Testing Lead-Lag Prediction Between FX Markets

Summary

The document frames a research question about whether movements in an FX pair’s midprice on a higher-volume exchange can predict subsequent movements on a lower-volume exchange. The proposed target is whether the second venue’s midprice rises within a chosen number of seconds after an increase on the first venue, with a specified statistical confidence. Simple returns are suggested to express movements in a common unit, including when comparing different currency pairs.

The example data suggests a one-second offset between the two price series, but it does not provide a model, statistical test, or evidence that the apparent lead-lag relationship is reliable. A useful analysis would need to define event timing and prediction horizons, account for asynchronous observations and market microstructure noise, and evaluate significance out of sample. The document asks for relevant literature rather than answering its own question, so it offers a research setup rather than a validated predictive method.

Key ideas

• The research question is whether a higher-volume FX venue leads price changes on another venue.
• The proposed prediction target is a future midprice increase within a specified time horizon.
• Simple returns can put price changes for different currency pairs on a comparable scale.
• An apparent delay in sample data alone does not establish predictive power or statistical confidence.

Tags

# Predicting midprice of FX pair increases in exchange Y if it increases in exchange X


# Predicting midprice of FX pair increases in exchange Y if it increases in exchange X












I want to make a predictive model to see if the mid-price of P (FX Pair) increases in Exchange X that P will increase in Exchange Y within n seconds with a certain statistical confidence.

Exchange X has the most trading volume for P and Exchange Y has a much lower volume.

I am using simple returns to make them the same unit in case I want to predict two different FX pairs.

My data looks like this

```
Time, X, Y
00:00:00, 1%, 0%
00:00:01, 0%, 1%
00:00:02, 2%, 0%
00:00:03, 0%, 2%
```

You can visually see that there's a 1 second delay I am interested to know what literature there is that have tackled this problem?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.