RDP 2004-06: Profitability of Reserve Bank Foreign Exchange Operations: Twenty Years After the Float Appendix A: Effectiveness of Australian Intervention – A Survey

Many authors have attempted to measure the effectiveness of foreign exchange intervention by central banks. Edison (1993) conducts a thorough survey of the early international literature. Sarno and Taylor (2001) discuss the progress made over the 1990s in international studies of the efficacy of intervention.

Edison, Cashin and Liang (2003, p 12) note a ‘general consensus in the literature’ that intervention via influencing market participants' portfolio decisions is ineffective, but that there is some evidence that intervention via the signalling channel is effective. Sarno and Taylor (2001) view the evidence from the 1990s as supportive of the effectiveness of intervention via both channels. Sarno and Taylor also suggest a third channel of intervention that may be relevant; the coordination channel.^[12]

A considerable literature has developed that studies the effectiveness of the intervention activities of the RBA. This survey outlines the methods used and empirical results found in this literature.^[13]

A1. Evaluating the Effectiveness of Intervention

Measuring the effectiveness of intervention is difficult for two reasons. Mainly this is because the counterfactual is unknown. That is to say, it cannot be reliably estimated how the exchange rate would have behaved in the absence of intervention.^[14] In addition, there is endogeneity between intervention and exchange rate movements; intervention will affect the exchange rate and the exchange rate is also an input into any decision to intervene. It is difficult to control for this, and in attempts to mitigate the problem, some studies have excluded the contemporaneous effect of intervention on the exchange rate, leading to possibly biased results. Many attempts to evaluate the effectiveness of intervention that have failed to capture the endogeneity have been unsuccessful.

Nonetheless, five main methods have been used in the literature in attempts to measure the effectiveness of intervention. These are:

• event studies;
• time series studies;
• measuring the profits from intervention;
• Generalised Methods of Moments (GMM); and
• survey methods.

A1.1 Event Studies

Event studies identify distinct intervention episodes and examine the effectiveness of intervention within the intervention window. Intervention is judged to be effective if it ‘has been successful at stopping or delaying any given trend in the exchange rate’ (Edison et al 2003, p 12). Fatum (2000) observes that ‘standard time-series techniques may not be well suited when dealing with the analysis of intervention vis-à-vis the behaviour of exchange rates’, whereas the event study method ‘seems to fit well’ given that ‘a cluster of intervention operations constitutes a natural candidate for identification as a single event’ (Fatum 2000, pp 5–6). In addition, examining intervention windows can also allow the use of intraday data on exchange rates if these data are available.

Edison et al (2003) undertook an event study of the RBA's intervention in the Australian dollar. The authors found 18 episodes of intervention between 1984 and December 2001 with 12 out of 18 episodes classified as successful. They conclude that there is some evidence that the ‘RBA has been effectively “leaning against the wind” in changing the trend movement in the Australian dollar exchange rate’ (Edison et al 2003, p 17).

The method of classification employed in event studies is often somewhat inflexible. An episode of intervention is classified as successful if the intervention immediately reverses the trend observed in the exchange rate prior to the intervention, or if it achieves a continued reversal of the trend, or both. However, the method does not take account of the purpose of the intervention.

In Edison et al's study, this inflexibility leads to two possible misclassifications of intervention episodes. Intervention in December 2000 is classified as a failure because it does not reverse the direction of the trend in the exchange rate. The authors note that it could also be classified as leaning with the wind: the RBA purchased Australian dollars at a time when the Australian dollar was appreciating (from a low level) and that appreciation continued following the intervention. That is, the RBA leaned with the wind successfully. Intervention in January 2001 is a second possible misclassification. Here, the RBA again purchased Australian dollars at a time when the Australian dollar was appreciating, a possible attempt to lean with the wind, but the currency depreciated over the intervention window and following the intervention. Because the exchange rate firstly appreciated and then depreciated, the intervention was classified as successful. However, as the RBA aimed to support the currency and this was not the result, using this methodology the episode should not be classified as a success.^[15]

A1.2 Time Series Studies

Various time series studies have also been conducted to evaluate intervention. Many of these have failed to find strong evidence to suggest that intervention is effective. This is most often due to the fact that these models do not capture the endogeneity between intervention and the exchange rate, or have excluded the problem of endogeneity by excluding contemporaneous intervention preventing ‘measurement of the immediate impact [of intervention]’ (Kearns and Rigobon 2003, p 5).

Consequently, the results produced in these models are likely to suffer from a negative bias which results in small and insignificant or incorrectly signed estimates for the impact of intervention. That is, the models predict that a purchase of the domestic currency has little effect on the currency or even depreciates the currency. Other coefficient estimates may also be biased due to the omitted variables. Kearns and Rigobon (2003) discuss the effects of the exclusions and provide an empirical example of the effect of ignoring the contemporaneous effects of central bank intervention.

