• Home
• About MED
• Publications
• Occasional Papers
• 08/02
• 6. Selective Hedging

• Publications

• Occasional Papers

• 09/01
• 08/09
• 08/08
• 08/07
• 08/06
• 08/05
• 08/04
• 08/03
• 08/02
• 08/01
• Archive

We analyse two features that may be relevant to selective hedging choices. The two features relate to exchange rates relative to perceived fundamentals and the level of forward points (i.e. short term interest rate differentials).

First, we examine whether the hedging ratio for a particular currency exposure increases at times when the NZD is low relative to that currency. Such behaviour may reflect management having a benchmark ratio of trade that it hedges, but choosing to lock in a higher ratio for what is perceived to be a favourable exchange rate on an opportunistic basis. Alternatively, management may have a hedged benchmark but choose to take its hedge ratio below the benchmark at times when it perceives the NZD to be high.

Figure 11 plots the hedge ratio for all USD-exposed transactions together with the USD/NZD exchange rate. Signs of selective hedging, especially in the first half of the sample, are apparent. For instance, the USD/NZD rose over the first half of 1999 and the hedge ratio fell from 36% to 24%; over 2000-2001, the USD/NZD fell sharply and the proportion of hedged USD transactions rose markedly (although, as discussed previously, this occurrence may be subject to a selection effect). Later in the sample, the trends in each of the series are, by contrast, strongly positively correlated although there are times when short run changes remain in the opposite direction (e.g. in early-mid 2004, when the hedge ratio rose from around 65% to 70% at a time when the USD/NZD rate had slipped from 0.692 to 0.616 over three months).

Figure 12 depicts comparable information for the AUD; Figure 13 presents information for "Other" currencies plotted against the Trade-Weighted Exchange Rate Index (TWI). Similar examples of short term selective behaviour are apparent in both graphs. So too is the longer run positive correlation between hedging ratios and currency movements. These results suggest that selective hedging behaviour, if based on perceptions of currency "opportunities", may be affected as much by short term currency changes as by longer term currency levels.

Figure 14 plots the hedge ratio for all USD-exposed transactions together with the short-term (30 day) interest rate differential (forward points) between New Zealand and the United States (FPNZUS). Figure 15 plots the comparable information for AUD-exposed transactions, where FPNZAU is the forward points against the AUD. In both cases, a normalised series for each variable is used to make the relationship scale-neutral.

A small positive relationship exists in each case. This relationship is consistent with New Zealand exporters increasing their hedging when forward points move further in their favour (noting that, for most of the sample, the forward points have consistently been positive for New Zealand exporters). The correlation in the USD case is 0.39, and in the AUD case is 0.44.

VAR Modelling

We investigate the presence (or otherwise) of selective hedging more formally using unrestricted vector autoregressions (VARs). Specifically, we examine the relationship between hedge ratios (for both values and transactions) and exchange rates, subsequently adding the influence of forward points. We concentrate specifically on USD and AUD exposures.

Prior to formulating the VAR specification, we test the variables for their time series properties. Table 5 reports p-values for Augmented Dickey-Fuller (ADF) tests, with the null of a unit root, for each of the variables of interest. We treat a variable as stationary where p<0.05.

On these tests, both the USDNZD and AUDNZD are non-stationary, as are the USD hedge ratios. The AUD hedge ratios, by contrast, are stationary.³⁴ It is important only to include stationary variables in the VAR (unless a cointegration relation is present). To ensure we use stationary variables, we consider that each variable may vary around a stochastic trend that can be approximated by a Hodrick-Prescott filter³⁵ passed through the series. We take the cycle series (i.e. the raw series minus the HP trend series) as our measure in the VAR (each detrended series, labelled with the suffix, _CYC, is stationary). For consistency, we use the HP-filtered series for the AUD hedge ratios as well as for the non-stationary variables. It makes little difference to our results whether we use the raw or filtered series for the AUD hedge ratios.³⁶

Definitions:

USDNZD is the USD/NZD exchange rate;

AUDNZD is the AUD/NZD exchange rate;

AUDhr_tra is the transactions-based hedge ratio for AUD exposures;

AUDhr_val is the value-based hedge ratio for AUD exposures;

USDhr_tra is the transactions-based hedge ratio for USD exposures;

USDhr_val is the value-based hedge ratio for ASD exposures;

fpnzau are the forward points for the NZD relative to the AUD;

fpnzus are the forward points for the NZD relative to the USD.

