Conventional macroeconomic theories of exchange rate determination have received only limited support from empirical studies, and there is now broad agreement that `fundamental' macroeconomic variables play relatively little role in determining the short-run movements of exchange rates. An improved understanding of exchange rate dynamics requires a closer examination of the functioning of foreign exchange markets. A joint conference with the National Bureau of Economic Research (NBER) and the Banca d'Italia, on `Microstructure of Foreign Exchange Markets', held at Perugia on 1–2 July, considered how the theories of `market microstructure' that have been developed to analyse equity markets may contribute to explanations of exchange rate behaviour. The conference was organized by Jeffrey Frankel, Senior Fellow at the Institute for International Economics, Washington DC, Professor of Economics at the University of California, and NBER Director for International Finance and Macroeconomics, Giampaolo Galli, Co-Director of the Research Department of the Banca d'Italia, and Alberto Giovannini, member of the Italian Treasury Ministry's Council of Experts, Jerome A Chazen Professor of International Business at Columbia University and Co-Director of CEPR's International Macroeconomics programme. Papers presented addressed both typical microstructure issues – such as the relationships between trading volumes and risk and between information and trading intensities – and those more specifically pertinent to exchange rates, such as the behaviour of market participants during the EMS crises of 1992–3, the mechanics of speculative attacks, and the effectiveness of the capital controls designed to prevent them.


In `Exchange Rate Economics: What's Wrong with the Conventional Macro Approach?', written with Robert Flood, Mark Taylor (University of Liverpool and CEPR) surveyed the traditional literature in order to highlight the remaining theoretical gaps and briefly document the poor empirical performance of the purchasing power parity (PPP) theory, flexible price and sticky price (overshooting) monetary models, and the portfolio balance and equilibrium models. Nevertheless, their estimates using pooled annual data for 21 industrialized countries from the floating-rate period revealed that the explanatory power of the fundamentals – although very poor over even a one-year horizon – was considerably greater for data averaged over periods of five, ten or twenty years, in applications of both the PPP theory and a simple flexible price model. Taylor concluded that the traditional approach can still play a useful role in explaining longer-run exchange rate behaviour.


Both Andrew Rose (University of California, Berkeley, NBER and CEPR) and Lars Svensson (Institute for International Economic Studies, Stockholm, NBER and CEPR) welcomed the paper's contribution in setting an agenda for future research. Studies adopting this microstructure approach should focus on how and why speculative attacks begin and why official interventions, although minute in comparison with daily turnover, exert significant effects. They should also assess the importance of such interventions' secrecy, the roles of rumours and heterogeneous beliefs, the effects of exchange rate volatility on hedging for trade, and the design of monetary policy to take account of markets' microstructure. Jeffrey Frankel noted that one of the reasons the data usually reject the PPP theory is that prices exhibit insufficient variance in the sample periods; indeed, the PPP theory tends to work better during periods of hyperinflation.


Charles Goodhart (LSE) then presented the preliminary results of his study, `One Day in June 1993: A Study of the Working of Reuters 2000-2 Electronic Foreign Exchange Trading System', joint with Takatoshi Ito and Richard Payne. They used prices from this new, electronic foreign exchange trading system as a bench-mark to assess the possible bias in the more commonly used data from the Reuters FXFX page, which reports only indicative bid and ask prices. Goodhart first stressed the need for caution about drawing general conclusions from this analysis of a data sample of only seven hours of trading activity on 16 June 1993, with no assurance that it was representative. He then showed that the means of the bid–ask spreads in the two series were essentially identical, even though their other characteristics, including their patterns of autocorrelation, were rather different. The time-paths of the indicative quotes therefore provide good proxies for those in the underlying series, although their spreads and their statistical behaviour are markedly different. Prices in Reuters 2000–2 are much more time-variant and dependent on the frequency of trade, while those in FXFX tend to cluster among a small number of standard values.


Alessandro Penati (Università Bocconi, Milano) questioned the robustness of the main result and suggested an alternative test based on agents' ability to profit from arbitrage between the means of the bid–ask spreads if the two series are not equivalent. He also questioned the authors' use of the 2000–2 system as a bench-mark for `real' prices; most trade is still conducted by telephone, so computerized trading systems cannot be a highly competitive mode of trading. Richard Lyons (University of California, Berkeley, and NBER) argued that the 2000–2 system may be considered competitive with a broker, but dealers generally work with smaller spreads among themselves.


In `Heterogeneous Behavior in Exchange Rate Crises', Fabio Bagliano and Andrea Beltratti (Università degli Studi di Torino) and Giuseppe Bertola (Università degli Studi di Torino, NBER and CEPR) used Italian data from the 1992 foreign exchange crisis to show that different classes of investors (domestic non-banks, foreign non-banks and banks) had behaved differently and to compute the gains and losses of the three classes of investors and of the central bank arising from the lira's devaluation. They then used an equilibrium model, constructed to reflect the heterogeneity of agents' objectives, to account for their observed behaviour in the run-up to the crisis. This enabled them to show that an exogenous rise in the devaluation probability – assumed to be uniform across the market – induced the different investors and the central bank to modify their portfolio positions, as well as yielding ex post speculative gains and losses as the devaluation took place.


