RDP 9406: Reserve Bank Operations in the Foreign Exchange Market: Effectiveness and Profitability 5. Alternative Tests of Stabilising Intervention

5.1 Wonnacott's Criterion

5.1.1 Methodology

Another test of whether exchange rate intervention is stabilising is if it acts to reduce the variance of the exchange rate around its equilibrium. This definition, however, has a major shortcoming in that it requires an exchange rate model to determine what the equilibrium rate should be, something which continues to elude the economics profession.

Wonnacott (1982) argued that a more practical definition is that intervention is stabilising if it reduces the variance of the exchange rate around its trend; this long-term trend is approximated with a centred moving average. The deviation of the exchange rate from its moving average is given by:

where:

MA_k is a centred moving average over k periods.

Wonnacott's test involves measuring whether the direction of intervention is consistent with pushing the exchange rate back towards its long-run moving average. If DMA is greater than zero, so that the exchange rate is above its long-term trend, intervention that is working in the right direction would involve the Reserve Bank selling foreign currency to the market. Similarly, intervention would be stabilising if the Bank purchased foreign currency from the market when DMA is negative. The performance of the monetary authority can be measured by the ratio of the number of days when intervention was in a stabilising direction to the number of days when intervention took place.

This success ratio is given by:

where:

d_i = 1 if intervention was in the right direction (and zero otherwise); and,

D_i = 1 if intervention took place (and zero otherwise).

Alternatively, the success ratio can be weighted according to the dollar value of intervention so as to account for the relative intensity of the intervention:

where:

n_i is the absolute amount of foreign currency purchased or sold in period i.

This pragmatic approach rests upon a number of assumptions. First, the moving average is a reasonable proxy for the equilibrium rate, given that this equilibrium cannot be otherwise quantified. Large and discrete one-off movements in fundamentals could require an abrupt shift in the equilibrium rate, something which cannot be picked up by a simple moving average. Second, intervention has no discernible impact on the longer run trend in the exchange rate. If intervention has some impact in the longer-term, then the centred moving average exchange rate would not be an exogenous parameter for the monetary authority, but would be endogenous to its own policies. Whilst these assumptions are fairly restrictive, we would suggest that violations to them have probably not been too great throughout the period; as a consequence, the success ratios (SR and SR($)) should allow us to reasonably determine whether intervention was working in the right direction^[15].

An underlying theme of Wonnacott's test is that intervention is more likely to be working in the right direction the larger the deviation of the exchange rate from its moving average. As a consequence, we have also calculated the success ratios using a restricted sample that excludes those observations which lie within an x per cent band of the moving average. Two alternative band widths were chosen: 1½ per cent and 3 per cent either side of the moving average. This is also more realistic because it allows the central bank some latitude in assessing where the longer-term exchange rate is likely to lie. In addition, because the lag length of the moving average is also somewhat arbitrary, the success ratios were calculated using 3-month, 6-month, and 12-month moving averages.^[16]

5.1.2 Empirical Results

Table 2 sets out the results for the total post-float period, for the various combinations of moving averages and band widths. The way these can be interpreted is as follows: row 3, for instance, shows results where a three-month moving average and a 3 per cent band width are used. The statistics show that over the whole period the exchange rate deviated from its moving average by more than 3 per cent on 139 days. On 94 of these days (i.e. 68 per cent) the Bank intervened in the foreign exchange market. Of these interventions, 80 per cent (91 per cent when weighted by the size of intervention) were in a stabilising direction. The final column shows that the average daily value of intervention that was in a stabilising direction was $A66 million.

Success ratios are all in excess of 50 per cent, which indicates that the Bank's intervention on average was in a stabilising direction on this test. If intervention had been random, success ratios would not have been significantly different from 50 per cent.^[17] The success ratios are significantly higher when weighted according to the size of intervention, suggesting that large interventions had a greater propensity to be in the right direction than smaller ones. Intervention tended to occur more frequently the further the exchange rate deviated from its moving average and the success ratios also tended to be higher in these cases.

The style of the Bank's intervention has changed during the period since the float. Up to late 1985, intervention was light. The average amount dealt on the days when the Bank intervened was about $A12 million, with the largest being on 8 November 1985, when the Bank purchased $A90 million. In 1986, the scale of intervention started to become bigger. The catalyst for the change in approach was recurring episodes of extreme foreign exchange market instability. Around the middle of the year, as the exchange rate had already fallen a long way and looked undervalued in relation to economic fundamentals, the Bank intervened heavily to provide a solid foundation. On several days, purchases of Australian dollars from the market approached $A250 million, and for July 1986 as a whole, net purchases amounted to $A1.2 billion. Through the second half of the 1980s, the Bank maintained a frequent presence in the market, intervening on 76 per cent of days. The average size of daily intervention was about $A50 million. In the 1990s, the frequency of intervention fell sharply to 26 per cent of days, though the average size increased to about $A110 million.

