Home › Blog › Quant Research
How Late Is Realized Volatility? Implied vs Realized Vol Across Markets, 2013–2026
When a shock hits, option prices reprice within minutes. Realized volatility has to wait for prices to actually move, and then it has to forget the move. I measured how large that gap is across stocks, credit, Treasuries, oil and gold over 13 years. The short answer: big surprises reach realized volatility the same day, but anticipated risks show up one to three weeks late, and the all-clear comes about two weeks late.
Why the question matters
Implied volatility, such as the VIX, is a price: what traders pay today for protection over the coming month. Realized volatility is a measurement: how much prices actually moved over a recent window. Many risk dashboards, cross-asset stress monitors and position-sizing rules rely on realized volatility because it can be computed for any asset with a price history, while implied volatility exists only where a liquid options market does. If realized volatility is a few days late, that is a detail. If it is weeks late, a monitor built on it will report "calm" for much of a stress episode and "stressed" long after the market has moved on.
Data and method
The test uses daily closes from March 2013 to October 2026, starting once every series had a full year of history. For implied volatility I used the Cboe volatility indices that have traded continuously over the period, and for realized volatility the matching ETF or index:
| Market | Implied volatility (Cboe) | Realized volatility from |
|---|---|---|
| US equities | VIX | S&P 500 (SPX) |
| Developed ex-US equities | VXEFA | EFA |
| Emerging-market equities | VXEEM | EEM |
| Long-dated Treasuries | VXTLT | TLT |
| High-yield credit | VIXHY | HYG |
| Investment-grade credit | VIXIG | LQD relative to IEF |
| Crude oil | OVX | USO |
| Gold | GVZ | GLD |
Realized volatility is an exponentially weighted moving average of squared daily returns with a decay factor of 0.94, the classic RiskMetrics setting and a common default in risk systems. To put markets with very different volatility levels on one scale, each series is scored by where today's reading ranks within its own trailing one-month window, from 0 (the highest reading of the window, most stressed) to 100 (the lowest). The composite is the average of the eight scores, built once from implied and once from realized volatility. Some tests below also use four more matched pairs with shorter histories (Nasdaq-100, Russell 2000, Dow and Brazil), for twelve markets in total.
Big shocks: priced the same day
Start with the cases that should favour realized volatility: the days with the largest one-day jumps in implied volatility, the top 2% in each of the twelve markets, 689 events in all. On those days the implied volatility score fell by 39 points on average and the realized volatility score by 33. Realized volatility captured about 85% of the move on the same day, because a day that frightens option traders usually comes with a large price move.
The difference is what happens next. Ten days later the implied score was back within 1 point of where it started. The realized score was still 18 points lower. Option traders mark the risk down as soon as it passes; the realized estimate keeps the large return in its memory for weeks.
Getting into stress, and getting out
Next, the composite. I marked "stress" as the composite falling into its own bottom tenth and compared the first stressed day on the implied side, 74 entries over the period, with the first stressed day on the realized side.
- Entry: realized volatility followed a median 5 trading days later. Only 35% of entries were matched within two days, and in 18% of cases the realized composite did not reach stress at all within 40 days.
- Exit: the realized composite climbed back above its median a median 11 trading days after the implied composite.
- Day to day: the cross-correlation of five-day changes peaks with realized volatility two days behind implied.
The smaller the warning, the longer the lag
Splitting the market-level stress entries (1,488 across twelve matched pairs) by how sharply implied volatility rose over the five days before shows that the lag is mostly a feature of modest moves. When implied volatility jumps by 40% or more, realized volatility is usually there the same day. When implied volatility rises by single digits, the median lag is eight trading days and most entries are not matched within a week.
| Five-day rise in implied vol | Entries | Median entry lag | Matched within 1 day | Within 5 days | Median recovery lag |
|---|---|---|---|---|---|
| Smallest half (about +7%) | 744 | 8 days | 26% | 40% | 10 days |
| 50th–80th percentile (+16%) | 446 | 4 days | 35% | 51% | 7 days |
| 80th–95th percentile (+25%) | 223 | 2 days | 44% | 58% | 5 days |
| Largest 5% (+41%) | 75 | 0 days | 55% | 67% | 5.5 days |
Twelve years of events
The pattern is clearest event by event. For each episode I counted trading days from the first day the implied composite entered stress to the first day the realized composite did.
