Volatility Targeting and Procyclical Deleveraging
A risk parity fund targets 8% annualized volatility and uses a 20-day realized volatility estimate to set leverage. Current allocations are 35% equities, 55% bonds, and 10% commodities; leverage is 1.4x. Realized portfolio volatility has jumped from 7% to 13% in three weeks. The equity trend model has turned negative, the bond-equity correlation has risen from -0.2 to +0.3, and commodities are in backwardation with positive momentum. To return to the 8% target, the model would cut leverage to about 0.86x, mostly by selling equities. The last two times the fund deleveraged after a volatility spike, it sold near a local low and underperformed. The CIO asks whether to follow the model, modify the lookback, or add a trend overlay.
1. Evaluate whether the risk parity fund should follow the mechanical volatility-target cut from 1.4x to 0.86x. Consider procyclicality, the trend signal, and the correlation shift.
Read the worked answer
The mechanical cut is defensible but incomplete. A 20-day realized volatility estimate rose from 7% to 13%, so to keep 8% target vol the leverage must fall roughly by the ratio 8/13, or to about 0.62x on the risk estimate; the stated 0.86x suggests the model uses a blended or longer estimate. Selling equities after a volatility spike is inherently procyclical: volatility clusters and often peaks near local lows, so the fund may sell low. But not cutting leaves the portfolio above its risk target, and the negative equity trend plus rising bond-equity correlation both argue that the risk increase is not merely noise. The correlation shift undermines the diversification that justifies leverage in the first place. A partial cut, not the full mechanical cut, is appropriate: reduce leverage to about 1.0-1.1x, sell the most volatile/correlated equity exposure, and keep some dry powder.
2. Propose modifications to the volatility-targeting process to reduce procyclicality without abandoning risk control.
Read the worked answer
Use a longer or blended volatility estimate, e.g., an exponential weighted average with a 60-day half-life plus a 20-day estimate, and apply a volatility floor based on long-run averages. Rebalance leverage in bands, not daily: only adjust when target leverage changes by more than 10-15%, and spread the adjustment over several days. Add a trend overlay that reduces equity risk when the 200-day trend is negative, independent of volatility, so the fund is not forced to sell only after vol spikes. Use correlation-adjusted risk budgets: when bond-equity correlation is positive, cut the assumed diversification benefit and lower overall leverage. Keep a hard maximum leverage cap. These changes make the process less reactive while preserving the link between risk and exposure.
3. What are the risks of overriding the model, and how would you govern the discretion?
Read the worked answer
The main risk is that discretion becomes discretionary bailout: the CIO overrides after every drawdown, the fund drifts above its risk target, and the long-run volatility objective is abandoned. There is also hindsight bias: the last two deleveragings looked bad, but a small sample does not prove the model is wrong. To govern discretion, pre-specify allowable overrides: only within a leverage band (e.g., 0.9-1.2x), only with a written rationale, and only when a second signal (trend or correlation) confirms. Require ex-post attribution: compare the override portfolio with the mechanical model over at least 12 months. Set a hard stop: if realized vol exceeds 10% for two consecutive weeks, the mechanical cut must be taken. Report to the investment committee monthly. This preserves accountability and prevents the override from becoming a permanent risk increase.