Five mistakes retail quants make

We have spent the last decade running, observing, and rebuilding systematic strategies. The same five mistakes show up over and over.

First: over-fitting. You add features, the in-sample Sharpe rises, you ship. Out of sample, the strategy is flat or losing. Fix: walk-forward, every time, no exceptions. Cap your model complexity by the size of the rarest event in your data.

Second: ignoring costs. The textbook Sharpe of a high-frequency strategy is enormous; the real Sharpe after spreads, slippage, and impact often goes below 1. If you do not subtract realistic execution costs from every backtest decision, you are building a strategy for a frictionless market that does not exist.

Third: leverage as a substitute for edge. Doubling leverage doubles CAGR and doubles drawdown. It does not improve Sharpe. The temptation is to lever an unimpressive strategy into impressive-looking returns; in a real drawdown, you blow up.

Fourth: failing to sanity-check the data. Adjusted prices that are computed wrong, corporate actions handled inconsistently, survivorship-biased universes that exclude the names that delisted — these are the most common sources of phantom alpha. If your backtest is unusually good, suspect the data before suspecting the model.

Fifth: regime confusion. A strategy designed during a bull market is not the same as a strategy that works through a cycle. Two years of out-of-sample data covering only a bull is not enough. Five years covering bull, bear, and chop is the minimum credible test.

Every one of these is a discipline problem, not a tooling problem. The hard work in quant is not the model — it is the patience to keep cutting through self-deception.