Hedge Slippage From Rebalancing and Transaction Costs

Why a Hedge Can Underperform Its Design

A hedge is usually designed using a target position, market sensitivities, and assumptions about how often the position can be updated. Actual implementation differs: trades cross a bid-ask spread, large orders may move the market, and a position held between rebalances can become stale as exposures change. These effects create hedge slippage: the difference between the intended hedge result and the result after trading and imperfect tracking.

A useful comparison must include both sides of the trade-off. Rebalancing more often can reduce tracking error, but it can also increase total spread costs and create additional opportunities for market impact. This tutorial develops a simplified, explicit framework for comparing those effects. It is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.

Definition: Hedge slippage is the shortfall of the implemented hedge relative to its intended or benchmark performance. In the model below, it has two distinct sources: residual tracking loss from a hedge that is not continuously updated, and execution costs from the trades used to rebalance.

A Model for Rebalancing and Execution Costs

Let a hedge be rebalanced $N$ times over a horizon of length $T$, with equal intervals $h=T/N$. In a stylized tracking-error model, the residual loss over interval $j$ is $e_j$, where the interval losses are independent, have mean zero, and have variance $a h^2$. The total residual is $R_N=\sum_{j=1}^{N} e_j$. The assumption that each interval’s error variance is proportional to $h^2$ is a calibration choice: it represents local exposure mismatch that grows with the time between updates. It is not a universal law for hedge error.

\operatorname{Var}(R_N) = \sum_{j=1}^{N}\operatorname{Var}(e_j) = N a h^2 = \frac{aT^2}{N}

For execution, suppose each trade changes the hedge by $\Delta q_j$ units. Let $c_s$ be the one-way half-spread cost per unit traded and $\kappa$ the coefficient in a local quadratic market-impact model. The estimated cost of one trade is

C_j=c_s|\Delta q_j|+\kappa(\Delta q_j)^2

The first term charges for crossing the spread; the second represents market impact that increases more than proportionally with trade size. This is a simplified cost model, not a general market-impact law. It is most appropriate for a specified trading range and should not be extrapolated to very large orders or illiquid markets without further evidence.

If target changes over interval $h$ are modeled as normal with mean zero and standard deviation $g\sqrt{h}$, their expected absolute size is $g\sqrt{2h/\pi}$, while their expected squared size is $g^2h$. Consequently, for equal intervals the expected total execution cost is

E[C_N] = c_s g\sqrt{\frac{2T}{\pi}}\sqrt{N} + \kappa g^2 T

This result highlights an important feature of the assumptions: expected spread cost rises with the square root of the number of rebalances, while expected quadratic impact is constant over $N$. Other patterns of target changes or execution costs can produce a different relationship. In particular, fixed charges per trade or costs that rise sharply with urgency can make frequent rebalancing less attractive.

Modeling convention: For a consistent planning comparison, let the residual tracking loss be $R_N\sim N(0,A/N)$, where $A=aT^2$, and approximate execution costs by their expected amount $C_N=E[C_N]$. Then the modeled total loss is $L_N=C_N+R_N$, so its $p$-quantile is $C_N+z_p\sqrt{A/N}$, where $z_p$ is the standard normal quantile. This approximation treats costs as fixed; it does not model their randomness or dependence on market moves.

Deriving a Cost-Aware Rebalancing Frequency

At confidence level $p$, the modeled loss quantile is a simple way to compare strategies when residual tracking losses are normal and costs are approximated as fixed. Write $b=c_sg\sqrt{2T/\pi}$ and $K=\kappa g^2T$. Then the quantile at level $p$ is

Q_p(N)=b\sqrt{N}+K+z_p\frac{\sqrt{A}}{\sqrt{N}}

The constant $K$ does not affect the best continuous value of $N$. To find that value, differentiate the two terms that do vary with $N$, set the derivative to zero, and check that the second derivative is positive:

\frac{dQ_p}{dN} = \frac{b}{2\sqrt{N}} -\frac{z_p\sqrt{A}}{2N^{3/2}}=0 \quad\Longrightarrow\quad N^*=\frac{z_p\sqrt{A}}{b}

For positive $b$ and $z_p\sqrt A$, the second derivative of $b\sqrt N+z_p\sqrt A/\sqrt N$ at the stationary point is positive, so the point is a minimum. In practice, $N$ must be a feasible positive integer. Compare the adjacent integer schedules around $N^*$, and apply any operational constraints. The formula does not establish an optimal schedule if the model assumptions fail or if costs and tracking risk have been calibrated inconsistently.

