Module 6 of 14 · Directional strategies

How fast should you trade?

A tempting belief is that markets are fractal: a pattern that works on daily bars works on hourly bars, and on two-minute bars, only faster and with more opportunities. Decades of futures data say otherwise. Which effect dominates depends on the horizon:

  • Several weeks to several months: trend following works best. This is the sweet spot.
  • Months to a year: trend still works, a little less well.
  • A few days: trend weakens sharply, and short-term mean reversion starts to appear (Module 11).
  • Intraday: directional trend rules on minute or hourly bars show little or no edge even before costs, and the costs are enormous.
  • Years: prices tend to revert toward long-run value, but the effect is too slow to measure reliably.

So the speed of a trend rule is not a free parameter to optimise. It interacts with two things at once: how much edge exists at that horizon, and how much turnover — and therefore cost — that horizon demands.

🎛 Speed ladder lab

0.004
0.15

Best speed after costs

32/128

net SR 0.85

Fastest rule

-0.85

gross -0.39 · turnover 114

Cost drag on fastest

0.46

Sharpe units / yr

Each bar averages four independent 20-year synthetic markets. Net Sharpe = gross Sharpe − turnover × cost per trade (in Sharpe units, from Module 4). The synthetic market has a little day-to-day reversal built in; raise it to mimic the noisier, mean-reverting behaviour real markets show at very short horizons. Daily data: an intraday rule would sit far to the left of the first bar, with turnover in the thousands.

The fastest rules lose twice

Even with costs set to zero, the fastest rule earns little or nothing: day-to-day noise partly undoes itself, so a hair-trigger crossover keeps buying just before small reversals. Add a realistic cost per trade and its turnover turns a small gross edge into a large net loss. Medium speeds hold up best; the slowest give back some edge because they react late.

The practical answer is to run several speeds together— for example fast spans of 8, 16, 32 and 64 days — dropping any speed that breaks a market's cost budget. Different speeds are only partly correlated, so the blend beats most single speeds (Module 9). A system that trades trend crossovers on two-minute bars sits off the left edge of this chart: it has the turnover of a market-maker without a market-maker's edge.

Things to try

  • • Set cost to zero. Is the fastest rule profitable even then? Now raise the reversal slider and watch it get worse while the slow rules barely notice.
  • • Find the cost per trade at which the 4/16 rule stops being worth trading. Compare that with the risk-adjusted costs you found in Module 4.
  • • Set reversal to 0 (a market with no short-term noise structure): fast rules recover — the damage really does come from the market's short-horizon behaviour.

Original interactive lesson. Labs run on a seeded synthetic market in your browser, built to illustrate mechanics, not to predict any real market. Contract figures are illustrative; check exchange specifications. Educational analysis, not investment advice.

How are we doing?