Time Aggregation: Why the Chart Clock Changes What You See

Two traders pull up the same stock on the same day and come away describing two different markets. One calls it a quiet session, up a fraction, nothing worth a second look. The other calls it violent: a sharp flush around lunch and a scramble back into the close. Both are right. One’s reading a single daily bar, the other’s reading five-minute candles. What they’re really arguing about is time aggregation, the choice of how finely you slice the trading day before you call the result a return, a volume, or a volatility number.

That choice sounds like plumbing, the sort of thing you set once and forget. It’s closer to a lens. Change the interval and the same sequence of trades produces different returns, different volume profiles, different volatility estimates, and different correlations. Compare two studies built on two different clocks and you can read a difference in sampling as a difference in the market. This is a guide to seeing the clock before you trust the chart.

What time aggregation actually throws away

Start at the bottom. The raw material of any market is a stream of individual transactions, each with a price, a size, and a timestamp. Those prints are irregular. Ten trades can hit in the same second, then nothing for a minute. They carry the bid-ask bounce, the back-and-forth between someone paying the offer and someone hitting the bid, which nudges the last price up and down even when nothing real has changed.

A bar takes a slice of that stream and compresses it to five numbers: open, high, low, close, and volume. A five-minute bar summarises every trade in those five minutes. A daily bar summarises the whole session. A monthly return summarises roughly twenty-one daily bars. Each step up throws information away on purpose, and the higher you climb the more of the path disappears.

The scale of that compression is easy to underrate. When I rebuild one US regular session at one-minute resolution I get 390 bars, one for each minute from the open to the close. Collapse the same session to a daily bar and I’m left with four prices and a volume. Every reversal and every spike between the open and the close now lives or dies inside the high and the low. The daily bar answers a narrower question than the minute series, and answering it cleanly means discarding the route price took to get there.

Why the same session can look calm or violent

Here’s the example I keep on my desk, built with round numbers so the mechanism is naked. A stock closed yesterday at 50.10. Today it opens at 50.00, sells off to 48.20 by half past twelve, then grinds back to close at 50.40. The daily close-to-close return is plus 0.6 percent. Quiet. On that one number, a trader would file the day as uneventful.

Now sample the same day in five-minute bars. The move from 50.00 down to 48.20 is a 3.6 percent drawdown the daily figure never mentions, and the recovery is a second leg of similar size in the other direction. A path that covered more than seven percent of round-trip range collapses into a placid plus 0.6 percent once you sample it once a day. The volatility you’d estimate from close-to-close daily data and the volatility you’d estimate from the intraday path describe the identical trades and still won’t agree.

A lot of readers reach for the wrong conclusion at this point. Shorter bars feel more accurate, closer to the truth, so surely a five-minute volatility estimate is the better one. Not automatically. As you shorten the interval you also sample more bid-ask bounce, and that microstructure chatter inflates the variance you measure with noise that has nothing to do with genuine price discovery. Range-based estimators like historical volatility read the same day differently again, because they lean on the high and the low rather than the close. There’s no single correct number, only a number produced by a sampling rule, and you have to know the rule.

Correlations drift toward zero as the clock speeds up

Aggregation doesn’t only bend volatility. It bends the relationships between instruments. Two stocks that look tightly linked on monthly returns can look almost independent measured minute by minute, and the reason is mechanical. At high frequency, trades in two different names rarely print at the same instant, so when you line their short-interval returns up side by side, half the time one has moved and the other hasn’t caught up yet. Measured correlation falls toward zero as the interval shrinks. The companies didn’t decouple; the clock simply outran the prints.

The trap is treating a correlation as a property of the pair rather than a property of the pair at a horizon. A book built to look diversified on daily co-movement can behave very differently through an intraday shock, when everything gaps together and the comfortable daily correlation is nowhere to be found. A trader reading a correlation table might ask what interval produced it before leaning on the number for risk. The same caution applies to anything downstream of order flow, where a reading taken over seconds tells a different story than the net print for the day.

The overnight gap is a decision

Every aggregation carries a boundary, and the most consequential one in daily data is where you put the overnight. Include it and your daily return runs close-to-close, yesterday’s late print to today’s late print, gap and all. Exclude it and your return runs open-to-close, regular hours only, and the jump between sessions vanishes from the series.

Take the same stock again. If it closed at 50.10, gapped down to open at 48.50 on overnight news, then recovered to 50.40, a close-to-close series records a mild plus 0.6 percent while a regular-hours-only series records a strong intraday climb of nearly four percent off the open. Same tape, two very different return paths, chosen entirely by where the boundary sits. If you’ve ever wondered why two data providers disagree about a single day’s return, this is often the answer. It’s the same question sitting underneath price gaps: the gap is either inside your number or outside it, and that’s a choice you made whether you noticed it or not.

Same market, different clock

The idea to hold onto is that changing the sampling interval isn’t the same act as changing the phenomenon you’re measuring. The stream of trades is fixed. Switching from daily bars to five-minute bars changes the resolution of your instrument, not the behaviour of the stock. Some real structure only shows up at one scale. Intraday mean reversion around a level can be obvious in one-minute data and invisible by Friday’s close, and a slow trend can be plain on the weekly chart and buried inside daily noise.

This cuts against a habit I have to watch in myself: assuming a relationship found at one interval carries to another. A signal that predicts next-day returns in daily data says nothing on its own about the next five minutes. Tools exist to test how returns scale across horizons, because random-walk behaviour at one interval can hide autocorrelation at another. Nassim Taleb made a blunter version of the point about attention rather than statistics: watch a portfolio tick by tick and you drown in noise, check it monthly and the signal comes back. His writing on the balance of noise and signal across time horizons reads well next to any thinking about aggregation.

What to ask before you trust a chart or a study

None of this tells you which interval to use, and it shouldn’t. The useful interval depends on the question. If you want to describe execution quality, seconds and VWAP are the right grain and the daily bar is useless. If you want to describe a multi-month market outcome, daily or weekly bars are the honest scale and the ticks are a distraction. It depends on data quality and the venue too, since thin instruments produce unreliable short bars no matter how badly you want the detail.

So the real work is descriptive, a short list of questions to run before you compare two results or act on someone else’s chart. What clock produced it: ticks, time bars, or something volume-based? Where are the session boundaries, and does the series carry the overnight? What price convention sits underneath, last trade or mid, raw or an adjusted price series back-adjusted for splits and dividends? What rule turned the prints into bars? Two studies are only comparable when those answers match. It’s the same discipline that keeps a backtest honest, which is why look-ahead bias and aggregation choices tend to surface in the same audit.

Read the clock before the candle

Every chart is an answer to a question nobody asked out loud: at what interval, between which boundaries, on what price. The tape of trades underneath is the same whichever way you slice it, and yet the returns, the volatility, and the correlations you pull out are made partly by the slicing. A trader who takes that on board stops treating two numbers as comparable just because they wear the same ticker, and starts asking what clock built each one. That single habit clears out a surprising amount of self-inflicted confusion.

Use it to read honestly across the intervals you already watch. There’s no single true timeframe waiting to be found, only clocks to keep straight. Learn the pattern. Ride the trend. Keep the gains.

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