Every backtest runs on real market history, after trading costs and regulatory fees. The strategies our research desk publishes go much further. Each one has to get through the kind of review an institutional quant team would run, and nearly every idea fails it.Real-world costs and fills. Every round trip pays the stock’s own measured bid-ask spread, never less than 2 basis points or half a cent a share a side at the price it traded at back then. On top of that come SEC and FINRA fees on every sale. Signals are calculated after the close and filled at the next trading day’s open, never at the closing price they came from.Total return, measured like a fund. Dividends are credited as they’re paid, uninvested cash earns the Treasury-bill rate, and Sharpe and Sortino ratios are measured against T-bills.Only what was knowable at the time. Earnings dates and financials are dated to when they were filed with the SEC, and short interest to when FINRA published it. Renamed companies keep their full history. Tests reach back as far as 2003, up to 23 years, and never cover less than 10.No survivorship bias. Each strategy is also run on up to 40 similar companies that were delisted during its test, and it’s refused if it lost money on them.A bar that rises with every idea tested. Test enough rules and a few will look brilliant by luck. So the desk counts every strategy it has ever tried and raises its bar to match. A strategy needs a t-statistic above 3.5 twice: once on its own trades, and again with its delisted look-alikes added in. That’s stricter than the 3.0 Harvey, Liu and Zhu proposed for new findings in finance.A report a professional would recognize. Every published strategy shows:its deflated and probabilistic Sharpe ratios, CVaR, Omega, Ulcer index and conditional drawdown at riskits rolling 12-month Sharpe, factor-adjusted alpha and after-tax returnits capacity and the trading cost it could absorbits correlation with every other published strategyPortfolios answer to their own tests. Their weights are fitted on the earlier years with shrinkage estimators (Ledoit–Wolf for risk, James–Stein for returns). They’re then judged only on the years held back. There a portfolio needs a Sharpe ratio of at least 1.0, an average correlation of 0.6 or less between members, and better results than its own members.Results are still hypothetical. No backtest captures every real-world detail, such as partial fills or the price impact of very large orders, and past performance doesn’t predict future returns.