Richard Lyons pointed out that the authors' hypothesis that the realignment probability is exogenous is very strong and suggested integrating the paper's theoretical and empirical sections more closely. Lorenzo Bini Smaghi (Banca d'Italia) stressed the need for care in drawing conclusions from the available data: balance-of-payments and even banking statistics are not very accurate in Italy, and the latter are not disaggregated by currency denomination and hence may not reflect exchange rate movements. Andrew Rose suggested that the heterogeneity of agents' behaviour derives largely from the heterogeneity of their beliefs on account of their having different information, which is not captured by allowing for differences in their risk aversion. Alan Kirman (European University Institute, Firenze) noted that all investors are on the same side of the market during speculative attacks – especially successful ones – so that discrimination among them is more difficult: stronger evidence of heterogeneity may often be found under normal conditions.


In `Risk and Turnover in the Foreign Exchange Market', Philippe Jorion (University of California, Irvine) used data on currency futures and options quoted on the Chicago Mercantile Exchange during 1985–92 to examine the empirical relationships between currency market turnover and unexpected volatility and between bid–ask spreads and expected volatility. Theoretical models of forex markets predict that both will display positive correlations: unexpected risk and turnover both depend on an information flow variable, while dealers must be compensated for deliberately assuming greater currency risk. Jorion used a time-series (GARCH) model to calculate expected volatility directly and the Black (1976) model for European options on futures to determine implied volatility by inverting an option pricing formula. His results revealed that implied standard deviations contain significant information on the future volatility of exchange rates and perform markedly better than time-series volatility forecasts. This reassuringly indicates that option traders form better expectations of risk over the next day than statistical models, and estimates based on this superior measure of expected risk also confirmed the predicted signs of the relationships among volatility, volumes and spreads.


Bernard Dumas (HEC School of Management, Jouy-en-Josas, CEPR and NBER) pointed out that the impact of news on volatility lasts for only a few minutes after its release. Analyses using higher-frequency data would probably require an option pricing model that allows for stochastic volatility.


In `Information Flows in the Foreign Exchange Market', written with Paolo Vitale, William Perraudin (Gonville and Caius College, Cambridge, and CEPR) contrasted `decentralized' foreign exchange markets, on which dealers are ignorant of other market makers' order flows, with most other markets, such as stock markets, on which such information is public knowledge. He examined the effects of asymmetric information in a model with many identical dealers and two types of customer: liquidity traders and those with access to privileged information. Dealers fix their spreads to protect themselves against informed customers and extract the maximum possible information from prices, and those that receive order flows from customers can `sell' this information to other dealers in the next period. Perraudin found that such decentralized markets exhibit wider bid–ask spreads (which discourages liquidity traders and hence makes order flows more informative), generate the maximum expected profits for dealers as a category (provided they can transact with all other market makers), and are less prone to crashes, although their prices (exchange rates) display higher variance.


Silverio Foresi (New York University) noted that in this model dealers act as a cartel to extract information from informed customers. The fact that none of them attempts to break the cartel may depend critically on an unstated assumption that the successful formation of a monopoly will not enhance their chances of trading with informed customers. Alan Kirman pointed out that there is no obvious reason why dealers cannot ask for quotes, without having to deal, so the information transfer mechanism must be specified more carefully. Richard Lyons noted that allowing the costs of inventory carrying and asymmetric information to account for the existence of bid–ask spreads would make the signal extraction problem much messier.


In `Foreign Exchange Volume: Sound and Fury Signifying Nothing?', Richard Lyons (University of California, Berkeley, and NBER) investigated whether trading is more informative when its intensity is high or low. Theoretical studies provide support for both possibilities, depending on the posited information structure, and Lyons sought to discriminate between them by analysing a data set comprising time-stamped quotes, prices and quantities of all the inter-dealer and brokered transactions of a single DM/dollar dealer at a major New York bank in a week. His results indicated that trades convey much less (more) information when transaction intensity is high (low), which supports the `hot potato' view: that dealers will repeatedly pass inventory imbalances among themselves following an initial customer order flow. If trading intensity is proxied by quoting intensity, however, the reverse applies, and `event uncertainty' prevails: a lower quoting intensity indicates that fewer information-based trades are on the market.


Both Mark Flood and António Mello (Banco de Portugal) stressed the extraordinarily informative content of the data set of these ultimately `real' prices, although the fact that they were available for one week only implies that the paper must be viewed only as a case-study. Both Giampaolo Galli and Marco Pagano (Università Bocconi, Milano, and CEPR) suggested investigating the extent to which the hot potato theory casts doubt on the high turnover of the foreign exchange market.