Despite the difficulties faced, time series studies have been undertaken to address the important question of the efficacy of foreign exchange intervention. Various methods have been used and some of those employed in the Australian context are considered here.

A1.2.1 Results from GARCH time series modelling

Many of the time series studies use models of the exchange rate with GARCH or EGARCH (exponential generalised autoregressive conditionally heteroskedastic) error structures. These models ‘allow the empirical testing of the effectiveness of intervention to be carried out simultaneously on both the mean and conditional volatility of exchange rate returns’ (Kim, Kortian and Sheen 1999, p 10). Testing the effects of intervention on the conditional mean of the exchange rate is one way to test if the intervention has had the effect of reversing or dampening a trend in the exchange rate.

Kim et al (1999) employ an EGARCH model to explain the percentage change in the A$/US$ exchange rate with intervention included as one of the explanatory variables. They use daily data from December 1983 to December 1997 and break the sample up into smaller sub-periods according to variations in the intervention style employed by the RBA. The authors find that the estimated contemporaneous effects of intervention were destabilising. However, the authors note that this is likely the result of the simultaneity of intervention and the exchange rate. When the slope dummy variables are examined they find ‘evidence of a stabilising influence on the $A/$US exchange rate process’ due to the RBA's intervention and that ‘a worse outcome on [the day of intervention] may have occurred if the Reserve Bank had not intervened’ (p 16).

A1.2.2 Results from Central Bank Reaction Function Studies

Other time series studies (Rogers and Siklos 2003, McKenzie 2004 and Kim and Sheen 2002) use a central bank reaction function to investigate the conditions under which the central bank may intervene. Identifying what prompts a central bank to intervene can provide evidence as to whether the central bank is acting to ‘lean against the wind’. These studies most often use a Probit model to predict the probability of intervention given movements and volatility in the exchange rate.

Rogers and Siklos (2003) study the intervention of the RBA using daily data for the period of January 1989 to September 1998. They find ‘considerable evidence’ of ‘leaning against the wind’ throughout the 1989–1997 period and intervention was ‘quantitatively larger’ in the 1989–1993 period.^[16] In addition, as would be expected, the stabilising effects of intervention are greatest on the day of intervention and decrease gradually on the days following the intervention.

Kim and Sheen (2002) use data from December 1983 to December 1997 ‘to estimate Probit models for purchases and sales of foreign currency (in US$) separately’ and also to ‘estimate a friction model of intervention whereby the Reserve Bank chooses to buy/sell only beyond threshold limits’ (p 627). The authors find in general ‘that a moderate appreciation (depreciation) of the $A from its 150–day average leads to an intervention purchase (sale) of foreign currency designed to slow the rise (fall) of the value of the $A’ (p 647). This form of intervention is ‘in accordance with the stated short horizon aim of leaning against the wind’. In addition, the friction model reveals that ‘intervention is strongly correlated with lagged intervention, which suggests that positive (negative) intervention was usually followed by positive (negative) intervention on the following day’ (p 643). Kim and Sheen suggest that this persistence may imply that RBA intervention tends to be carried out over a few days and they propose that this may improve its effectiveness as their model gives a significant estimate for the stabilising effect of cumulative intervention.

McKenzie (2004) conducts analysis using a Probit analysis of the reaction function; an ‘ex post government intervention decision against a measure of exchange rate volatility’ (p 62). In addition, he proxies for exchange rate volatility using a GARCH model. McKenzie finds that the ‘dynamics of the foreign exchange market are significantly different on the days’ on which the RBA intervenes (p 72). He notes that it could be that RBA intervenes on these days because volatility in the market is high, or it may be that volatility is high because the RBA intervenes. However, distinguishing between these is not possible within his analysis due to the likely endogeneity of intervention and the exchange rate.

A1.2.3 Other time series approaches

Hopkins and Murphy (1997) undertake a case study using regression analysis to determine the effect of intervention on the market. They examine the 7.5 per cent depreciation of the Australian dollar during July to October 1993 and aim to identify the relative importance of information in the market at that time when ‘uncertainty regarding the passage of the Federal Budget through the Senate was reflected in the foreign exchange market’ (p 199). They find, for this period of uncertainty in the market, that ‘intervention operations and associated statements by the RBA did provide some stability to the market’ (p 217). However, their results should be interpreted with caution as only a small sample size is used due to the restricted time period studied, and as the authors include contemporaneous intervention but ignore the possible simultaneity issues that arise as a result.