Two groups separate cleanly. Sudden shocks such as the COVID-19 crash, the April 2025 tariff announcement, the 2024 yen carry unwind and the SVB failure hit prices and option markets together, and realized volatility was at most four days behind. Anticipated risks were different. Option prices started rising ahead of the Brexit vote, implied stress began on June 15, 2016, while realized volatility did not get there until the result on June 24. Before the invasion of Ukraine, implied volatility rose for weeks while prices barely moved, and realized volatility arrived 14 trading days later.
Why realized volatility is late
Three mechanisms explain almost all of the gap.
- Anticipated risk is not in prices yet. A vote, a central bank decision or a military build-up is priced in options before anything happens. A measure built from past returns cannot see it, whatever the estimator.
- Memory. An exponentially weighted average with a 0.94 decay keeps a large return in the estimate for weeks. Implied volatility drops as soon as traders believe the danger has passed.
- Direction. Equity and credit implied volatility respond to the direction of the market, not just the size of the move. For the S&P 500, the daily change in log VIX has a correlation of about −0.76 with the index's return: a one-standard-deviation down day lifts VIX by about 12%, and an up day of the same size lowers it by about 9%. Realized volatility treats both days the same, so the sharp rebound days that end a selloff push realized volatility up while implied volatility falls. That is the main reason recoveries lag.
Can a faster estimator close the gap?
Partly. Shortening the memory of the estimator speeds it up at the cost of a noisier reading. The table compares estimators on the eight-market composite against the implied composite.
| Estimator | Entry lag | Within 2 days | Never followed | Recovery lag | Correlation with implied | Daily noise |
|---|---|---|---|---|---|---|
| Implied volatility (benchmark) | 0 | 100% | 0% | 0 | 1.00 | 10.5 |
| EWMA, decay 0.94 (RiskMetrics) | 5 | 35% | 18% | 11 | 0.48 | 6.6 |
| EWMA, decay 0.90 | 4 | 43% | 11% | 6.5 | 0.53 | 7.2 |
| EWMA, decay 0.85 | 3 | 46% | 7% | 4 | 0.55 | 7.9 |
| EWMA, decay 0.80 | 1 | 51% | 4% | 3 | 0.55 | 8.5 |
| 21-day simple realized vol | 7 | 24% | 30% | 12.5 | 0.34 | 6.4 |
| High-low (OHLC) EWMA, decay 0.94 | 4 | 30% | 27% | 10 | 0.52 | 6.3 |
| High-low (OHLC) EWMA, decay 0.85 | 3 | 42% | 14% | 3.5 | 0.61 | 7.7 |
Lags are medians in trading days. Daily noise is the typical one-day change in the composite score, in points.
Moving from a 0.94 to a 0.85 decay cuts the recovery lag from 11 to 4 days and the share of missed stress episodes from 18% to 7%, and the result held in both halves of the sample (2013–2019 and 2020–2026). Even the fastest estimators are less noisy day to day than implied volatility itself, so the usual worry about a jumpy signal is smaller than it looks. The popular 21-day window is the slowest of all. Using daily highs and lows improves the correlation with implied volatility but does about as much for timing as simply shortening the memory. None of this fixes the first mechanism: no estimator built from past prices can see an anticipated risk.
A calibration trap
There is a second problem that has nothing to do with timing. Because a composite of realized volatilities moves more smoothly, it spends less time at the extremes. Over the full sample the implied composite scored below 20 on the 0–100 scale on 14.2% of days, the realized composite built from the same markets on only 6.9%. A dashboard that uses the same fixed thresholds for both will raise a realized-volatility stress alarm about half as often, and a broader realized-only basket even less. Changing the decay factor does not fix it (6.5–8.4% across settings). The remedy is to set thresholds from each composite's own distribution, for example calling the bottom tenth of its history "stress", rather than reusing levels designed for implied volatility.
Key takeaways
- For large surprises, realized volatility is not late: it captures about 85% of the implied move on the day.
- For anticipated risks it is one to three weeks late, and no realized estimator can fix that. Where an options market exists, use implied volatility.
- Recoveries lag by about two weeks with a standard 0.94 decay, mostly because realized volatility ignores the direction of the market.
- A 0.85 decay roughly halves the lag into stress and cuts the recovery lag from 11 to 4 days, with modest extra noise. A 21-day window is the slowest choice.
- Calibrate thresholds for realized-volatility composites separately. Thresholds set for implied volatility will miss many stress days.
To see implied and realized volatility side by side for the S&P 500 today, the VIX term structure page charts VIX against 21-day realized volatility, and every Daily Market Review tracks the gap after the close. For the theory behind implied volatility, see Implied Volatility and the Smile.
Data: Cboe daily index closes and ETF closes, March 2013 to October 2026. Results are historical and descriptive; nothing here is investment advice.