Non-Synchronous Trading
In textbooks, we assume all prices are recorded at exactly t.
In reality, Asset A might trade at 10:00:01 and Asset B might trade at 10:00:59. This timing mismatch is called Non-Synchronous Trading.
The Problem
When you calculate the Correlation between two assets using high-frequency data, Non-Synchronous trading causes a Bias.
Example
- Reliance (Liquid): Trades every second. (Reacts to market news instantly).
- SmallCap Co (Illiquid): Trades once every 10 minutes. (Reacts to market news with a "lag").
If the market crashes at 10:05:
- Reliance drops at 10:05.
- SmallCap doesn't trade until 10:10. So at 10:05, its price looks "unchanged".
Result: If you calculate correlation at 10:05, it looks like they are Uncorrelated. This is false. They ARE correlated, but there is a lag.
The Epps Effect
This phenomenon is known as the Epps Effect: "Correlations between stock returns tend to decrease as the sampling frequency increases."
- Daily Correlation: 0.8 (Strong)
- 1-Minute Correlation: 0.1 (Weak - due to non-synchronous bias)
Solutions / Adjustments
- Hayashi-Yoshida Estimator: A statistical fix to calculate "true" correlation even with asynchronous data.
- Previous Tick Interpolation: For the illiquid asset, carry forward the last traded price to fill the gaps (fill-forward).
Takeaway: Never blindly calculate correlation on raw tick data. You will drastically underestimate the relationship between assets.
Loading quiz…