Building a US-equity bot, I bolted on nine timing structures. The result was the humbling kind.
Nine timing structures, all below SPY
Trend-following, regime switching, volatility targeting, momentum ranking — I compared all nine against SPY on raw CAGR. None of them won. With SPY as the benchmark over the backtest window, every layer of timing added friction (trades, cash-waiting, whipsaws) and no excess return.
The only winner: static 1.2x
Exactly one thing beat SPY — no timing, enter once, hold at 1.2x margin leverage. In a long bull market, of course it did.
That isn't alpha
Here's where you have to be honest. The 1.2x won not because it forecast the market but because it simply took 1.2x the exposure. Leverage wins in a bull — and breaks harder in a bear for the exact same reason. That's scaled beta, not manufactured alpha. Calling it "the strategy won" is self-deception.
The discipline it left
- Compare against the benchmark (SPY) on raw CAGR first. Before any risk-adjusted metric, check whether absolute return even clears the benchmark.
- If leverage won, write down "leverage won." Don't credit the timing.
- The nine timing structures are shelved as Phase 2, not killed. The verdict isn't done until I see whether a defensive premium shows up in a different regime (a slow, large bear). This window just didn't have one, so defense never got paid. The same rule as killing a strategy that fails its pre-registered gate applies — except this isn't "it lost," it's "this regime can't judge it."
The lesson: before you celebrate a backtest beating the benchmark, separate skill from leverage. It's usually leverage. And without the habit of pre-registering to block hindsight selection, it's easy to hand leverage's credit to timing.