Our proposed agenda is organized around a simple but provocative idea: order flow matters for exchange-rate determination. The idea is provocative because order flow plays no role in most exchange rate models. For order flow to play no role, however, requires strong assumptions about how market-clearing prices are actually found. On theoretical grounds, these assumptions are difficult to defend. They are difficult to defend on empirical grounds too. There is strong evidence now that order flow plays an important role, a role that traditional models are not designed to capture. By relaxing these strong assumptions, our models provide a vehicle for understanding how price-setters find the clearing price, and what that price is based upon. Empirically, our initial results show that order flow can account for half of daily exchange-rate variation, a far greater share than traditional macro fundamentals such as interest rates. Our project represents a distinctly different research agenda because these results are hard reconcile with either the existing macroeconomic or microstructure approach. Our research agenda borrows liberally from both the macroeconomic and microstructure approaches, but cannot be viewed as encompassed by either. The agenda does not treat exchange rates as decoupled from macro fundamentals, as is typical within the microstructure approach. It is thus firmly anchored in the broader context of asset pricing. Nor does the agenda treat exchange rates as determined in a common-knowledge environment, as is typical within the macroeconomic approach. Departing from common knowledge requires a treatment of information aggregation. We make ample use of tools from the microstructure literature for addressing this aggregation. The project uses two new data sets that cover activity in the spot foreign exchange market over a four-month period, May 1 to August 31, 1996. One data set contains time-stamped, tic-by-tic data on actual transactions for nine currencies. Three features of this data set are noteworthy. First, it provides transaction information for the whole interbank market over the full 24-hour trading day whereas earlier data sets typically cover one dealer, and only part of the day. Second, individual participants do not observe these market-wide data in real time, which allows us to get inside the "black box" of price determination. Third, multiple currencies and a relatively long time span allow us to address price determination from more of an asset-pricing perspective. The second of our two data sets is much less extensive than the first, but covers a segment of the market that has, until now, remained beyond the reach of empiricists - trading by the public. Until now, all transaction data in foreign exchange have come from interbank trading. This second data set contains all the public trades in DM/$ received by a large bank over the same four-month period covered by our first data set. That bank is number one worldwide in terms of foreign exchange market share. This data set allows us, for the first time, to connect dealers' trading and prices to the underlying sources of demand in the economy. The questions we will address with the proposed research include: What share of exchange rate variation is attributable to order flow? What components of order flow convey the most information? Are the price effects of macroeconomic announcements uncorrelated with order flow? Does exchange rate determination work from macroeconomic fundamentals, to order flow, to price? How does order flow link to learning, herding, and noise trading in foreign exchange markets? Does feedback trading account for part of the correlation between order flow and price, e.g., Friedman's stabilizing speculators? When is order flow from central banks likely to have the most price impact?