Worked Example: Comparing Rebalancing Frequencies on Net 95% VaR

A hypothetical insurer evaluates hedge schedules over one year. Its calibrated residual-loss model is $R_N\sim N(0,360{,}000/N)$, with dollar losses. Its expected execution-cost model is $C_N=120\sqrt{N}+18$, also in dollars. Costs are treated as fixed at this expected amount for the comparison. Compare annual, quarterly, and nine-times-per-year rebalancing using 95% VaR. Use $z_{0.95}=1.645$, and compare each schedule with a no-hedge benchmark whose loss is normal with mean zero and standard deviation \$1,000.

Method. For each schedule, add the expected cost to the 95th percentile of residual tracking loss. Under the stated normal model, that percentile is $1.645$ times the residual standard deviation.

Annual rebalancing ($N=1$). The expected cost is $120\sqrt{1}+18=\$138$. The residual standard deviation is $\sqrt{360{,}000/1}=\$600$, so the residual 95% VaR is $1.645(600)=\$987.00$. Net 95% VaR is therefore $\$138+\$987=\$1{,}125.00$.

Quarterly rebalancing ($N=4$). The expected cost is $120\sqrt{4}+18=\$258$. The residual standard deviation is $\sqrt{360{,}000/4}=\$300$, so residual 95% VaR is $1.645(300)=\$493.50$. Net 95% VaR is $\$258+\$493.50=\$751.50$.

Nine rebalances ($N=9$). The expected cost is $120\sqrt{9}+18=\$378$. The residual standard deviation is $\sqrt{360{,}000/9}=\$200$, so residual 95% VaR is $1.645(200)=\$329.00$. Net 95% VaR is $\$378+\$329=\$707.00$.

The no-hedge 95% VaR is $1.645(\$1{,}000)=\$1{,}645.00$. The modeled net risk reductions versus that benchmark are $\$520.00$, $\$893.50$, and $\$938.00$ for the three schedules, respectively. Among these choices, nine rebalances has the lowest net VaR. The continuous optimum is $N^*=1.645\sqrt{360{,}000}/120=1.645(600)/120=8.225$, consistent with a schedule near nine rebalances. The check is that each net VaR equals its cost plus its residual 95% VaR.

Common mistake avoided: Comparing only residual standard deviations would favor more frequent rebalancing without limit. Adding execution cost shows why the risk measure can eventually rise as rebalancing becomes more frequent.

Separating Spread Cost from Market Impact

For a one-way trade of $q$ units, the model charges the half-spread on the absolute quantity traded and impact on the square of that quantity. If a same-direction order is split into $m$ equal pieces, total spread cost is unchanged under this model, but quadratic impact falls. The reason is that $m(q/m)^2=q^2/m$, whereas $m(q/m)=q$. That is a mathematical property of the assumed quadratic cost function, not a guarantee that splitting an order is beneficial in practice: delay, changing prices, information leakage, or fixed charges can offset the reduction.

Worked Example: Cost of a Large Trade Versus Staged Execution

A hypothetical portfolio must increase a hedge by 12 contracts over a year. For each execution, the one-way half-spread is \$1.50 per contract and the estimated impact cost is $\$0.08q^2$, where $q$ is the number of contracts in that execution. Compare one trade of 12 contracts with four equal trades of 3 contracts. Separately, the calibrated residual tracking-loss standard deviation is \$500 for the one-trade plan and \$250 for the four-trade plan. Assume normal residual loss, fixed execution costs, and use $z_{0.95}=1.645$.

One trade. Spread cost is $1.50(12)=\$18.00$. Impact cost is $0.08(12^2)=0.08(144)=\$11.52$. Total cost is $\$18.00+\$11.52=\$29.52$. The residual 95% VaR is $1.645(\$500)=\$822.50$, giving net 95% VaR of $\$29.52+\$822.50=\$852.02$.

Four equal trades. Each trade has spread cost $1.50(3)=\$4.50$ and impact cost $0.08(3^2)=\$0.72$. Across four trades, spread cost is $4(\$4.50)=\$18.00$, impact cost is $4(\$0.72)=\$2.88$, and total cost is $\$20.88$. The residual 95% VaR is $1.645(\$250)=\$411.25$, so net 95% VaR is $\$20.88+\$411.25=\$432.13$.

Under these assumptions, staging the trade reduces net 95% VaR by $\$852.02-\$432.13=\$419.89$. The spread cost is the same because the total quantity traded is unchanged; the modeled impact is smaller for four pieces. The check is that total spread cost equals \$18.00 under either schedule, while the four impact charges sum to \$2.88.

Common mistake avoided: Applying the quadratic impact rate to the total quantity and then multiplying by the number of executions would overstate the staged-trade impact. Apply the rate to each execution’s quantity, square that quantity, and then sum.

