The most reliable way to make a number lie is to pay people for hitting it. That is the disarming core of Goodhart’s Law: once a measure becomes a target, the act of optimization changes the system and breaks the original relationship. Markets, bureaucracies, and companies are adaptive organisms. Give them a yardstick and they will learn to game it faster than the yardstick can learn to adapt.
What Goodhart really meant
The idea emerged from monetary policy in the 1970s, when central banks tried steering the economy by targeting money-supply aggregates. It worked until it didn’t. As the targets hardened, financial firms innovated around the definitions—reshuffling liabilities, inventing substitutes, and shifting activity off the measured path. The statistical regularity that seemed stable during observation unraveled under control. The lesson was not “don’t measure,” but “don’t mistake a proxy for the thing you actually want.” In macroeconomics, most of our dials—price indices, credit aggregates, labor-market tallies—are proxies. They are good servants and terrible masters.
When finance optimizes the yardstick
Bank regulation offers a recurring case study. Risk-weighted capital was designed to make banks hold more equity against riskier assets. Predictably, banks sought assets blessed with low weights: highly rated securities with thin apparent risk, until the veneer cracked. A similar dynamic dogs popular risk metrics: when managers are judged on short-horizon value-at-risk, portfolios converge on the same trades, creating cliff effects exactly when liquidity vanishes. Benchmarks meant to discipline behavior can end up coordinating it. Even reference rates and performance thresholds become targets; once bonuses or funding costs hang on a figure, the figure attracts pressure. None of this requires malice. It simply reflects incentives at scale.
Designing metrics that fight back
You cannot opt out of measurement. You can design for adaptation. The practical toolkit is unglamorous but effective:
- Use multiple, independent measures; if one is gamed, others light up.
- Prefer outcomes over inputs and rotate secondary KPIs on a schedule.
- Combine rules with judgment: pair model-based limits with simple backstops (e.g., leverage caps alongside risk weights).
- Stress test the metric itself; run red-team exercises to surface gaming strategies before they scale.
- Audit and lag incentives so short-term tweaks don’t pay immediately.
In policy, this points to flexible frameworks—average targets, ranges, and explicit tolerance for model error—rather than single-point hero numbers. In firms, it argues for fewer universal dashboards and more context-rich reviews where data informs but does not dictate. The point is not to retreat from quantification, but to be honest about how quickly proxies decay under pressure.
Our take
In an economy built on optimization, Goodhart’s Law is not a curiosity; it is the background radiation. Leaders who treat KPIs and macro dials as oracles will keep getting surprised. The edge now lies with those who design incentives that are hard to hack, accept that no single metric can be sovereign, and reserve space for judgment when the dashboard starts to look too perfect.
