Module 14 of 14 · Running it for real

Risk overlays and hard limits

Volatility targeting manages risk market by market and on average over time. On any particular day the portfolio can still be carrying far more risk than intended: forecasts may all be strong at once, positions may line up with correlations in the worst possible way, or quiet markets may have pushed leverage very high. And an automated system has a different kind of risk too — the risk that something breaks and it trades when it shouldn't.

Any risk control needs four parts: something you measure, a level that triggers action, the action, and when to undo it. A portfolio risk overlay does all four systematically. Measure several things that scale with position size:

  • Expected portfolio risk from the current positions and covariance matrix.
  • Jump risk — the same calculation with every market's volatility set to an extreme percentile of its own history.
  • Correlation-shock risk — every correlation set to whichever of ±1 hurts you most.
  • Gross leverage — total notional ÷ capital.

Set each limit near a high percentile of its own backtest history, compute limit ÷ value for each, and multiply all positions by the smallest ratio when it is below 1. When all measures are back inside their limits, the multiplier returns to 1 by itself.

Then add hard limits per market: a maximum position (what the sizing formula can produce at the forecast cap, plus headroom, and no more than a leverage cap allows) and a maximum number of contracts traded per day. These are not about market risk; they are guard rails against bugs, bad data and broken state.

🎛 Risk overlay lab

1 · Portfolio risk overlay

24%
30%
55%
75%
70%
65%
9×
12×

Expected portfolio risk

1.00

24% vs 30%

Jump risk (vols at extremes)

1.00

55% vs 75%

Correlation shock risk

0.93

70% vs 65%

Gross leverage

1.00

9× vs 12×

Applied to all positions

0.93×

binding: correlation shock risk

2 · Hard limits for one instrument

250000
10%
2
4000
10
18%
2×

Max at forecast cap

2.8

+50% headroom → 5

Max at leverage cap

12

contracts

Position limit

5

contracts, whichever is lower

Daily trade limit

2

contracts / day

Overlay: each measure gets a multiplier = min(1, limit ÷ today's value); all positions are multiplied by the smallest one, so the overlay can only ever reduce risk. Calibrate each limit from a high percentile of your own backtest so it bites rarely. Limits: the position limit is the smaller of the maximum position your sizing formula can produce (forecast at its cap of 20, plus 50% headroom) and the most contracts allowed by a per-instrument leverage cap; the daily trade limit here is a third of the position limit. Values are illustrative; risk target 20%.

Limits are cheap insurance

A well-calibrated overlay is idle most of the time, so it costs almost nothing in performance, and it bites in precisely the situations a volatility estimate misses. Trade limits cost even less: a daily limit of a few contracts is invisible to a strategy that trades a few times a week, and it caps the damage when a restart, a stale price, or a mis-specified contract multiplier makes the system try to trade something absurd.

Also watch margin. A volatility-targeted futures portfolio normally uses well under half its capital as margin; if usage climbs toward that, cut all positions proportionally rather than dumping the single biggest margin user. And keep the account fully capitalised — free cash is what absorbs variation margin on a bad day.

Things to try

  • • With the defaults, the correlation-shock measure binds and every position is cut to about 93%. Raise its limit until no multiplier bites.
  • • In the limits section, raise the instrument weight to 40% at $250,000: the leverage cap becomes the binding limit, telling you that market is too large a share of the account.
  • • Turn the micro into a full-size contract (multiplier 100) and set capital to $50,000: the leverage cap allows zero contracts. If the limit is zero, the market should not be in the portfolio at all.

Original interactive lesson. Labs run on a seeded synthetic market in your browser, built to illustrate mechanics, not to predict any real market. Contract figures are illustrative; check exchange specifications. Educational analysis, not investment advice.

How are we doing?