71 days in. Four of our six strategies are flat or red.

We committed publicly on January 1 to running 6 strategies with their DEFAULT academic parameters, every day, until further notice. 71 days later, here's where we are:

• XGBoost ML: +5.5% | Sharpe 2.14 | 7 trades | -3.0% max DD
• TSMOM: +2.8% | Sharpe 1.01 | 2 trades | -5.7% max DD
• HMM Regime: +1.8% | Sharpe 1.04 | 2 trades | -3.3% max DD
• Cross-Asset Pairs: +0.3% | Sharpe 0.53 | 1 trade | -1.0% max DD
• Cointegration Basket: 0.0% | Sharpe 0.00 | 0 trades | 0.0% max DD
• Pair Trading: -0.9% | Sharpe -1.89 | 6 trades | -1.2% max DD

Equal-weighted portfolio return: +1.6%. Equal-weighted Sharpe: ~0.47. That's an annualized return of about 8% with modest risk. Not glamorous. Not embarrassing either.

The honest part nobody wants to say

Four strategies out of six are flat or red. If a signal service was showing you this, they'd hide it. We're showing it because hiding it would defeat the whole point of what we're building.

Here's the thing: this is exactly what you should expect from a portfolio of academically-grounded strategies running in any single 3-month window. The academic papers behind TSMOM, Pair Trading, and Cointegration all report long-run Sharpes in the 0.8-1.2 range — on samples of 10-30+ years. Inside any 3-month slice of those samples, you'll find periods where individual strategies are losing 5%+ and the portfolio is flat.

Strategies don't 'perform'. Portfolios of strategies sampled over long horizons perform. A 3-month window is noise.

What I'd change (spoiler: nothing)

Every week since Jan 1, I've had some version of the same thought: 'should we drop pair trading? It's dragging the portfolio.' The answer every week is no, and the reason is important.

Removing pair trading now — after it has underperformed for 71 days — would be the textbook quant failure mode. You're not removing a bad strategy. You're performance-chasing. You're overfitting your strategy selection to your live window, which is survivorship bias with extra steps. The strategies you end up with will look amazing in hindsight and fail the moment the regime rotates.

The same trap applies in reverse: DOUBLING UP on XGBoost because it's +5.5% is just as wrong. These strategies were chosen because their academic justification is strong. Their live P&L is not new information — it's a sample of noise around the long-run distribution we already understand.

When I would actually change something

• A strategy's Sharpe drops below -1.0 AND the drawdown exceeds 15% — that's outside normal sample noise and warrants investigation (bug? regime breakdown?).
• The hash chain detects drift > 0.5% on any historical data point — that means something in the data pipeline changed and we need to understand why.
• A new academic paper cites a fundamental flaw in one of our implementations. We update the code, we document the change, we note inception is reset for that strategy.
• Two years have passed. After 2 years we'll have enough data to make real judgments about whether the long-run expectations match our live results.

See you in 29 days for the next update. The portfolio will have either crept up, crept down, or stayed flat, and whichever one happens, we'll still be running the same six strategies with the same parameters on the same assets.