Blog - Volatility Targeting and Drawdown Control: A Practical Framework for Systematic Portfolio Risk

Abstract

Volatility targeting has become a central tool in systematic portfolio management, yet its interaction with path-dependent drawdown risk remains imperfectly understood in practice. This article examines a framework that combines ex-ante volatility targeting with explicit drawdown-sensitive constraints to manage portfolio risk in a disciplined, model-based way. The discussion distinguishes between genuine forward-looking risk budgeting and ex-post return smoothing driven by discretionary de-risking after losses. Building on the literature on conditional volatility modelling, volatility-timing, and risk-managed factor strategies (Engle, 1982; Bollerslev, 1986; Fleming, Kirby and Ostdiek, 2001; Barroso and Santa-Clara, 2015; Moreira and Muir, 2017), the article outlines how realised, EWMA and GARCH volatility estimates can be embedded in a practical exposure-scaling and drawdown-control protocol. The central trade-offs concern responsiveness versus turnover, the stability of parameter estimates, the treatment of crises and regime shifts, and the interaction of drawdown rules with investor behaviour, funding constraints and market liquidity. The main conclusion is deliberately modest: volatility targeting and drawdown controls can improve the governance and robustness of systematic portfolios when implemented transparently, tested rigorously out of sample and evaluated net of realistic frictions, but they do not eliminate tail risk or guarantee superior long-run performance.

Keywords

volatility targeting; drawdown risk; GARCH models; systematic investing; portfolio risk management; volatility timing; conditional drawdown; transaction costs

1. Introduction

Volatility and drawdown describe different dimensions of portfolio risk and should not be treated as interchangeable. Annualised volatility summarises the average dispersion of returns around their mean, while drawdown captures the realised peak-to-trough path experienced by capital over time, including its depth and duration (Chekhlov, Uryasev and Zabarankin, 2005). A portfolio may exhibit an apparently acceptable long-run volatility yet still suffer large, prolonged drawdowns that are operationally and behaviourally unacceptable.

This disconnect is particularly relevant for systematic portfolios that employ leverage, derivatives, or concentrated factor exposures. Even when unconditional volatility is well-controlled, sudden regime shifts, correlation spikes, and liquidity dry-ups can generate sharp equity losses, margin stress and forced deleveraging (Longin and Solnik, 2001). Investors and allocators typically react to drawdowns rather than to abstract volatility metrics, raising the risk of redemptions and strategy abandonment at precisely the worst time.

The objective of this article is to propose a research and implementation framework, rather than a guaranteed loss-control mechanism, for combining volatility targeting with drawdown-sensitive rules. The emphasis is on ex-ante risk budgeting, empirical validation and implementation realism, not on engineering smooth performance paths or implying that volatility-managed strategies are universally superior (Fleming, Kirby and Ostdiek, 2001; Moreira and Muir, 2017).

2. Volatility Targeting as an Ex-Ante Risk Control

Volatility targeting refers to the systematic adjustment of portfolio exposure so that the ex-ante volatility of returns is kept near a specified target. A simple implementation chooses the portfolio weight in the risky asset at time t as

w_t = σ_target / σ̂_t

where σ̂_t is an estimate of conditional or recent realised volatility and σ_target is the desired risk level (RiskMetrics Group, 1996; Moreira and Muir, 2017). When estimated volatility rises, exposure is reduced; when volatility falls, exposure is increased, potentially through leverage.

In practice, volatility-targeted portfolios face constraints: maximum leverage caps, minimum exposure floors, and discrete rebalance frequencies (e.g., daily or monthly). During sudden volatility spikes, the estimate σ̂_t may lag realised risk, leading to delayed de-risking or forced trading into illiquid conditions (Fleming, Kirby and Ostdiek, 2001; RiskMetrics Group, 1996). Moreover, stale estimates can cause the portfolio to be over-exposed heading into a regime shift or under-exposed during early recovery phases.

Economically, volatility-managed portfolios exploit the empirical observation that factor and market volatilities vary more over time than their conditional expected returns; scaling exposure inversely with volatility tends to allocate risk away from high-volatility states without proportionally giving up expected return (Moreira and Muir, 2017; Fleming, Kirby and Ostdiek, 2001). However, evidence on risk-adjusted performance is mixed across markets and implementations, and it is incorrect to treat volatility targeting as a free Sharpe-ratio improvement (Barto, 2019; Moreira and Muir, 2017).

