The determination of foreign exchange rates has long been an important but empirically challenging topic in international economics. Models that aim to relate foreign exchange (FX) rates directly to macroeconomic fundamentals tend to have disappointing out-of-sample and forecasting properties (Meese and Rogoff, 1983; Cheung et al., 2005, 2017). Instead of contin uing to try to model exchange rates primarily as variables that equilibrate exports and imports of goods and services across countries, economists turned to modeling exchange rates as the relative prices of assets denominated in various currencies, with trade flows and financial flows both responding to asset demands and supplies. Moreover, beginning with the work of Glosten and Milgrom (1985), Kyle (1985), and Admati and Pfleiderer (1988), it has been argued that in order to understand the price formation in a financial market more fully, one needs to distinguish between the private and public information sets of market participants. Because published data about macroeconomic fundamentals constitute public information, models that rely exclusively on macroeconomic fundamentals are bound to disappoint as the miss the influence investors’ private information. There is by now broad agreement among researchers that FX order flow, defined as the difference between buyerinitiated and seller-initiated transaction volumes in the FX market, helps explain exchange rates because it conveys investors’ private information. Market-relevant private information can pertain to future changes in aggregate economic activity and inflation as well as to firm-level cash flows and discount rates. FX order flow that is driven by market participants’ private information (‘‘informed trades”) should have, in principle, a permanent effect on exchange rates. In contrast, FX order flow that is not driven by private information (‘‘noise trades”) should have at most a temporary impact on exchange rates.