سیاست های پولی و انتظارات استنباطی از نرخ ارز

We present a macroeconomic market experiment to isolate the impact of monetary shocks on the exchange rate, as an alternative to SVAR identification. In a non-stochastic treatment, covered interest rate parity holds and predicted exchange rates are tracked well. In a stochastic treatment, we model expectations using a Neyman–Pearson hypothesis test (inferential expectations) and find evidence of belief conservatism and uncovered interest rate parity failure. The market environment magnifies belief conservatism, which is opposite to the standard claim that markets tend to eliminate individual choice anomalies.

The debate about monetary policy's impact on exchange rates has spanned the whole floating rate era (Kim and Roubini, 2000 and Chinn, 2006). In his seminal overshooting model Dornbusch (1976) elegantly melded the central bank's influence over interest rates and the future nominal exchange rate into a theory of foreign exchange. His setup has been a building-block ever since, and is imbedded in the New Open Economy Macroeconomic paradigm (Lane, 2001). Yet it is no secret that OLS regression tests have weighed heavily against the key Dornbusch exchange rate equation – uncovered interest parity (UIP). Under the joint hypotheses of rational expectations, risk neutrality and zero transactions costs, the current interest differential ought to be an unbiased predictor of the future change in the nominal spot rate. That is, the regression slope in Eq. (1) should be unity2 equation(1) View the MathML sourceΔst+1=β(it−it*)+ut+1 Turn MathJax on Instead, the estimated coefficient in Eq. (1) is less than unity, and sometimes negative (Frankel and Rose, 1995 and Froot and Thaler, 1990).3 However, Eq. (1) is not a strict test for the appropriateness of the Dornbusch (1976). This model inhabits a world where the only shocks are nominal, so there is no reason to suppose (1) will hold in general. Fortunately, structural vector autoregressions (SVARs) enable researchers to sift pure monetary shocks from all others and (1) can be tested for this subset. One such study (Bjornland, 2009, p. 1403) supports UIP, claiming that “Dornbusch was right after all”. However, SVARs can only test joint hypotheses: the hypotheses explicitly in view, plus any identifying restrictions. Unfortunately, SVAR results about UIP are beholden to different identifying restrictions (Faust and Rogers, 2003 and Fry and Pagan, 2007). The contribution of this paper is to detail a market experiment where agents must trade foreign exchange in the light of information about a stylized economy. Our experimental conditioning is an alternative to the econometric SVAR conditioning, because the only shocks affecting the economy are monetary. Furthermore, unlike a SVAR, our experiment allows the complete set of agents’ buying and selling intentions to be observed. As we shall see, agents exhibit belief conservatism. That is, they change their beliefs (about future nominal exchange rates) only when the evidence passes a threshold. We operationalize this by treating belief formation as a statistical hypothesis test. Under inferential expectations (IE), agents with a high evidence threshold act as though they conduct a hypothesis test with a low test size α our measure of belief conservatism ( Menzies and Zizzo, 2009). 4 One surprising result we find is that belief conservatism for individuals is less pronounced than it is in a market. This contrasts with the folk wisdom that markets should drive out individual-choice anomalies. A caveat for this result is that our metric for individual belief conservatism is based on the bid and offer that the experiment elicits from them. As a robustness check, we confirm the existence of individual choice belief conservatism more precisely in a non-market supplementary experiment. The paper is structured as follows. Section 2 outlines the notion of inferential expectations, which can be used to quantify belief conservatism while retaining rational expectations as a special case. In Section 3 we build a two-period theoretical economy which becomes the environment for the experiment, discussed in Section 4. The results are outlined in Section 5, with the last sub-section providing the results from a supplementary experiment. Section 6 concludes.

We have presented a novel macroeconomic market experiment on the determination of exchange rates using financial data (such as interest rates, and exchange rate forecasts). We used a Walrasian market mechanism to implement the exchange rate market, because it has very high efficiency properties in terms of convergence to competitive prices and quantities (Holt, 1995). Our subjects received considerable opportunities to gain experience over about 4 h of experiment, in two sessions, spaced one to three days apart; they were very motivated and no one dropped out between sessions. They had good understanding of the macroeconomic environment and exchange rates were always in a reasonable range. In non-stochastic sessions exchange rates tracked the covered interest parity prediction very closely. In stochastic sessions, inferential expectations with low αs predicted exchange rates better than rational expectations in 10 sessions out of twelve. Relative to the individual choice αs estimated by bids and offers, the Walrasian market mechanism does not reduce but actually magnifies the predictive power of belief conservatism. This magnification was apparent even when we compared the αs to those obtained in a non-market supplementary experiment. This is a striking result since some individual choice anomalies may be substantially reduced in a market environment, and, for many economic issues – such as exchange rate determination – it is clearly market behavior that matters, not individual choices as such. It would have been legitimate to expect that, although inferential expectations may outperform rational expectations for a significant fraction of agents in an individual choice experiment (as detailed in an online appendix), at the market level rational expectations may still be a better predictor of exchange rates. Thus, by testing a non-standard model of expectation formation by looking at market prices, ours was designed as a potentially harsher test than that provided by a number of individual choice experiments. We found that uncertainty in the macroeconomic environment roughly doubled the downward bias in uncovered interest rate parity regressions. While obviously further research is needed, we suggest that belief conservatism, as modeled by inferential expectations, may be part of the explanation why nominal exchange rate forecasts are not calculated using rational expectations, at least until a threshold of evidence is reached. An implication is that sharp movements in exchange rates may sometimes reflect switches in beliefs about future nominal rates as a threshold of evidence is attained. In those instances, the search for contemporaneous causal variables may be misconceived. We initially motivated our experiment by noting some of the shortcomings of SVARs, but we cannot deny that experiments have their own limitations as a methodology. The amounts of money paid to the participants, most of whom will never be foreign exchange traders, were small. However, the divergent UIP verdicts arising from our experiment and the SVARs are surely worthy of further research.