Module 7 of 14 · Directional strategies
Buffering — the no-trade zone
A forecast that changes every day, multiplied by a volatility estimate that changes every day, produces an ideal position that wiggles constantly. Chasing every wiggle generates a stream of tiny trades whose expected value is roughly zero and whose cost is certain.
Buffering puts a band around the ideal position — say ±10% of your average position. While the position you actually hold sits inside the band, do nothing. When the ideal moves far enough that you fall outside, trade only back to the edge of the band, not all the way to the centre. That second detail matters: trading to the edge means a small reversal puts you back inside the band with no further trade.
With whole contracts the band is computed in contracts: round the lower edge down, the upper edge up, and only trade if your current holding is outside that range. For small positions this rounding is often the whole buffer.
🎛 Buffer lab
Turnover
26 → 20
25% less trading
Gross Sharpe
0.34 → 0.37
before costs
Net Sharpe
0.23 → 0.29
Annual cost
0.132 → 0.099
Sharpe units
Positions are measured in “average positions” (forecast 10 = 1.0). With a buffer of b, nothing happens while the held position stays within ±b of the ideal; when it escapes, you trade only back to the nearest edge of the band, not all the way to the ideal. The chart shows the last 180 days of a 20-year synthetic run.
Free money, within limits
A modest buffer cuts turnover substantially while barely touching gross performance, because most of the trades it skips were noise. With realistic costs the net Sharpe ratio rises. Push the buffer too wide and you start ignoring real signal: the held position lags the ideal for too long and gross performance falls.
Buffering is the cheapest improvement in this course — a few lines of code — and it matters most for faster rules and expensive markets. Strategies that rely on precise positions every hour, like the fast mean reversion in Module 11, are the exception: there the buffer destroys the edge, and execution tricks do the cost-cutting instead.
Things to try
- • With the default fast span of 8, move the buffer from 0% to 10% and read the turnover tile. Then try 50%.
- • Set cost to zero: the buffer's benefit almost disappears. It is a pure cost-saving device, not a source of edge.
- • Use a fast span of 2 with a 30% buffer. Does buffering rescue the fastest rule from Module 6? (It helps, but it can't create edge that isn't there.)
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.