Inverse volatility weighting sizes each position by the reciprocal of its volatility:
w_i = (1 / σ_i) / Σ (1 / σ_j)
where σ_i is the trailing volatility of stock i, usually the annualised standard deviation of daily returns over a lookback window of 60 to 252 trading days. A stock that moves half as much as another gets twice the weight.
The reasoning is that equal capital does not mean equal risk. If you put 4 percent into a steady FMCG name and 4 percent into a volatile smallcap, the smallcap contributes several times more to your portfolio’s daily swings. Inverse volatility equalises the standalone risk each position brings, rather than the rupees.
Worked example
Three holdings with annualised volatilities of 18, 30 and 45 percent.
| Stock | Volatility | 1 / σ | Weight |
|---|---|---|---|
| Low-vol large cap | 18% | 0.0556 | 50.0% |
| Mid-vol large cap | 30% | 0.0333 | 30.0% |
| High-vol small cap | 45% | 0.0222 | 20.0% |
This is illustrative arithmetic, not a backtest result. The calm name gets two and a half times the capital of the volatile one, and each position contributes roughly the same standalone risk.
Parameters that change the answer
| Parameter | Typical range | Effect |
|---|---|---|
| Lookback window | 60 to 252 days | Short windows react faster and churn more |
| Return frequency | Daily, weekly | Daily is noisier; weekly smooths gap risk |
| Volatility floor | 5% to 10% annualised | Stops a temporarily frozen stock taking a huge weight |
| Rebalance schedule | Monthly, quarterly | More frequent means more turnover |
Where it breaks
Inverse volatility ignores correlation entirely. It treats two banking stocks that move together as two independent risk sources, so a portfolio of HDFCBANK, ICICIBANK and KOTAKBANK can end up heavily weighted simply because private banks were calm during the lookback, while the concentration in a single sector goes unmeasured. Risk parity fixes this by using the full covariance matrix; inverse volatility is its diagonal-only shortcut, and the two agree only when all pairwise correlations are equal.
Volatility is also autocorrelated but not persistent forever. A stock that was quiet for six months before an accounting disclosure will have carried a large weight into exactly the wrong event. Circuit-limited stocks and thinly traded names post artificially low measured volatility because the price simply does not print, so screen for a minimum traded value before letting the weighting scheme reward stillness.
Finally, low volatility is not the same as low drawdown. A slow, persistent decline registers as calm.