It is crucial to distinguish:

• Volatility targeting as a forward-looking or contemporaneous risk budget, applied systematically via pre-defined rules.
• Ex-post volatility reduction resulting from discretionary de-risking after adverse performance, often driven by behavioural or institutional pressures.
• The difference between genuinely robust portfolio construction and the cosmetic smoothing of reported returns, which can obscure underlying tail risks and leverage exposure.

Only the first corresponds to a coherent ex-ante risk-management process; the others often reflect hidden timing bets and governance weaknesses.

3. Measuring Volatility: Realised, EWMA and GARCH Approaches

3.1 Realised or rolling historical volatility

Realised volatility uses recent returns over a fixed window, for example the standard deviation of daily returns over the past 20–60 days. Its intuition is straightforward: recent dispersion is informative about near-term risk. Strengths include simplicity, transparency and limited model dependence. Weaknesses arise from the arbitrary choice of window length, abrupt jumps when observations enter or leave the window, and limited responsiveness to fast-changing conditions (RiskMetrics Group, 1996).

Realised volatility can be highly sensitive to regime shifts: a low-volatility window just before a crisis will underestimate risk, while a high-volatility window after a shock may keep exposure suppressed long into recovery (Fleming, Kirby and Ostdiek, 2001). In a systematic strategy, realised volatility-based scaling is often suitable when transparency and robustness are prioritised over precise volatility forecasts, but its implementation must be accompanied by stress tests for different lookback horizons and sampling frequencies (Moreira and Muir, 2017).

3.2 Exponentially weighted moving average (EWMA) volatility

EWMA models, popularised in the RiskMetrics framework, compute volatility as an exponentially weighted average of past squared returns, with more recent observations receiving higher weight (RiskMetrics Group, 1996). Intuitively, this captures the persistence of volatility while allowing for gradual adaptation to new information.

The strengths of EWMA include ease of implementation, a single decay parameter, and a closer connection to conditional volatility than simple rolling windows. However, the choice of decay factor is non-trivial, and parameter instability can arise across asset classes and time periods (RiskMetrics Group, 1996). EWMA estimates can still react too slowly to abrupt structural breaks and may overemphasise transient spikes, depending on the calibration.

In systematic strategies, EWMA volatility is often used as a baseline risk estimator because it is simple, transparent and compatible with covariance-matrix construction. Implementation risks include look-ahead bias if parameters are calibrated on full-sample data and then applied retrospectively, and the temptation to overfit decay parameters to maximise backtest performance.

3.3 Conditional volatility models: ARCH/GARCH

ARCH and GARCH models, introduced by Engle (1982) and Bollerslev (1986), model conditional variance as a function of past squared returns and past conditional variances. The intuition is that volatility is clustered and mean-reverting, and that a parametric model can capture this dynamic more efficiently than ad hoc smoothing.

Strengths of GARCH-type models include their superior in-sample fit and, in many applications, improved out-of-sample forecasting performance compared with simple historical estimators (Bollerslev, 1986; Engle, 1982). They support extensions such as asymmetric responses, long memory and multivariate specifications. Weaknesses involve parameter estimation uncertainty, sensitivity to model specification, and the risk that structural breaks or regime shifts invalidate historical relationships (Engle, 1982; Bollerslev, 1986).

For systematic portfolios, conditional volatility models can provide more refined risk signals for volatility targeting or for constructing dynamic covariance matrices (Fleming, Kirby and Ostdiek, 2001). Implementation risks include overfitting, the use of non-stationary samples, and underestimation of uncertainty around volatility forecasts—particularly in crisis periods when model residuals and correlations behave in extreme ways (Longin and Solnik, 2001). In practice, many managers combine model-based forecasts with simpler realised measures and conservative overrides.

4. Drawdowns as a Path-Dependent Portfolio Constraint

Maximum drawdown is the largest peak-to-trough decline in portfolio value over a specified period. Rolling drawdown tracks the current distance from the most recent equity peak, while drawdown duration and recovery time measure how long it takes to return to the high-water mark. Conditional drawdown measures, such as Conditional Drawdown-at-Risk (CDD), focus on the expected severity of extreme drawdowns (Chekhlov, Uryasev and Zabarankin, 2005).

Drawdown is inherently path-dependent: two strategies with identical mean and volatility can have very different drawdown profiles depending on the sequencing of returns. Standard deviation, Value at Risk (VaR) and a single annualised volatility estimate cannot fully capture this path dependence, especially when return distributions show skewness, kurtosis and serial dependence (Chekhlov, Uryasev and Zabarankin, 2005; Rockafellar and Uryasev, 2000). Measures like Conditional Value at Risk (CVaR) and CDD attempt to focus explicitly on tail and path risk (Rockafellar and Uryasev, 2000, 2002).

