Variance Ratio: Is Your Market Trending or Reverting?

You’ve got a chart in front of you and a decision to make before you touch a single indicator: reach for a breakout tool that assumes the move keeps running, or a mean-reversion oscillator that assumes price snaps back. Most of the time that choice gets made on gut. The instrument feels trendy, or it feels choppy, so you pick the tool that matches the feeling. The trouble is that the feeling and the data disagree more often than anyone likes to admit.

The variance ratio gives that decision a number. It’s one figure that tells you whether an instrument’s returns are behaving like a trend, like a spring, or like a coin flip, at the exact time scale you care about. And you can work out a rough one from a price series with nothing more than a spreadsheet.

What the variance ratio actually compares

The calculation lines up two quantities that should be equal if price movement is random. The first is the variance of k-period returns. The second is k times the variance of one-period returns. Divide the first by the second and you’ve got the variance ratio.

Take k equal to 5, which turns daily returns into weekly ones. You measure the variance of your weekly returns, then you measure the variance of your daily returns and multiply it by 5. The variance ratio is the weekly variance divided by that scaled-up daily variance. If the two match, the number lands on 1.0. That single comparison is the whole idea, and every interpretation below hangs off how far the reading sits from 1.0.

The reason this particular comparison is worth the effort comes from how variance behaves when price changes are independent of each other. When today tells you nothing about tomorrow, variance grows in a straight line with time. Five days of independent moves carries exactly five times the variance of one day. So the honest benchmark for weekly variance is five times the daily figure, and the variance ratio measures how badly reality misses that benchmark.

Why a random walk sits at exactly 1.0

Picture a market where each day’s return is drawn fresh, with no memory of the day before. Some days up, some days down, no pattern connecting them. Under that pure random walk the variance of a five-day return is precisely five times the variance of a one-day return, and the variance ratio is 1.0 by construction. This is the null case, the reading you get when there’s no exploitable structure in the returns at that horizon.

Now let the days start to lean on each other. When a positive return tends to be followed by another positive one, moves in the same direction stack up. Five-day windows travel farther than five independent days would, so the variance of the longer return swells past five times the daily variance and the ratio climbs above 1.0. That’s the momentum case, the statistical signature of a trend. It’s the same persistence a trend intensity index tries to capture from a different angle.

Flip the autocorrelation and the picture inverts. When a positive return tends to be followed by a negative one, the moves partly cancel inside each window. Five-day returns end up smaller than five independent days would produce, the longer-horizon variance falls short of the benchmark, and the ratio drops below 1.0. That’s the mean-reversion case.

One misread is worth flagging from the start: the variance ratio measures persistence, not direction. A reading of 1.3 tells you moves cluster and follow through. It says nothing about whether price is climbing or falling. A market can grind lower for months with a variance ratio well above 1.0, because a persistent downtrend is still persistence. If you treat a high reading as a bullish signal, you’ve asked the number a question it never answers.

Calculating one by hand

None of this needs a statistics package. The steps are plain enough to run in a spreadsheet over a coffee.

• Build the daily return series from your closes, each day as a percentage change on the last.
• Compute the variance of that daily series.
• Build the weekly return series by compounding each block of 5 daily returns.
• Compute the variance of the weekly series.
• Divide the weekly variance by 5 times the daily variance.
• Read the result against 1.0.

A worked example makes the arithmetic concrete. Say your daily returns carry a standard deviation of 1.5 percent. Squaring that gives a daily variance of 0.000225. Five times that figure is 0.001125, and that’s the weekly variance you’d expect from a random walk. Now suppose the measured weekly returns come in with a standard deviation of 3.7 percent, for a weekly variance of 0.001369. Divide 0.001369 by 0.001125 and the variance ratio is 1.22.

A reading of 1.22 says weekly moves are running about twenty-two percent wider than independence would predict. At that horizon the instrument is carrying real momentum, and a trend-continuation tool has a statistical footing under it rather than a hunch. What I look for first when I run this is exactly that gap between the measured weekly variance and the five-times-daily benchmark, because the size of the gap is the strength of the signal.

The time scale changes the answer

The same instrument will usually hand you a different variance ratio at daily, weekly, and monthly return horizons. This is the part that trips people up. There’s no such thing as the variance ratio of a stock, only the variance ratio at a stated scale. The horizon where the reading pushes above 1.0 is the scale where momentum tools have their strongest foundation. The horizon where it falls below 1.0 is where mean-reversion tools have theirs.

That single fact resolves an argument traders have with themselves constantly. A classic empirical pattern is a variance ratio comfortably above 1.0 at the weekly horizon and below 1.0 at the daily horizon. It means the instrument trends week to week while overreacting and snapping back day to day. Under that reading a daily RSI oscillator flagging short-term reversions and a weekly momentum filter riding the trend aren’t contradicting each other at all. They’re each correct at the scale they measure, which is the whole premise behind multi-timeframe analysis.

It’s also why the great trend followers work on the horizon they do. When a trader like Ed Seykota holds a position for weeks or months, he’s operating at the scale where return persistence is strongest and the variance ratio sits highest, not acting on stubbornness. Shorten the horizon far enough and that same instrument may cross into reversion, where the trend-following logic quietly stops paying.

Where the reading lies to you

The variance ratio has a weakness you have to respect, and it’s the one most people skip. Estimated from a short window, the number is noisy, and its confidence interval is wide. A variance ratio of 1.08 pulled from 60 days of data isn’t convincingly different from 1.0 by most statistical tests. Read as a trend signal, it’s close to reading tea leaves.

The reliability grows with the sample. Hundreds of observations tighten the estimate; a couple of months of data leaves it loose enough to mislead. The other place to lean is the tails. A reading clearly above 1.2 or clearly below 0.8 carries far more weight than anything hovering near 1.0, because the extremes are harder for small-sample noise to manufacture. The discipline I follow is simple: trust the tails on long samples, treat the middle ground as undecided, and never dress up a 1.06 as evidence of a trend.

The second trap is the one from earlier, worth repeating because it’s so common. People report a single variance ratio as if it were a fixed property of the instrument. It’s a property of the instrument and the horizon together. Quote the number without the time scale and you’ve said almost nothing. The formal statistical version of this tool, the variance ratio test developed by Lo and MacKinlay to probe whether markets follow a random walk, was built around exactly this horizon dependence, and recent research has extended the same framework to separate long-horizon return memory from volatility memory across multiple scales.

Fitting it next to the tools you already use

If you already run trending-versus-ranging indicators, the variance ratio is the statistically grounded member of that same family. The choppiness index and the vertical horizontal filter both answer the trend-or-range question from the geometry of the price path. The variance ratio answers it from the autocorrelation of the returns, which is the same language academics use to define what trending and mean-reverting actually mean in the data.

The practical use is a pre-trade check. Before you decide whether an instrument suits a momentum entry or a reversion oscillator, run a quick variance ratio at the horizon you plan to trade. Above 1.0 on a decent sample, the momentum tool has a data-based reason to be there. Below 1.0, the reversion tool does. Near 1.0, you’ve got your answer too, and it’s that neither edge is present right now and the honest move is to wait. That’s a firmer footing than starting from an assumption about the regime and hunting for an indicator that agrees with it.

Run the number before you pick the tool

The variance ratio won’t predict the next move, and it doesn’t try to. What it does is replace a guess about the current regime with a measurement, at a scale you choose, from data you already have. It tells you when momentum has a foundation and when mean reversion does, and it’s honest enough to tell you when neither does. Calculate it, respect the sample size, read the tails, and let the number decide which half of your toolkit comes off the shelf. Learn the pattern. Ride the trend. Keep the gains.

Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.