Testing Whether Positioning Data Leads Currency Prices

Summary

The document considers whether currency positioning data precedes price movements. It describes applying Granger causality tests to paired positioning and price series, with several lags, and advises interpreting the p-values against the null hypothesis of no lead-lag relationship. The questioner supplies no reported test results, so the exchange does not establish that positioning actually predicts prices.

The answer also suggests making both series stationary and examining lagged correlations, transfer entropy, and Lévy area as complementary measures. These approaches rely on different assumptions and capture different kinds of dependence, so they should be treated as diagnostic alternatives rather than interchangeable proof of predictive value. The discussion does not explain implementation, multiple-testing concerns, or out-of-sample validation; evidence of a statistical lead would still need careful assessment before drawing a trading conclusion.

Key ideas

• Granger tests assess whether past positioning values add information about currency prices.
• The p-values are used to evaluate the null hypothesis of no lead-lag effect.
• Stationarity should be considered before testing relationships between the series.
• Lagged correlation, transfer entropy, and Lévy area are proposed as additional diagnostics.
• The exchange provides no empirical results or out-of-sample evidence.

Tags

# Lead/Lag statistical test in Python


# Lead/Lag statistical test in Python












I have two series of data, one showing positioning data and one showing price data. Generically the format is similar to the below:

Price:

| Date | EUR | JPY | NOK | SEK |
| Date | price | price | price | price |
| Date | price | price | price | price |

Positioning:

| Date | EUR | JPY | NOK | SEK |
| Date | positioning | positioning | positioning | positioning |
| Date | positioning | positioning | positioning | positioning |

what i have done so far is the below, but i'm not sure how to interpret these results or if there are better tests to show whether the positioning data leads the price data

```
for currency in currencies:
    print(f'\nGranger Causality Test for {currency}:')
    test_result = grangercausalitytests(pd.concat([positioning_data[currency], price_data[currency]], axis=1).dropna(), maxlag=4)
```

## Answer by Nicolás Zanni (score 2)

https://quant.stackexchange.com/a/81154

For the granger causality, check the p-values and determine if the null hypothesis( no lead lag effect) is rejected or not.

I would try to make both time series statationary and see more metrics like:

- Correlation with lags

- Transfer Entropy

- Levy Area (Paper)

Each metric has its own assumptions, but they will give you enough to see if there exists a lead lag effect besides granger causality.

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.