9 Trading Costs, Backtests, and Model Risk
9.1 The market charges for implementation
A portfolio return before fees and a strategy return that can actually be captured are different objects. Common costs include commissions, bid-ask spreads, slippage, financing, exercise handling, market impact, and taxes. Options may also introduce complex expiration and assignment scenarios.
9.2 Research pitfalls
- Look-ahead bias: Using a datum before it would have been observable.
- Survivorship bias: Testing only securities that remain available today.
- Selection bias: Choosing a good-looking result after comparing many specifications.
- Overfitting: Learning the history instead of a transferable relationship.
- Turnover underestimation: Ignoring the trades needed to move from yesterday’s weights to today’s.
A credible experiment states its data cutoffs, timing conventions, transaction-cost assumptions, benchmark, and failure criteria before inspecting the final results.
9.3 Decision-making under imperfect models
One response is sensitivity analysis: vary inputs, transaction costs, correlations, and scenario probabilities. Another is to prefer portfolios or strategies whose usefulness is not dependent on one fragile estimate.
Lab idea: Compare the same rebalanced portfolio before and after a hypothetical 10-basis-point cost on the amount traded.
Living chapter. This is an initial working treatment. Derivations, examples, and references will be expanded in future editions.