Case Study 1: Epicycles in Economics — When Models Only Explain the Past

The Setup

Macroeconomic models are among the most mathematically sophisticated tools in the social sciences. They use complex differential equations, stochastic processes, and computational simulations to represent the behavior of entire economies. They are built by some of the most highly trained quantitative minds in academia. And their track record of predicting major economic events is remarkably poor.

The Pattern

Consider the history of macroeconomic forecasting around major turning points:

  • 1973 oil crisis: Not predicted by mainstream models
  • 1987 stock market crash: Not predicted
  • 1997 Asian financial crisis: Not predicted by IMF models
  • 2001 dot-com crash: Timing and severity not predicted
  • 2008 financial crisis: Not predicted by the models used by central banks, rating agencies, or major financial institutions
  • 2020 COVID economic impact: Severity not predicted (though the pandemic itself was the exogenous cause)

After each of these events, the models were retrofitted — new variables were added, new mechanisms were incorporated, and the models were shown to "explain" the event in retrospect. But the ability to explain the past did not translate into the ability to predict the future.

This is the epicycle pattern operating in real time.

The DSGE Example

Dynamic Stochastic General Equilibrium (DSGE) models are the workhorses of modern central bank macroeconomics. After the 2008 crisis, DSGE models were widely criticized for failing to anticipate the collapse. The response was not to question the framework but to add new components: financial frictions, housing markets, behavioral elements, network effects. Each addition improved the model's ability to explain 2008 — after the fact.

The question Lakatos would ask: did these additions generate novel predictions that were subsequently confirmed? Or did they only improve retrospective fit? If the latter, the additions are epicycles — increasing complexity without increasing predictive power.

The Unfalsifiability Structure

Macroeconomic models exhibit several mechanisms of unfalsifiability:

Post-hoc rationalization (Mechanism 1): After any economic event, economists can construct a model that "explains" it. The explanations are compelling but were not available before the event.

Ad hoc auxiliary hypotheses (Mechanism 2): When models fail, new variables are added: "exogenous shocks," "structural breaks," "regime changes." Each is a legitimate concept individually, but their accumulation functions as epicycles that protect the core framework.

Moving goalposts (Mechanism 3): When prediction fails, the goal shifts from "predicting" to "understanding": "Our models aren't meant to predict specific events; they're meant to illuminate mechanisms." This is a legitimate epistemic function — but it's a retreat from the original claim that models could guide policy by predicting economic outcomes.

The Defense and Its Limits

Defenders of macroeconomic modeling make important points:

  • Economies are genuinely complex systems with emergent properties that resist prediction
  • Weather forecasting also fails at long time horizons, but meteorological models are still useful
  • "Understanding mechanisms" is valuable even without precise prediction
  • Some macro models have been useful for specific purposes (inflation targeting, monetary policy rules)

These defenses have merit. The question is not "Are macro models useless?" (they aren't) but "Are they held to appropriate standards given their actual track record?" If a model's defenders constantly shift between "this model predicts" (when pitching it) and "this model explains" (when it fails), that oscillation is a diagnostic signal.

Discussion Questions

  1. Apply the Epicycle Test to DSGE models. How many additions have been made since the original formulation? Were they predicted or reactive?
  2. Is macroeconomics at Level 3 (indirectly falsifiable through its larger framework) or Level 4 (not yet falsifiable)? Argue your case.
  3. What would it take to make macroeconomic predictions genuinely falsifiable? Design a specific test.
  4. Compare the epicycle accumulation in Ptolemaic astronomy with the variable accumulation in DSGE models. What structural similarities do you see?

References

  • Tetlock, P. E. (2005). Expert Political Judgment: How Good Is It? How Can We Know?. Princeton University Press. (Tier 1 — relevant for forecasting accuracy)
  • Romer, P. (2016). "The Trouble with Macroeconomics." (Tier 1 — an internal critique by a Nobel laureate)
  • Research by the Federal Reserve and other central banks has documented the forecasting track record of DSGE models. (Tier 2)