Abstract

Abstract Currency economists are puzzled by the relatively high frequency of massive, abrupt exchange-rate changes, such as dollar-yen's 11 percent decline on October 7, 1998. This paper provides evidence that such moves may be partly caused by stop-loss orders, which create rapid, self-reinforcing exchange rate movements or "price cascades." The central hypothesis, which is taken from prominent theoretical research in finance, suggests that positive feedback trading, of which stop-loss trading is an example, can cause discontinuities manifested as price cascades. The paper's empirical analysis, which uses high-frequency exchange rates and order records from a major foreign exchange bank, indicates that stop-loss orders propagate trends and are sometimes triggered in waves, contributing to price cascades. Since price cascades are inconsistent with standard structural exchange rate models, they may be a factor behind the "exchange rate disconnect" problem. Stop-loss propagated price cascades could also be a mechanism through which exchange rates shift among multiple equilibria. Price cascades may help explain the well-known "fat tails" of the distribution of exchange-rate returns. The paper also provides evidence that exchange rates respond to noninformative order flow. (Key words: exchange rates, currency market microstructure, order flow, high-frequency, stop-loss, portfolio insurance, information) (JEL codes: F1, G3.) STOP-LOSS ORDERS AND PRICE CASCADES IN CURRENCY MARKETS On October 7, 1998, the dollar-yen exchange rate fell 11 percent. On March 7, 2002, the rate dropped over 3 percent. These moves, which dwarf the 0.7 percent standard deviation of daily returns in dollar-yen since 1990, are symptomatic of a broader phenomenon: the wellknown "fat tails" of exchange rate returns. Since 1990, dollar-yen returns above four standard deviations have occurred 85 times more frequently than predicted by the normal distribution; under a normal distribution, daily returns above 3 percent would occur fewer than once every 100 years. 1 Dramatic exchange rate moves are as puzzling to economists as they are disruptive to markets. According to standard exchange rate models, the main force behind them must be news. Yet Cai et al. (2002) find that the arrival of news was of only "secondary importance" for extraordinary yen volatility throughout 1998. Likewise, Evans (2001) finds that "public news is rarely the predominant source of exchange rate movements over any horizon" (p. 1, italics in the original). Of greater importance, these authors suggest, is order flow. Researchers have also turned to order flow to account for the stock market crash of 1987, another dramatic price move that cannot be explained by news Because these schemes involve price contingent, positive feedback trading, they can contribute to market discontinuities-that is, crashes (Genotte and Leland (1990); Easley and O'Hara (1991); Jacklin et al. 2 In the most commonly cited scenario, a price decline from any source triggers portfolio insurance sales, causing further price declines, which trigger additional portfolio insurance sales, etc. This type of self-reinforcing price dynamic will be referred to here as a "price cascade." Since information about portfolio insurance and stop-loss orders is not public, rational trading by uninformed agents could intensify such a price cascade (Genotte and Leland (1990); Easley and O'Hara (1991)). 1 The normal distribution used for comparison here has the same mean and standard deviation as actual returns. This paper attempts to provide statistical evidence of this phenomenon in currency markets, where stop-loss orders are commonplace. 4 To identify when such orders are executed, I turn to evidence that stop-loss orders cluster in predictable ways near round numbers (Osler The statistical methodology is a variant of the bootstrap (Efron (1979(Efron ( ), (1982). The analysis first shows that exchange rates tend to move rapidly after reaching levels where stop-loss orders cluster. 5 This indicates that a trend can be prolonged by the execution of stop-loss orders triggered by that trend, consistent with the paper's main hypothesis. However, this result need not indicate that the execution of stop-loss orders at one level sometimes propels rates to new levels, triggering more stop-loss orders, as described by Deutsche Bank. To evaluate whether stop-loss orders are actually triggered in waves, the paper undertakes two tests in which exchange rate behavior after reaching points where stop-loss 3 There is ample additional evidence that market participants take the contribution of stop-loss orders to large price moves for granted. The publication "Currency Network Daily Briefs" reported that "stops were triggered," or some equivalent, on at least 16 of the approximately 190 trading days from December 2000 through August 2001, or at least once every two weeks. Market lore suggests that participants sometimes intentionally trigger a series of stoploss orders, and the activity has its own name: "running the stops." Major movements when stop-loss orders are triggered are characterized by market participants as extremely rapid and "gappy," meaning that individual prices are skipped as the rate moves from one price level to another. 4 Note that currency stop-loss orders include buy orders triggered by rate increases, as well as sell orders triggered by rate decreases.