Buffering and No-Trade Bands
Rules that stop a strategy from re-trading every time a signal or rank flickers slightly, by requiring a move past a buffer zone before a position is added, trimmed, or reversed.
A raw signal or rank estimated from noisy data will jitter slightly from one rebalance to the next even when nothing meaningful has changed, and a strategy that mechanically re-trades to match every tiny wiggle pays trading costs for no real information gain. Buffering, or a no-trade band, fixes this by requiring a signal to move past some threshold — beyond its current position, not just any change — before the portfolio actually trades.
A common version ranks stocks and only sells a currently-held name once its rank falls outside, say, the bottom 25% (rather than the bottom 20% used to buy new names), and only buys a new name once it enters the top 15% (rather than the top 20% at which existing holdings are kept). This asymmetric band means a stock hovering right at the boundary doesn't get bought and sold repeatedly as its rank flickers by one or two positions each period.
The trade-off is real: wider bands cut turnover and transaction costs but let the portfolio drift further from the "ideal" ranking the raw signal implies, so the band width is itself a parameter to tune against the strategy's estimated cost per trade.
A no-trade band requires a signal to move past a buffer zone — not just change at all — before a position is traded, cutting the turnover caused by pure signal noise at the cost of letting the portfolio drift somewhat from the raw ranking.
Practice in interviews
Further reading
- Grinold & Kahn, Active Portfolio Management, ch. 16