a reality: expectations can be self-fulfilling. This is benign provided that expectations are stabilizing – preserving current growth in the economy.
Figure 1: The expectations channel
But expectations may also be destabilizing when they become strongly divorced from reality. This can lead us to make decisions with actual risks and consequences that we haven’t counted on. When those risks are realized, we change our expectations and take evasive action. The result is usually instability. The boom and subsequent collapse of the US sub-prime mortgage market is a case in point.
My research project, funded by Rebuilding Macroeconomics, will investigate whether expectations drive big events like the Financial Crisis. Expectations will be modelled to include the cognitive limitations of individuals and capture the feedback loop between expectations and outcomes (Figure 1). In particular, agents will form expectations using simple heuristics, or rules of thumb.
Some of the heuristics will be based on economic fundamentals. Others will be extrapolative by taking recent outcomes and projecting them into the future. The fraction of agents using each rule is determined by past forecast performance. Hence, agents will gravitate to those rules that forecast best, and away from those that do poorly.
Agents may choose between extrapolative rules based on different sample sizes. To see why this matters, suppose house prices increase following good news about the economy. In response, agents will initially switch to “extrapolative rules” with relatively short horizons that capture the recent trend in prices. This in turn will fuel additional price increases and extra switching to short horizon extrapolative rules, which better track the most recent observations. This starts a multiplier effect that works through the expectations channel, and amplifies the impact of the initial news.
Although this mechanism is potentially destabilizing at the aggregate level, it is sensible from an individual perspective because heuristics are chosen based on forecast performance. In this context, myopic behaviour results not from simple overconfidence or pessimism, but from meaningful attempts to forecast where the economy is heading. These ‘animal spirits’ are generated by the model.
Figure 2 plots the value of the US stock market since 2002 (left panel) and survey expectations of its value next year (right panel). There is clear evidence of heterogeneity: some households are optimistic about the stock market and others are relatively pessimistic. Once the stock market began rising in 2003, more households adopted optimistic expectations and fewer held pessimistic expectations.  This rise in optimism halted with the Financial Crisis. Indeed, once the stock market began falling in 2007-8, many households switched from optimistic to pessimistic expectations.
Although, expectations appear to respond to shifts in the stock market as opposed to anticipate them. Waves of optimism and pessimism may exacerbate market fluctuations by driving investors to buy or sell. From a modelling perspective, the trick is to ‘write down’ this simple intuition in equations which are also a reasonable description of reality. The simple heuristics approach to expectations is one way to do this.
Figure 2 – US stock market and expectations
To capture the links between asset prices, expectations and the wider economy, models built in the project will include a stock market and a housing sector in which house purchases are financed by credit. The illiquidity of housing will be modelled explicitly: some houses will go unsold for several periods. All models will be subject to empirical evaluation, and the results will be made publicly available. The main aim is to improve our understanding of what causes big events – in particular the extent to which expectations matter.
The project also aims to address some important policy questions: What types of policies are desirable when expectations are extrapolative rather than rational? Are booms built on rising house prices fundamentally different from those built on rising stock prices? What are the implications for monetary, fiscal and macro-prudential policy?
Visit Michael’s Rebuild Macro Research Project: Endogenous Extrapolation: Implications for Boom-Bust Cycles and Macroeconomic Policy
 This view has received some attention from academic economists. Examples include A Crisis of Beliefs by Gennaioli and Shleifer and Animal Spirits by Akerlof and Shiller.
 The heuristic approach contrasts with the mainstream assumption of rational expectations whereby agents use all available information and hence avoid predictable forecast errors. Gigerenzer and Gaissmaier (2011) note that simple heuristics are used by both individuals and organizations. Frankel and Froot (1990) point to the use of extrapolative and fundamental forecasting techniques by FX traders.
 Cars Hommes focuses on this class of models, plus experimental and empirical evidence, in Behavioural Rationality and Heterogeneous Expectations in Complex Economic Systems. See also Hommes (2005).
 Expectations are taken from the Survey of Consumers Table 20: Probability of Increase in Stock Market Next Year (quarterly series). Optimistic expectations are defined as reported probabilities of 75% or more. Pessimistic expectations are defined as reported probabilities of 24% or less. S&P 500 is the average closing value each quarter.
You may also like…
15 November 2018