A regime model that only shows you historical averages is a chart, not a strategy. This page runs a walk-forward backtest: at each monthly rebalance, classify the regime using only data available on that date, rank sectors by their prior-period forward returns in that regime as known at that date, and hold the top-3 equal-weight for the next month. No look-ahead; every decision uses only information a real allocator would have had. Benchmarks: SPY buy-and-hold, and a 60/40 SPY+IEF monthly-rebalanced portfolio. Related: regime view, cycle position, inflation persistence.
Log scale. Each strategy starts at $100 on the backtest-start date. Regime Rotation = 100% in the 3 sectors with the highest prior-period forward-6m return in the current regime, equal-weighted, rebalanced monthly. SPY = buy-and-hold. 60/40 = 60% SPY + 40% IEF (7-10Y Treasury), rebalanced monthly.
CAGR is compound annual growth. Annualized volatility and Sharpe use monthly returns (Sharpe assumes 2% risk-free). Max drawdown is worst peak-to-trough decline on the equity curve. Beta is regression slope vs SPY. Hit rate is % of months the strategy’s monthly return exceeded SPY’s. Turnover is the average absolute weight change per rebalance, annualized.
When is the regime tilt actually earning its keep vs. just owning the index? Each point is the strategy’s trailing-12m return minus SPY’s over the same window. Above zero = the tilt added value; below zero = the tilt cost you. Consistent small positive readings > big positive spikes (which usually mean big negative spikes are coming).
Realized annualized return of each strategy, decomposed by the regime it was operating in. The regime tilt should materially outperform in regimes where sector dispersion is highest (typically Reflation and Stagflation) and approximately match SPY in Goldilocks.
Walk-forward discipline. At rebalance month t, the sector ranking uses only historical months before t that share the current regime. No future information ever enters a decision. This is the minimum bar for an honest backtest.
What's included. Adjusted close prices (Yahoo), dividends and splits reflected. Monthly rebalance (last business day). Regime classification identical to the main dashboard.
What's excluded. Transaction costs (would dock ~10-30bp/yr at monthly turnover levels). Taxes. Sector-ETF expense ratios (~10-15bp). Slippage. Rebalance-timing effects (month-end is used; intra-month wouldn't change the picture materially but changes the numbers).
Sample-size handling. If a regime has fewer than 6 historical observations at rebalance time, the strategy falls back to equal-weighting all sectors (a neutral stance) for that month. This prevents whipsaw on thin priors. At the start of the backtest roughly ~8-12% of months may use the fallback; by 2015 onward the prior is always well-populated.
Known limitations. The regime classifier itself is rebuilt monthly using all available history through t, but the methodology (inflation gauge, growth composite, z-window) is fixed. Real-time regime identification is noisier than the smoothed historical view suggests.
What this does and doesn't prove. A positive result shows the regime framework contained some signal over this sample period. It does not prove the signal persists. A negative result would be a reason to demote the framework. Read it as Bayesian evidence, not forecast.