Transaction Costs & Market Impact: Models & Analysis › Limitations and Practical Considerations

While the models presented in this chapter provide valuable frameworks for understanding and estimating transaction costs, several important limitations deserve attention.

Transaction cost parameters are not constants; they vary with market conditions, time of day, and broader volatility regimes. The square-root exponent $\gamma \approx 0.5$ is an empirical average that can range from 0.3 to 0.7 depending on the market and time period. Impact coefficients $\eta$ vary even more widely. A model calibrated during normal market conditions may dramatically underestimate costs during periods of stress or low liquidity.

This parameter uncertainty compounds with the inherent noisiness of cost measurement. Individual trades have highly variable execution quality due to randomness in order flow, market maker inventory, and timing. Reliable estimates require large samples, but by the time you have enough data, the market may have changed.

The models presented assume a passive approach where you accept market prices. In practice, execution algorithms can significantly reduce costs through techniques like:

• Splitting orders to reduce instantaneous impact
• Using limit orders to capture spread instead of paying it
• Timing execution to coincide with periods of high liquidity
• Exploiting dark pools and alternative venues

These techniques, covered in the upcoming chapter on execution algorithms, can reduce realized costs below what simple models predict, but they also introduce new complexities and potential failure modes.

The market impact models presented treat impact as a function of trade size alone, ignoring the strategic environment. In reality, other market participants observe and react to your trading. Predictable execution patterns can be exploited by high-frequency traders; information about your intentions can leak through broker channels or observable patterns.

For large institutional traders, information leakage can be a cost equal to or greater than direct market impact. This is another reason why actual trading costs often exceed model predictions, particularly for traders with significant market presence.

A subtle but important issue is that transaction costs and alpha are not independent. High-alpha opportunities often arise precisely in situations where costs are also high: illiquid markets, stressed conditions, or concentrated positions. A strategy that only generates alpha when you need to trade large size in illiquid conditions may have much lower capacity than a naive cost model suggests.

Conversely, the act of trading on alpha erodes that alpha through information revelation. If your signal is truly informative, your trades will permanently move prices; the permanent impact component represents the market learning from your trading. This creates a fundamental tension between exploiting alpha quickly (before others discover it) and trading slowly (to minimize impact).