Risk management: models, limits and stress tests
Understand what risk measures reveal and what they can miss.
Define the exposure before modelling it
List open positions, contract sizes, account currencies, leverage, funding costs and margin requirements. Include positions that share a market driver or depend on the same liquidity source.
A risk limit should be understandable before it is automated. Decide who or what will act when a limit is reached, and how an outage or unreliable data feed will be handled.
Read VaR and expected shortfall with their assumptions
Value at Risk estimates a loss threshold over a stated horizon and confidence level under a model. For example, a modelled 95% one-day VaR of $10,000 describes a threshold that the model expects losses to exceed on 5% of days. It is not a maximum possible loss or a verified probability for tomorrow.
Expected shortfall considers the average loss beyond a selected tail threshold. Both measures depend on data and modelling assumptions; neither removes uncertainty about extreme events.
Position sizing is more than a formula
Position size connects a market move to a cash outcome. A calculation based on a planned stop must still allow for slippage, gaps and costs. Volatility-based methods change exposure when measured volatility changes, but historical volatility may understate what comes next.
Kelly-style sizing depends on estimated probabilities and payoff distributions. Errors in those inputs can produce impractically large exposure. A mathematical result is not a personalised allocation recommendation.
Stress the conditions the backtest did not show
Consider wider spreads, disappearing liquidity, correlated losses, execution delays, market gaps and connection failures. Test combinations of adverse conditions, not just one parameter at a time.
Maximum drawdown measures a decline from a prior peak in the observed series. A historical maximum does not bound future losses. Define review or exposure-reduction procedures before a drawdown occurs.
Use several measures and review the process
Return, volatility, downside variability, drawdown and costs answer different questions. A Sharpe or Sortino ratio does not establish safety, and a favourable ratio can conceal a small or unrepresentative sample.
Review assumptions as the market and account change. Automation can help enforce rules, but incorrect settings or data can make the same error repeatedly and at speed.
Further reading
This guide explains concepts and ways to organise your research. It is not a recommendation to trade, a personal allocation plan or a promise of returns. Read the Terms of Business and your account conditions.