Operationally, drawdowns matter because deep or extended losses can trigger:

• Capital-preservation concerns and risk-budget breaches.
• Margin calls and leverage constraints for derivatives and leveraged portfolios.
• Investor behaviour such as redemptions, mandate terminations or governance interventions.
• Strategy abandonment at precisely the point where expected risk premia may have improved.
• Liquidity stress and forced deleveraging when market depth is impaired (Longin and Solnik, 2001; Chekhlov, Uryasev and Zabarankin, 2005).

A drawdown-control rule—such as scaling down exposure once drawdown exceeds a threshold—can mitigate further losses in protracted adverse regimes. However, it may also crystallise losses, miss sharp recoveries, increase turnover and embed model risk in the choice of thresholds and recovery criteria. These trade-offs must be analysed explicitly rather than assumed away.

5. Integrating Volatility Scaling, Leverage Limits and Drawdown Controls

A coherent framework for systematic risk management combines:

• A volatility estimator σ̂_t (realised, EWMA or GARCH-based).
• A target risk budget σ_target.
• Maximum and minimum exposure limits, including leverage caps.
• A drawdown-sensitive exposure adjustment, based on rolling or conditional drawdown metrics.
• A re-entry or recovery rule governing when exposure is restored.
• Explicit transaction-cost and turnover controls.

In such a framework, the baseline exposure is given by volatility scaling, for example w_t = σ_target / σ̂_t, truncated to the leverage and exposure bounds (RiskMetrics Group, 1996; Moreira and Muir, 2017). Drawdown enters as an additional state variable: when drawdown exceeds a moderate threshold (e.g., 10–15 per cent), exposure may be gently reduced relative to the volatility-only rule; when a deeper threshold is breached, more aggressive scaling or hard caps may apply. Recovery rules specify how exposure is gradually reintroduced as drawdown heals or as volatility normalises.

The portfolio should not mechanically react to every price tick. Smoothing mechanisms include:

• Using slower-moving volatility and drawdown estimates (e.g., daily rather than intraday).
• Hysteresis or bands around thresholds to avoid frequent regime switching.
• Minimum time between rebalances.
• Governance rules requiring human review for large changes in risk.

Consider a purely conceptual example. Suppose a strategy targets 10 per cent volatility, with leverage capped at 2× notional and a minimum exposure of 20 per cent. When estimated volatility rises sharply (e.g., from 10 per cent to 20 per cent annualised), the volatility rule alone would cut exposure from 100 per cent to 50 per cent. If at the same time the portfolio is in a modest 8 per cent drawdown, no further adjustment is triggered. If drawdown deepens beyond 20 per cent, an overlay rule might cap exposure at 30 per cent until drawdown improves, even if volatility falls back.

Later, volatility may normalise while the portfolio remains in a 15 per cent drawdown. A pure volatility-targeting rule would suggest re-leveraging, but a drawdown-aware framework might restore exposure in stages (e.g., from 30 per cent to 50 per cent) contingent on partial recovery and liquidity conditions, thereby balancing opportunity with capital preservation. This example is illustrative only and not a prescriptive strategy.

6. Turnover, Transaction Costs and Implementation Risk

Dynamic risk scaling is particularly vulnerable to implementation shortfalls. Paper portfolios often ignore or understate the impact of:

• Bid–ask spreads, which widen during volatility spikes.
• Market impact from trading larger or more leveraged positions.
• Financing costs and borrowing constraints for leveraged or short positions.
• Futures roll costs when using derivatives for exposure management.
• Tax considerations, which are highly jurisdiction-dependent and can be materially affected by turnover.
• Execution delays, partial fills and slippage, especially during stress episodes.
• Correlation and liquidity breakdowns, where assets assumed liquid become costly or impossible to trade.

Realistic evaluation must focus on net performance after estimated trading costs, financing costs and slippage. Empirical work on volatility timing and risk-managed factor strategies typically includes transaction-cost assumptions and finds that net benefits can remain positive but are reduced relative to frictionless simulations (Fleming, Kirby and Ostdiek, 2001; Barroso and Santa-Clara, 2015). However, these results are sample- and implementation-dependent, and there is no guarantee that similar benefits will persist under different market conditions or for different universes.

