Stale Prices and Nonsynchronous Trading

A liquid stock like a large index constituent might trade dozens of times in the final minute before the market closes. A thinly-traded small-cap might not trade at all in that same minute — its "closing price" is really the price of its last trade, which could have happened ten or twenty minutes earlier. Both are recorded in the database with a 4:00pm timestamp, but they don't represent the same moment in market time. The small-cap's price is stale: it hasn't updated to reflect anything that happened to the broader market in the intervening minutes, even though the timestamp implies it has.

Why this quietly breaks statistics

This staleness doesn't just add noise — it introduces a systematic bias. When you compute the correlation between the liquid stock and the stale small-cap using their timestamped closing prices, the small-cap's price is effectively reacting to news with a lag, because its "close" is really a snapshot from earlier. That lag makes the measured correlation between the two lower than the true, contemporaneous correlation, because part of the small-cap's reaction to any shared market move hasn't happened yet by the time its stale price was recorded. This is called the nonsynchronous trading problem, and it's a well-documented reason why less liquid assets tend to show artificially low correlations and artificially low measured volatility in daily data, and why cross-sectional momentum or lead-lag signals built from illiquid names can pick up a spurious pattern that's really just an artifact of trading frequency, not genuine predictability.

A concrete case

Suppose a strategy looks for cases where a liquid ETF moves and a basket of small, thinly-traded holdings "lags" it by a day, then trades the laggards expecting them to catch up. Some of that apparent lag is real information diffusion. But some of it is nothing more than nonsynchronous trading: the small stocks' recorded prices are stale relative to the ETF's, so of course they appear to lag — their last trade simply happened earlier in the day. A backtest that doesn't account for this can look like it found a tradable lead-lag effect when it has actually just rediscovered the mechanical fact that illiquid names trade less often.

What this means in practice

Before trusting a correlation, lead-lag, or beta estimate involving instruments of very different liquidity, it's worth checking how often each one actually trades within the sampling window. Using longer aggregation windows (weekly instead of daily returns), or estimators specifically designed to correct for nonsynchronous trading, reduces the bias, though it never fully eliminates it for genuinely illiquid names.