A variable with a _cyc suffix in subsequent analysis is the detrended series (using an HP filter).

Prior to estimating and presenting the VAR results, we outline what we may expect to find. First, in the two variable (hedge ratio and exchange rate) VAR, selective hedging will be indicated where there is a significant negative response of the hedge ratio to the exchange rate. This result would indicate exporters locking in perceived low exchange rates for their exports, while remaining exposed when the exchange rate is perceived to be abnormally high.

Second, we test the relationships using both the transactions-based and value-based hedging ratios. The value-based ratios weight large (relative to small) exporters more heavily than in the transactions-based measure. Thus differences in large versus small exporter behaviour may be inferred from different reactions of the two hedge ratio measures to exchange rates. For instance, if the value-based measure shows a stronger reaction to exchange rates than does the transactions-based measure we can infer that large exporters are more likely to engage in selective hedging behaviour.

Third, after estimating the two variable VAR, we add in forward points as a third variable. We would anticipate a positive relationship between the hedge ratios and forward points (if there is a relationship). Again the relationship may differ between the transactions-based and value based measures, indicating different behaviour of large versus small exporters.

Each VAR is estimated using three (monthly) lags of each variable.³⁷ Once estimated, impulse response functions (IRFs) are calculated, based on each of the estimated systems.

Appendix 1 presents graphs of the effect of a one standard deviation change in the (detrended) exchange rate on the relevant hedge ratio. In each graph, the solid line depicts the estimated response; the dashed lines indicate two standard error bands (calculated analytically). We treat any relationship as "significant" where zero sits outside the confidence bands for at least one month.

Initially we conduct the analysis for the full period, being 1997:11 – 2007:02 (we lose the initial three months due to the lag structure of the VAR). We also split the sample at the mid-point (after 2002:06) to test for changes in the relationship over time; the full period contains 112 observations, with 56 observations in each sub-period. The latter sub-period escapes the data problems that may be associated with the 2000 hedging ratio spike; most of this sub-period also has full electronic capture of the data. For these reasons, greater weight may be placed on the second sub-sample results.

Figure A1 in the Appendix indicates that for the full period, the value-based hedging ratio for AUD exposures (AUDhr_val_cyc) responds significantly to the AUDNZD exchange rate (AUDNZD_cyc). The maximum response occurs three months following an exchange rate change, possibly reflecting the hedging horizon of exporters; i.e. hedging decisions are made three months prior to the export transaction.

When transactions are used as the measure (Figure A2), a similar relationship, albeit shallower and with longer lags (and not quite significant) is obtained. The differences between the two suggest that larger exporters are both quicker at responding to perceived exchange rate misalignments (between the NZD and AUD) and respond more aggressively than do smaller exporters.

Figures A3 and A4 plot the corresponding IRFs for USD exposures. The value-based measure shows a similar, but insignificant, pattern to the AUD results, but the transactions-based measure shows no clear direction of response (at least over the initial months). The descriptive graphs (Figures 11, 12 and 13) suggest that hedging behaviour, especially for USD exposures, may have changed between the first and second halves of the sample. We examine this possibility by presenting the results for the value-based hedging measure for the two sample halves.³⁸

The first half-sample results for AUD and USD exposures are presented as Figures A5 and A6 respectively. Figures A7 and A8 present the second half-sample IRFs. Over the first half of the sample, both the AUD and USD responses are significantly negative, with both responses peaking three months after the exchange rate change. In the second half of the sample, the AUD response is again significant (peaking after two months), but the USD response (which is now slightly positive) is well within the confidence bounds around zero. Together, these results indicate that exporters with AUD exposures continue to adopt selective hedging positions, but those with USD exposures no longer do so in a material fashion.

