Position and Risk Limits in Simulation, Explained

A signal-only backtest sizes positions purely off conviction: the stronger the signal, the bigger the bet, with no upper bound. Real books don't work that way. A risk desk imposes a maximum position size per name, a maximum gross and net exposure for the book, and often a value-at-risk limit that shrinks the whole book automatically when realized volatility rises. None of that shows up in a backtest that only checks whether the signal fired — and the gap matters most in a crisis, which is exactly when real risk limits bind hardest.

Worked example: the drawdown a real desk would have been forced into

A trend-following strategy sizes positions inversely to recent volatility, so as markets get calmer it takes larger positions — a reasonable design on its own. Backtested with no risk limits, gross exposure reaches 340% of NAV during an unusually calm stretch that precedes a volatility spike.

Add a realistic constraint: a VaR limit that caps the book at a 2% daily 99% VaR, and a hard gross cap at 250% of NAV.

	No limits (naive backtest)	With risk limits
Gross exposure entering the vol spike	340%	250% (capped)
Realized loss on the spike day	−6.8% of NAV	−5.0% of NAV
Position the strategy wanted to add mid-spike	Adds to the trend	Forced to cut, VaR limit binds
Result	Full participation in the reversal that follows	Missed part of the recovery, having been forced to de-risk

The naive backtest overstates the crisis-day loss because it lets exposure run to a level no real book would carry — but it also overstates the recovery, because a real risk limit would have forced deleveraging into the spike, cutting exposure right before the rebound the naive backtest is fully positioned to capture. Both directions of the bias matter, and they don't cancel: the honest simulation shows a shallower drawdown and a slower recovery, which changes the shape of the entire equity curve, not just its worst day.

Without a limit, gross exposure drifts past what any real desk would carry, right before the spike where it matters most.

Encode the real book's position, gross, and VaR limits directly into the simulation loop, checked and enforced at every rebalance, rather than sizing purely off signal conviction and adding a limit as an afterthought — the limits change when exposure gets cut, not just how large it can get.

Risk limits interact with strategy design in a way a post-hoc cap doesn't capture: a real desk forced to de-lever mid-crisis often misses part of the subsequent recovery, so applying the limit only to the drawdown and not to the exposure path afterward overstates the strategy's real-world Sharpe.

Build limits as constraints inside the position-sizing step of the simulation loop, not as a clip applied to P&L after the fact — the sequencing of when a limit binds is what determines the real equity curve.

Position and Risk Limits in Simulation

Worked example: the drawdown a real desk would have been forced into

Related concepts

Practice in interviews

Further reading