When Execution Costs Can Erase a Risk Reduction

A reduction in tracking risk is not automatically an economically useful hedge action. If the risk measure is a loss quantile, compare the decrease in the residual-loss quantile with the execution cost under the same market scenario. A positive difference indicates a modeled reduction in total quantile; a negative difference indicates that the cost exceeds the measured risk reduction. This is a decision aid, not a complete measure of value: a single quantile does not describe all outcomes, and the result depends on the selected confidence level and scenario.

Worked Example: Testing a Hedge Adjustment Under Routine and Stressed Costs

A hypothetical insurer considers a hedge adjustment that reduces the standard deviation of its residual loss from \$40,000 to \$25,000. Residual losses are modeled as normal with mean zero. The adjustment trades 1,000 contracts. The one-way half-spread is \$10 per contract. Estimated quadratic impact is $\kappa q^2$, with $\kappa=0.008$ in routine conditions and $\kappa=0.020$ in a specified stress. Treat costs as fixed within each scenario and use $z_{0.95}=1.645$. Determine whether the adjustment lowers modeled net 95% VaR in each case.

Gross reduction in residual 95% VaR. Before the adjustment, residual 95% VaR is $1.645(\$40{,}000)=\$65{,}800$. After the adjustment, it is $1.645(\$25{,}000)=\$41{,}125$. The gross reduction is $\$65{,}800-\$41{,}125=\$24{,}675$.

Routine execution cost. Spread cost is $\$10(1{,}000)=\$10{,}000$. Impact cost is $0.008(1{,}000^2)=\$8{,}000$. Total cost is $\$18{,}000$, so the net 95% VaR reduction is $\$24{,}675-\$18{,}000=\$6{,}675$. In routine conditions, the adjustment lowers the modeled net quantile.

Stressed execution cost. Spread cost remains $\$10{,}000$. Impact cost is $0.020(1{,}000^2)=\$20{,}000$, giving total cost of $\$30{,}000$. The net 95% VaR reduction is $\$24{,}675-\$30{,}000=-\$5{,}325$. In this specified stress, the cost exceeds the gross reduction, so the modeled net quantile is higher after the adjustment.

The check is that the same risk reduction is used in both cases; only the impact assumption changes. This example also shows why execution costs should be stressed alongside hedge effectiveness rather than assumed constant.

Exam Traps and Practical Interpretation

Using the full spread instead of the half-spread. If the quoted bid-ask spread is \$2.00 and a trade executes once against the midpoint benchmark, the one-way spread cost is typically modeled as \$1.00 per unit, not \$2.00. A round trip may incur both sides, depending on the benchmark and trades.
Ignoring turnover. Rebalancing frequency alone does not determine cost. Costs depend on the quantity traded at each rebalance, and repeated reversals can produce much higher turnover than a one-direction adjustment.
Comparing gross risk to net risk. A smaller residual variance can coexist with a larger loss quantile once execution costs are included. State the risk measure and include costs consistently.
Treating the cost formula as universal. Linear spread plus quadratic impact is a local approximation. It may be unsuitable when liquidity is discontinuous, market impact is asymmetric, or the order changes market prices for a prolonged period.
Adding deterministic cost to variance. A fixed cost shifts the loss distribution and its quantiles; it does not increase the residual variance. If costs are random or correlated with market losses, their distribution and dependence must be modeled rather than added as a constant.

Key takeaway: Compare rebalancing schedules using a common net-risk measure. State how residual tracking risk changes with frequency, calculate spread and market-impact costs from actual turnover assumptions, and test whether the risk reduction remains worthwhile under stressed execution costs.

Check Your Understanding

Use the stated cost and residual-risk assumptions in each question. Unless specified otherwise, treat costs as fixed and residual losses as normal; use $z_{0.95}=1.645$.

A one-way trade of 8 units has a half-spread cost of \$1.25 per unit and quadratic impact of $0.10q^2$ dollars. What is its total execution cost? (Answer: \$16.40)
A required 10-unit trade is split into five equal executions. The half-spread is \$2 per unit and impact is $0.30q^2$ per execution. What is total cost? (Answer: \$26.00)
For $N=4$, expected cost is $75\sqrt N+12$ dollars and residual standard deviation is $300/\sqrt N$ dollars. What is net 95% VaR? (Answer: \$408.75)
Compare $N=1$ and $N=9$ when expected cost is $50\sqrt N+5$ dollars and residual standard deviation is $400/\sqrt N$ dollars. Which schedule has lower net 95% VaR, and what is that VaR? (Answer: $N=9$, with net 95% VaR of \$374.33)
A continuous-frequency model has variable 95% VaR terms $80\sqrt N+1.645(720)/\sqrt N$, plus a constant cost independent of $N$. What continuous value minimizes the expression, and which nearby integer frequency has the lowest value? (Answer: $N^*=1.645(720)/80=14.805$; the lowest nearby integer frequency is $N=15$)

What You'll Learn