Regime-Conditional Performance

A strategy that delivered +12% CAGR over 10 years may have delivered +40% in 2020 and −8% in every other year. The aggregate number hides the regime structure. Regime-Conditional Performance breaks results down by macro regime — expansion vs. recession, high vs. low volatility, trending vs. range-bound — so you can see whether the strategy works broadly or only in specific conditions.

Regime definitions

FM103 classifies each rebalance period into a regime using three macro signals (see Macro Regime Classification):

1. Volatility regime: high vs. low using the CBOE VIX index. VIX > 22 is "high vol"; below is "low vol."
2. Yield-curve regime: normal vs. inverted using the 10Y−2Y Treasury spread. Negative spread is "inverted."
3. Trend regime: bull vs. bear using S&P 500 200-day moving average crossover.

The composite label combines the three (e.g., "low-vol expansion," "high-vol bear"). The number of distinct regimes observed depends on backtest length; a 5-year backtest typically sees 2–4 regimes.

Per-regime metrics

For each regime observed, the sub-pill reports:

• Number of rebalance periods in the regime
• Mean per-period return
• Annualised Sharpe ratio (rescaled by periods per year)
• Hit rate — fraction of periods with positive return
• Max drawdown within the regime

The headline: "Strategy posted positive Sharpe in 3 of 4 regimes" — the regime breadth statistic. This is the single number that goes into the Risk sub-score of the Strategy Health Card.

What regime breadth tells you

Conditional deployment

A strategy that works in 2 of 4 regimes is not useless — if you can detect the favourable regime in real time, you can deploy conditionally. The pattern: "deploy this momentum strategy only when VIX < 22 and the curve is not inverted." The platform doesn't automate this overlay, but the sub-pill output is the input for designing it.

Limitations

• Regime sample size. A 20-period backtest may see one regime only 3 times. Sharpe estimates with N=3 are essentially useless. The sub-pill flags regimes with N < 5 as "low confidence."
• Regime persistence. Adjacent rebalance periods often fall in the same regime, so within-regime observations are autocorrelated — effective sample is smaller than period count.
• Regime taxonomy is a model. The high-vs-low VIX threshold of 22 is conventional but not god-given. Sensitivity to the threshold is worth checking.
• Recent backtests miss recessions. Backtests starting after 2009 likely include zero NBER recessions; their regime breadth is partial by construction. Hamilton (1989) was first to formalise regime detection in macro time series.

Visualisation

The sub-pill shows two charts:

• A regime ribbon along the timeline coloring each period by its assigned regime.
• A per-regime Sharpe bar chart with confidence-interval error bars based on within-regime sample size.

Connecting back to factor return series

Regime-conditional analysis is most useful in combination with factor attribution. Pattern to look for:

• Momentum factor delivers positive spread in trending regimes, near-zero in choppy regimes.
• Value factor delivers positive spread in recovery regimes (post-recession), negative in late-cycle bull markets.
• Quality factor delivers positive spread consistently but with biggest contribution in high-vol regimes (flight to quality).

If your strategy's regime-conditional performance does not match the factor literature's pattern, either you have a different factor implementation or the strategy is not behaving as advertised. The cross-check is valuable.