Rebalance frequency is a first-order design choice. High-frequency volatility targeting can more closely track the desired risk budget but induces high turnover, higher transaction costs and greater exposure to temporary noise. Lower-frequency rebalancing reduces trading but risks being “behind the curve” in fast-moving regimes. The optimal compromise is context-specific and should be informed by sensitivity analyses and robustness checks, not by maximisation of a single performance metric in one backtest.

7. A Practical Research and Validation Protocol

A rigorous protocol for testing volatility-targeting and drawdown-control frameworks should include at least the following elements:

1. Define the universe and strategy. Specify instruments, data frequency, base trading rules and whether the risk overlay is applied to a benchmark, factor strategy or multi-asset portfolio (Fleming, Kirby and Ostdiek, 2001; Barroso and Santa-Clara, 2015).
2. Use point-in-time data. Employ datasets that preserve historical index membership, corporate actions and delistings to avoid survivorship and look-ahead bias.
3. Build volatility estimates using only information available at decision time. For realised, EWMA or GARCH estimates, ensure that only past returns are used, with parameters fixed or updated according to a realistic estimation schedule (Engle, 1982; Bollerslev, 1986).
4. Apply chronological, rolling or walk-forward validation. Separate in-sample model development from out-of-sample evaluation in time, using rolling windows or expanding windows to mimic live deployment (Fleming, Kirby and Ostdiek, 2001; Moreira and Muir, 2017).
5. Include realistic costs and constraints. Model transaction costs, bid–ask spreads, financing, futures rolls, margin requirements and borrowing limits; where possible, anchor assumptions to empirical studies of trading costs and price discovery (Fleming, Ostdiek and Whaley, 1996).
6. Test across market regimes. Evaluate behaviour in calm markets, crises, sharp recoveries and volatility spikes, including periods of elevated cross-asset correlation (Longin and Solnik, 2001).
7. Stress-test key parameters. Vary volatility lookback windows, EWMA decay factors, GARCH specifications, volatility targets, drawdown thresholds and rebalance frequencies to assess robustness (Barroso and Santa-Clara, 2015; Barto, 2019).
8. Report a rich set of metrics. Beyond average returns and Sharpe ratios, report realised volatility, maximum drawdown, drawdown duration, tail-risk metrics such as CVaR and CDD, turnover, and performance stability across subperiods (Rockafellar and Uryasev, 2000; Chekhlov, Uryasev and Zabarankin, 2005).
9. Avoid parameter mining. Do not select parameters solely because they maximise in-sample Sharpe or minimise drawdown; instead, prioritise configurations that perform reasonably across multiple samples and regimes.
10. Document governance and override rules. Specify how the risk framework is monitored, when human overrides are permitted, and under what circumstances the framework may be suspended (for example, in extreme market closures or structural breaks).

The overarching principle is robustness, not optimisation. The aim is to build a framework that is acceptable across a range of plausible scenarios rather than one that performs best in a single historical sample.

8. Limitations and Failure Modes

Volatility targeting and drawdown control rest on models that necessarily simplify reality. Volatility itself is an incomplete representation of risk: it does not directly capture liquidity risk, gap risk, jump risk, counterparty risk or model risk (Rockafellar and Uryasev, 2000). Estimated volatility can change too slowly, leaving portfolios over-exposed heading into crises, or too quickly, prompting whipsaw trading in response to transient shocks (Fleming, Kirby and Ostdiek, 2001; RiskMetrics Group, 1996).

Regime changes can invalidate historical relationships, leading to miscalibrated volatility and correlation estimates. During stress periods, correlations between risky assets often increase, reducing diversification precisely when it is most needed (Longin and Solnik, 2001). Dynamic deleveraging can amplify selling pressure if many participants employ similar volatility-sensitive rules, potentially exacerbating price moves.

Drawdown rules can both mitigate and create problems. They may reduce exposure in adverse conditions, but they can also lock in losses, increase trading, and cause strategies to miss sharp recoveries, particularly when markets mean-revert strongly after crises (Barroso and Santa-Clara, 2015; Barto, 2019). Backtests of drawdown-sensitive overlays are especially prone to overfitting because threshold choices, lookback windows and re-entry rules can be tuned to particular historical episodes.