Karunaratne (1996) employs multicointegration techniques to test the hypothesis that RBA intervention has been ineffective. The author breaks up the period from December 1983 to May 1993 into five sub-periods and models the rationale for RBA intervention using a quadratic loss function where the RBA aims to minimise ‘losses due to target missing and losses because of exchange rate instability’ (p 409). The results of the Johansen multicointegration tests are ‘favourable to the proposition that RBA intervention was effective from a long-run perspective’ (p 415). However, the methods used do not take account of the endogeneity of the exchange rate and intervention. Consequently, as Karunaratne notes, the long-run equilibrium relationships found between the nominal exchange rate, the foreign interest rate as a proxy for the unobservable target exchange rate target variable, and the domestic interest rate (with net purchases of foreign currency included as a control variable), may ‘have occurred even without intervention’ (p 415).

A1.3 Profits from Intervention

The thinking underlying this line of research was outlined in the body of this paper. As noted, it is an adaptation of the argument put by Friedman (1953) that stabilising speculation should be profitable. Until now, Andrew and Broadbent (1994) had been the only Australian study to conduct an analysis of the profits from intervention. They find significant profits from intervention between December 1983 and June 1994 and conclude that RBA intervention had a stabilising influence on the Australian dollar exchange rate against the US dollar.

A1.4 Generalised Method of Moments (GMM) Studies

There has been a recent innovation in the study of intervention developed by Kearns and Rigobon (2003). The authors use simulated Generalised Method of Moments to evaluate the effectiveness of intervention while explicitly allowing for the endogeneity of the intervention and the exchange rate.

Kearns and Rigobon (2003, p 1) recognise that ‘once a central bank has decided to intervene, the quantity of currency it buys or sells and the decision as to whether to engage in further intervention will typically depend on the response of the exchange rate to [the central bank's] trades’. The advantage of GMM is that the model allows estimation of a set of simultaneous equations and explicitly captures the interdependence of the exchange rate and the central bank's intervention and the contemporaneous impact of intervention.

Their results provide empirical econometric evidence to support the description of RBA intervention as ‘leaning against the wind’. The results of the paper suggest that there is a strong positive effect of intervention (where Australian dollars are purchased and foreign exchange sold); particularly on the day it is conducted, with a smaller positive effect for a few days afterwards.^[17] They find point estimates to suggest that RBA intervention has an economically and statistically significant contemporaneous effect on the exchange rate, such that a US$100 million purchase of Australian dollars will appreciate the Australian dollar by between 1.3 per cent and 1.8 per cent.

A1.5 Survey Methods

One other method that has been applied to gauge the effectiveness of RBA intervention is to survey market participants. Hutcheson (2003) surveyed foreign exchange dealers licensed by the RBA as at 12 July 1999.^[18] The tabulated responses imply that the RBA's intervention transactions and their motivations are viewed as credible, with 86 per cent of respondents stating that intervention is usually conducted at the appropriate moment, 77 per cent saying that intervention achieves the desired goals and 73 per cent answering that intervention moves exchange rates towards their fundamental value. Fifty per cent of respondents thought that intervention increases exchange rate volatility. However, this result should be interpreted with caution. If the exchange rate falls intraday and intervention successfully reverses the depreciation, intraday volatility will have increased due to the intervention. Alternatively, as Hutcheson suggests, intraday volatility may increase if intervention is not well anticipated by dealers. In addition, this result is difficult to interpret as no time frame for volatility was stated in the survey question.

Footnotes

The authors suggest that the central bank has a role to remedy coordination failures in the foreign exchange market when the exchange rate is overvalued or experiencing a bubble, but where traders are unwilling to be the first to break the trend. The central bank could use publicly announced intervention operations to coordinate ‘smart money’ traders to enter the market together and break the bubble (Sarno and Taylor 2001, p 863). [12]

For background information on the RBA's foreign exchange operations, refer also to Fraser (1992), Macfarlane (1993, 1998) and Rankin (1998). [13]

Sarno and Taylor (2001) refer to studies by Dominguez and Frankel in which exchange rate expectations are used to proxy for the counterfactual. However, to our knowledge no studies of this sort have been carried out in the Australian context. [14]

These reclassifications would not have altered the conclusions of the study as 12 out of 18 intervention episodes would still be classified as successful. [15]

Despite noting the possibility of simultaneity between their chosen estimation equations, due to the endogeneity of exchange rates and intervention, the authors estimate equations separately on the basis of Hausman specification tests. However, if simultaneity was in fact a problem, the estimates may be biased. [16]

This result is consistent with the findings of other studies where significant effects are not found if the contemporaneous effects of the intervention are excluded as these studies only pick up the smaller effects on subsequent days. [17]

Fifty-nine surveys were sent to foreign exchange dealers and 39 of these were completed. The respondents were mostly senior employees in their institutions' treasury departments. [18]