It is possible to split the floating rate period into three broad sub-periods, corresponding to different phases in the Bank's intervention:

• from the float until late 1985, when intervention in the market was very light;
• from late 1985 to early 1991, when the Bank intervened frequently in the market with moderate volumes; and
• since early 1991, when the Bank has been in the market less frequently but has dealt in large amounts when needed.

The results for these three sub-periods are shown in Table 3. These results show that the success rate of intervention, as measured by SR and SR($), was a little higher in the second sub-period than in the first period, but increased noticeably in the latest sub-period when the Bank intervened less frequently but in larger amounts. Most success ratios in this last period were above 70 per cent and some were 100 per cent. This suggests that the Bank was most successful when it entered the market less frequently but on a significant scale.

Studies of other major foreign exchange markets have come up with similar conclusions. Wonnacott (1982) applied his test to US intervention in Deutschemarks over the late 1970s, Mayer and Taguchi (1983) studied intervention by Germany, Japan and the United Kingdom between 1974 and 1982 while Murray, Zelmer and Williamson (1990) examined Canadian intervention over the period 1975 to 1988. Their results showed that intervention was helpful, at least on some simple tests. For example, the percentage of days when intervention occurred, the percentage of intervention that was in the right direction and the average value of intervention all tended to increase as the exchange rate moved further away from its moving average.

5.2 Hybrid Criterion

5.2.1 Methodology

A variant to Wonnacott's tests was proposed by Mayer and Taguchi (1983). They distinguished two exchange rate zones by fitting a band around the trend in the exchange rate. Outside this band, intervention is judged according to Wonnacott's criterion. Within the band, however, intervention is judged according to a “leaning against the wind” criterion, the so-called LAW criterion. Under this criterion, successful interventions push the rate towards the last observed level. In other words, the central bank should buy the local currency when it is falling and sell it when it is rising, regardless of where it stands vis-a-vis the moving average trend. Mayer and Taguchi argued that this hybrid criterion more accurately reflected central bank behaviour.^[18] They suggested that if the exchange rate is a long way from its equilibrium level but is moving towards it, then Wonnacott's criterion should prevail. But as the exchange rate moves closer to its equilibrium level, then intervention should be assessed according to the LAW criterion. This is because the equilibrium level cannot be ascertained exactly, and there is the likelihood of misjudgment. Moreover, in order to obtain a “smooth landing” in the equilibrium zone, it might be advisable to lean against the wind to ensure that momentum does not build up to push the exchange rate through the equilibrium zone. Figure 6 provides a graphical comparison of these criteria.

5.2.2 Empirical Results

In testing the Bank's intervention using the hybrid criterion, two bands were chosen around the long-run trend within which the stabilising impact of intervention was judged according to the LAW criterion; they are 1½ per cent and 3 per cent on each side of the moving average.

Table 4 shows results using this criterion for the total post-float period and for the three sub-periods. Qualitatively, these results are not very different from using Wonnacott's criterion:

• success ratios were again well above 50 per cent;
• in all periods, SR($) is higher than SR. This suggests that intervention for large amounts was more successful than intervention for small amounts. Partly this reflects the fact that large intervention tended to take place when the exchange rate was a long way from its long-term average (i.e. when the exchange rate was more likely to be out of line);
• intervention was generally more successful in the second sub-period than in the early period; and
• the success ratio SR($) tends to be highest in the recent sub-period. In contrast, however, the success ratio, SR, is little different from the earlier periods and is well down compared with the success ratios measured using Wonnacott's criterion. This suggests that, in this period, the Bank's operations when the exchange rate was within the band were not characterised by the LAW criterion.

Footnotes

The success ratios measure whether the direction of intervention is correct. They do not test whether intervention influences the level of the exchange rate. [15]

The tests were performed on series that contain 6 months fewer observations than the original series in order to allow for the calculation of a 12 month centred moving average. [16]

Given the sample size, any value more than 2 per cent away from 50 per cent would be significantly different from a random outcome. [17]

Mayer and Taguchi argued that the LAW criterion made little sense applied to all exchange rate movements because it implied that the authorities always took the view that the prevailing exchange rate is the best one. Under this approach, therefore, all exchange rate movements should be dampened, if not suppressed. [18]