Backtests in general can be distorted by data-quality issues, survivorship bias, the omission or underestimation of trading costs, and post-hoc parameter selection (Fleming, Kirby and Ostdiek, 2001; Moreira and Muir, 2017). A lower realised volatility profile does not automatically imply a superior economic outcome: investors care about net returns, tail behaviour, capital preservation, and the interaction with their broader constraints and liabilities (Moreira and Muir, 2017; Barroso and Santa-Clara, 2015).

No risk-control framework can fully eliminate tail risk, gap risk, liquidity shocks or model failures. At best, volatility targeting and drawdown-aware controls provide a structured way to allocate and monitor risk, while acknowledging that extreme scenarios remain possible.

9. Conclusion

Volatility targeting and drawdown-sensitive controls offer a disciplined way to manage the risk of systematic portfolios, but they are not panaceas. When implemented as ex-ante risk budgets grounded in transparent volatility estimates and explicit constraints, they can help align realised risk with investor objectives and mitigate some of the most damaging path-dependent losses (Moreira and Muir, 2017; RiskMetrics Group, 1996). When misused as ex-post smoothing tools or overfitted overlays, they can obscure leverage and tail exposure, creating a false sense of security.

The literature on volatility timing, risk-managed factor strategies and conditional tail and drawdown measures provides a rich foundation for designing and evaluating such frameworks (Engle, 1982; Bollerslev, 1986; Fleming, Kirby and Ostdiek, 2001; Barroso and Santa-Clara, 2015; Moreira and Muir, 2017; Rockafellar and Uryasev, 2000; Chekhlov, Uryasev and Zabarankin, 2005). The most credible implementations are those that are systematically defined, tested out of sample, evaluated net of realistic costs and embedded within a robust governance framework.

Ultimately, volatility targeting and drawdown control should complement—not replace—core portfolio principles: diversification across independent risk premia, careful liquidity and funding planning, sound position-sizing and independent risk oversight. A systematic risk framework is valuable not because it promises a smooth ride, but because it makes explicit the trade-offs, assumptions and limits of what risk management can realistically achieve.

This article is for educational purposes only and does not constitute personalised investment advice or a recommendation to implement any specific strategy.

References

Barroso, P. and Santa-Clara, P. (2015) ‘Momentum has its moments’, Journal of Financial Economics, 116(1), pp. 111–120. https://doi.org/10.1016/j.jfineco.2014.11.010

Barto, M. (2019) Various volatility-managed portfolios and their performance. Master’s thesis. University of Gothenburg. Available at: https://gupea.ub.gu.se/handle/2077/61344

Bollerslev, T. (1986) ‘Generalized autoregressive conditional heteroskedasticity’, Journal of Econometrics, 31(3), pp. 307–327. https://doi.org/10.1016/0304-4076(86)90063-1

Chekhlov, A., Uryasev, S. and Zabarankin, M. (2005) ‘Drawdown measure in portfolio optimization’, International Journal of Theoretical and Applied Finance, 8(1), pp. 13–58.

Engle, R.F. (1982) ‘Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation’, Econometrica, 50(4), pp. 987–1007. Stable URL: https://www.jstor.org/stable/1912773

Fleming, J., Kirby, C. and Ostdiek, B. (2001) ‘The economic value of volatility timing’, The Journal of Finance, 56(1), pp. 329–352. https://doi.org/10.1111/0022-1082.00327

Fleming, J., Ostdiek, B. and Whaley, R.E. (1996) ‘Trading costs and the relative rates of price discovery in the stock, futures, and option markets’, Journal of Futures Markets, 16(4), pp. 353–387.

Longin, F. and Solnik, B. (2001) ‘Extreme correlation of international equity markets’, The Journal of Finance, 56(2), pp. 649–676. https://doi.org/10.1111/0022-1082.00340

Moreira, A. and Muir, T. (2017) ‘Volatility-managed portfolios’, The Journal of Finance, 72(4), pp. 1611–1644. https://doi.org/10.1111/jofi.12513

RiskMetrics Group (1996) RiskMetrics™ – Technical Document. 4th edn. New York: J.P. Morgan/Reuters. Available at: https://www.msci.com/research-and-insights/paper/1996-riskmetrics-technical-document

Rockafellar, R.T. and Uryasev, S. (2000) ‘Optimization of conditional value-at-risk’, Journal of Risk, 2(3), pp. 21–41. https://doi.org/10.21314/JOR.2000.038

Rockafellar, R.T. and Uryasev, S. (2002) ‘Conditional value-at-risk for general loss distributions’, Journal of Banking and Finance, 26(7), pp. 1443–1471. https://doi.org/10.1016/S0378-4266(02)00271-6