Sounds like you’re on the right track, but I’d recommend tightening the confidence bounds first. Lower the alpha level, maybe to 1% instead of 5%, and double‑check the standard error on your residuals. If the series is highly autocorrelated, consider adding an AR term or a GARCH component so the model can capture clustering. Also run a back‑test with a rolling window—if the Sharpe ratio drops after the tweak, it’s probably too aggressive. Keep the risk parameters conservative until the out‑of‑sample performance stabilises.