In `Bid-Ask Spreads in Foreign Exchange Markets: Implications for Models of Asymmetric Information', joint with David Hsieh, Allan Kleidon (Cornerstone Research) investigated the ability of `standard' asymmetric information models, which assume that all traders have perfect knowledge of each others' preferences and beliefs, to account for exchange rate volatility, bid–ask spreads and volumes at the opening and close of each trading day. He extended this framework to consider the London and New York forex markets jointly, since they are open simultaneously and trade in essentially the same `commodity'. These models attribute the high volatility and wide spreads observed when New York opens to the arrival of new information, which suggests that this should have world-wide repercussions, but the data indicate that this has little effect on trading in London. Kleidon argued instead that volatility is higher when New York opens because the current price per se does not reveal all the relevant information, and dealers need time to `get the feel of the market'. The assumption made in standard models that all traders have perfect knowledge of a market's structure must therefore be relaxed.


Zhaouhui Chen (LSE and CEPR) stressed the role of learning in this context: agents entering the New York market must learn a rule to judge the price before they can use the information it conveys. Charles Goodhart noted that the need to `get the feel of the market' should not apply to truly global banks that operate around the clock and thus do not have to close their positions at the end of the day. Richard Lyons stressed that Hsieh and Kleidon's data probably referred to customer–dealer trading, while the `real' (intra-dealer) market may display different behaviour.


In `Dynamic Hedging and the Interest Rate Defense', Peter Garber (Brown University) and Michael G Spencer (IMF) considered the impact of dynamic hedging strategies on the efficacy of a central bank's use of the interest rate to defend a fixed exchange rate. In conventional models, raising short-term interest rates reduces pressures on the exchange rate by squeezing the holders of short positions, but the use of option hedging strategies may also lead to high levels of automatic selling, if sceptical forex market agents interpret this action as the last rearguard action before abandoning the fixed rate. The interest rate differential is assumed to reflect the expected rate of depreciation, and so long as the exchange rate remains unchanged, a rise in the probability of depreciation increases the likelihood that traders will exercise their options to sell; the underwriter will then require a higher hedge ratio to sell the weaker currency. In an exchange rate crisis, a defensive rise in interest rates can therefore trigger large sales of the weaker currency and potentially have the perverse effect of weakening it further. Unfortunately, no data are available to assess the quantitative relevance of this effect.


Paolo Kind (Salomon Brothers) noted that heavy selling because of dynamic hedging can only happen if the official band limit provides a reference point and, moreover, heavy positions against a given currency have accumulated over time. He reported that this had applied to the lira and the Swedish krona in 1992 but much less to sterling. Also, the effect of dynamic hedging is stronger when there is a large number of options with a strike price at the weak edge of the band.


In `Is there a Safe Passage to EMU? Evidence from the Markets', joint with Barry Eichengreen and Andrew Rose, Charles Wyplosz (INSEAD and CEPR) assessed the effectiveness of capital controls by examining the behaviour of macroeconomic variables – such as budget deficits and money growth rates – around both successful and unsuccessful speculative attacks during the 25 years after 1966 in 22 countries with and without capital controls to show that they can play a significant role even if they do not reduce the likelihood of such attacks. Capital controls may well have reconciled a modicum of policy autonomy with the commitment to fixed exchange rates, provided the authorities with a breathing space in which to organize orderly realignments, and made it easier to rebuff speculative attacks that are not grounded in the fundamentals. Focusing on the channels of speculation, Wyplosz then noted that speculative positions must be financed through bank loans, at least in the long run, whose cost cannot be increased by sterilized interventions. He discussed a series of alternative policy proposals to support the achievement of EMU. The most original of these was raising the cost of financing speculative attacks by imposing a tax on the lending of domestic money to foreign residents.


The discussion focused on this policy proposal (which two of the authors had advanced previously). Vittorio Grilli (Ministero del Tesoro, Roma, and CEPR) noted that the distinction between resident and non-resident institutions is becoming increasingly blurred which complicates the application of such a tax. Moreover, this tax would also impose costs on bona fide investors in home government bonds as well as speculators. Jose Viñals (Banco de España and CEPR) also stressed that it is virtually impossible to design a tax that affects only speculators in a highly integrated market. Andrew Crockett (Bank for International Settlements) pointed out that this proposal ignored the purpose of the Maastricht criteria for EMU, i.e. to force credibility through convergent policies.


The conference ended with a panel discussion on `Is the Growth and Evolution of Foreign Exchange Markets a Concern for Policy Makers?' This focused on the high exchange rate volatility which may lead in turn to large misalignments, external constraints on and repercussions of domestic stabilization policies, and dangers to systemic instability. Antonio Fazio (Banca d'Italia) contrasted the weak coordination of national monetary policies with the increasing globalization of financial markets. Andrew Crockett stressed that market expectations play a key role and warned against the systemic risk arising from the complexity of certain financial instruments. David Mulford (Credit Suisse First Boston) maintained that markets behave irrationally at most for short periods but usually have a disciplinary effect on monetary policies. Ian Plenderleith (Bank of England) pointed out specific aspects of the foreign exchange market that need to be understood better: in particular the roles played by information, expectations, hedging strategies and chartism. Fabrizio Saccomanni (Banca d'Italia) also advocated conducting surveys more frequently in order to counteract the current lack of data on activity.


The papers and proceedings of this conference will be published in a joint CEPR/NBER volume, edited by Jeffrey Frankel, Giampaolo Galli and Alberto Giovannini.