One possible criticism of our exchange rate measure is that the HP filter uses actual future values of the rate in calculating the trend and cycle series. We test the robustness of our results by calculating backward looking measures of exchange rate misalignment. Specifically we subtract from each exchange rate the lagged one year mean, three year mean, five year mean and ten year mean respectively of that exchange rate. These rates are denoted AUD1, AUD3, AUD5, AUD10, USD1, USD3, USD5 and USD10. For instance, in period 2003:1, AUD1 equals AUD/NZD in 2003:1 less the mean value of AUD/NZD over the twelve months from 2002:1 to 2002:12.

The previous results are robust to using these backward-looking measures. Figures

A9 - A12 present the IRFs for AUD value-based hedge ratios to AUDNZD shocks, using each of the four backward-looking measures for the full period. The significance and shape of the responses is very similar throughout. The strongest relationship is obtained using the three year lag. Figures A13 – A16 present the IRFs for the USD value-based hedge ratios to shocks to USDNZD shocks, using each of the four backward-looking measures. We present these results just for the first half-sample consistent with the prior finding that USD selective hedging was not apparent in the second half-sample. The significance and shape of the responses is again very similar throughout. The strongest relationships in this case are obtained using the longer lags (five and ten years).³⁹

The difference in backward-looking lag structure between the AUD and USD may reflect the differing time series properties of the two exchange rates. As shown in Figures 11 and 12, deviations of USDNZD from its sample mean are much larger and longer than is the case with AUDNZD. (This is reflected also in Table 5, where the unit root is rejected at p=0.249 for AUDNZD compared with p=0.798 for USDNZD.) Selective hedgers may therefore consider that if cycles are mean-reverting, the adjustment is much faster for the AUDNZD than for the USDNZD.

Hitherto, we have not considered forward points explicitly in the analysis. While there is no theoretical case for hedging ratios to change when forward points change (in an efficient market), we did find in relation to Figures 14 and 15 that both AUD and USD hedge ratios have been positively correlated with the respective forward point level. When we include the forward points measures in the VAR specifications (with alternative exchange rate measures) we find no case (whether expressed as raw or filtered data) where the forward points effect is remotely significant for either currency. For instance, Figure A17 presents the full period response of the AUD value-based hedge ratio to forward points (fpnzau). The lack of response to forward points implies that any forward points effect on hedging is correlated with, and swamped by, the effect of the exchange rate cycle on exporters' behaviour.

Does Selective Hedging Work?

Finally, we examine whether selective hedging reflects an ability on the part of exporters who adopt the practice to predict exchange rate movements. Specifically we examine whether variations in the hedge ratio from its trend value "predicts" future exchange rate changes. To do so, we estimate the following regression for each hedge ratio and currency, for the full sample and for each split-sample period:

Δlog(ER_t) = β₀ + β₁*HEDGE_t-1 + β₂*HEDGE_t-2 + β₃*HEDGE_t-3 + ε_t (1)

where ER is variously AUDNZD and USDNZD; and HEDGE is variously AUDhr_tra_cyc, AUDhr_val_cyc, USDhr_tra_cyc, USDhr_val_cyc.

For each regression, we test the joint significance of β₁, β₂ and β₃ (using the equation F-statistic with the null hypothesis: β₁=β₂=β₃=0). Note, that unlike the VAR specification or a Granger causality test, we do not include lagged exchange rates in (1). The reason is that we are testing whether exporters benefit by selective hedging, not whether the hedge ratio predicts future exchange rate changes over and above what can be explained by past changes. Indeed, if past exchange rate changes help to predict future exchange rate changes, selective hedging may be a profitable strategy.

Table 6 presents the sign of the sum of the β coefficients which should be positive if selective hedging is a profitable strategy. We also present the p-values for the t-tests on the individual β coefficients and for the F-test for the combined effect of the βs; the Adjusted R² is presented as a measure of explanatory power of the hedge ratio for future exchange rate changes. Results are presented for the full sample and for the two sample halves.

The results in Table 6 indicate unambiguously that selective hedging has not been successfully practiced by New Zealand exporters as a whole. The only significant t-statistics and F-statistics have the wrong sign for the variables in question. In these cases, on average, hedging has risen (fallen) just prior to an exchange rate fall (rise). In all other cases, no significant relationship is obtained. Explanatory power of the equation (Adj. R²) is uniformly low (and sometimes negative), with the highest value for an equation having the correctly signed variables being 0.019. These results are robust across samples, across hedge ratio measures and across currencies.