High-frequency information content in end-user foreign exchange order flows

Abstract

This article considers the impact of foreign exchange (FX) order flows on contemporaneous and future stock market returns using a new database of customer order flows in the euro-dollar exchange rate market as seen by a leading European bank. We do not find clear contemporaneous relationships between FX order flows and stock market changes at high frequencies, but FX flows do appear to have significant power to forecast stock index returns over 1–30 min horizons, after controlling for lagged exchange rate and stock market returns. The effects of order flows from financial customers on future stock market changes are negative, while the effects of corporate orders are positive. The latter results are consistent with the premise that corporate order flows contain dispersed, passively acquired information about fundamentals. Thus, purchases of the dollar by corporate customers represent good news about the state of the US economy. Importantly, though, there also appears to be extra information in corporate flows which is directly relevant to equity prices over and above the impact derived from stock prices reacting to (predicted) exchange rate changes. Our findings suggest that financial customer flows only affect stock prices through their impact on the value of the dollar.

Keywords:

• foreign exchange
• order flow
• equity returns
• forecasting

JEL Classifications: :

• F31
• G15

Acknowledgements

We thank Richard Payne, Geoffrey Kendrick, Lucio Sarno, and an anonymous referee for comments, and the anonymous bank for data provision and several explanatory conversations. All errors are our own.

Notes

Osler (2008) surveys this literature in detail.

We accept that it is somewhat surprising that it is priced as rapidly as our results suggest. However, our results are limited to very short horizons by data limitations and while they suggest that there is some very high-frequency forecasting power from FX flows to equity returns, they do not rule out the possibility of greater forecasting power over longer horizons. We are investigating this possibility in a separate paper (Marsh and Miao 2010).

These two papers explain equity returns using FX order flow. A third paper, Albuquerque, Francisco and Marques (2008), test the relation in the opposite direction, considering the effect of order flows in stock markets on exchange rates.

Trades with inter-dealer network banks initiated by the data supplier are unfortunately not available.

Below, we forecast stock returns using lagged FX returns, lagged stock returns, and lagged FX order flow. For symmetry, we should include lagged stock returns in equation (2). However, lagged US stock returns are only available for a small portion of the trading day, which dramatically reduces our sample.

We thank the referee for clarifying our thinking on this, and other, issues in the article.

The only exception to this is in the very short horizon where both of the end-user customer flows show positive point estimates which subsequently decline. However, none of these estimates is significant.

Lyons (1998) suggests an inventory half-life from an active dealer of around 10 min. However, the move to electronic inter-bank dealer systems has improved dealers’ opportunities to pass inventory into the market faster and cheaper, so this 10 min level might be seen as an upper limit. Discussions with bankers suggest that inventories are now managed over 1–2 min horizons.

We note that this is not the case for inter-bank flows, however, which retain negative and significant forecasting power even after accounting for future exchange rate movements.