5.1 Short-run exchange rate pass-through

We closely follow Gopinath et al. (2010), modifying their approach to account for the change in perspective from import to export pass-through. Specifically, our simplest regression is of the form

where the log change in NZD-converted unit values within a specific firm-country-good relationship (ΔP_fcgt) is regressed on the cumulative (normalised by M) log difference in the bilateral exchange rate with the destination country (Δe_ct) since the last observed trade, and a set of control variables Z_cgt. Following Gopinath et al. (2010), Z includes destination×HS4-digit product dummies, and log changes in destination country GDP and CPI, and New Zealand CPI (all normalised by M).

β reflects the extent to which the NZD-converted unit value received by the exporter is influenced by the bilateral exchange rate. When β = 0, NZD-converted unit values are unaffected by the bilateral exchange rate. Conversely, when β = 1, unit prices in NZD respond one-for-one with the bilateral exchange rate so that the unit price in the importer's currency remains unchanged. More generally, β = 1 - β′ where β′ is the estimate generated from the importer perspective following the same Gopinath et al. (2010) approach.

To assess the potential importance of invoice currency to ERPT, we then allow β to differ according to the invoice currency

where δ^p, δ^l and δ^v are dummies indicating that the trade is invoiced in producer, local, or vehicle currencies respectively. Table 5 reports the sample size in each of these groups for both short-run and long-run populations. As discussed earlier, over half the Δ_SRP observations are producer-denominated, with 25 percent in local currency, and the remaining 20 percent in vehicle currencies (mainly USD).[18] In the long run, these ratios shift more towards producer-currency invoicing (71.7 percent of the sample), since less frequent traders tend to price in NZD. In both the short and long run, the restricted sample of firms with lagged trade data has very similar currency composition to the full sample.[19]

Table 5 : Sample size for ERPT regressions by invoice currency group

	Short run				Long run
	All		Lagged trade		All		Lagged trade
Producer	662,400	0.549	558,700	0.543	82,400	0.717	69,500	0.702
Local	303,800	0.252	262,100	0.255	15,300	0.133	13,800	0.139
Vehicle	240,900	0.200	208,300	0.202	17,300	0.150	15,700	0.159
Total	1,207,100	1.000	1,029,100	1.000	115,000	1.000	99,000	1.000

Reported sample size and invoice currency shares for columns labelled "All" relate to short-run (Table 6) and long-run (Table 11) regression samples. "Lagged trade" columns provide the same statistics for thepopulation for which lagged export values are available (used in tables 8 and 12, respectively).

Table 6 shows the results of estimating equations 5.1 and 5.3 for short-run unit value changes. Column 1 reports the average β across all currency groups, showing that 47.5 percent of the average bilateral exchange rate movement is absorbed into the NZD-converted unit value. Column 2 then allows β to vary along the lines of Gopinath et al. (2010), into trades invoiced in the producer currency and in other currencies. This specification represents an intermediate stage between equations 5.1 and 5.2, produced solely for comparison with Gopinath et al. (2010) (their table 2). Since that paper produces parameters from the importer's perspective, the comparable coefficients to our β_producer (0.09) is one minus their coefficient for non-USD trades (1 - β_ND, or 1-0.92=0.08). For trade invoiced in non-NZD, the coefficient (0.80) is comparable to Gopinath et al.'s coefficient on USD trade (1 - β_D, or 1-0.24=0.76). In both cases, pass-through behaviour of New Zealand exporters to all countries is quite consistent with the previously estimated pass-through behaviour of exporters from all countries to the United States.

Table 6 : Short-run ERPT by invoice currency group

	(1)	(2)	(3)	(4)
β	0.475** [0.015]
βnon-producer		0.804** [0.021]
βproducer		0.092** [0.022]	0.092** [0.022]	0.086** [0.022]
βlocal			0.909**	0.901**
			[0.029]	[0.029]
βvehicle			0.700** [0.029]
βvehicle(p/v)				0.825** [0.030]
βvehicle(v/l)				0.065 [0.047]
N(ΔSR)	1,207,100	1,207,100	1,207,100	1,207,100
Within R2	0.013	0.013	0.013	0.013

Regressions include unreported HS4-destination fixed effects and macroeconomic variables as outlined in themain text. Standard errors in brackets (** denotes significance at the 1% level).β coefficients all significantly different from each other at the one percent levelwith the exception of β_producerand β_vehicle(v/l)in column 4 (p-value 0.682).

Column 3 represents our first step beyond Gopinath et al. (2010), where we now allow separate coefficients for each currency group (as in equation 5.2).[20] Allowing this distinction it is apparent that β is higher for contracts invoiced in the local currency than the vehicle currency – with coefficients of 0.909 and 0.700 respectively, significantly different at the one percent level. If, as Table 4 implies, nominal prices are most sticky in the invoice currency, this would explain the relatively lower coefficient on β_vehicle compared with β_local. To test this hypothesis, we note that, in the case of vehicle currency use, the bilateral exchange rate movement can be decomposed into two (log) additive parts

where we replace the destination index c by the appropriate currency indexes. Given this decomposition we estimate

so that pass-through for vehicle currency users has two components: one related to the bilateral exchange rate between the producer and vehicle currencies β_vehicle(p/v), and one related to the exchange rate between the vehicle and local currencies β_vehicle(v/l). If stickiness in the invoice currency is an important factor, and these exchange rates have a degree of independent movement, then we would expect β_vehicle(p/v) to be the higher of the two coefficients. Table 7 shows the correlation between observed exchange rate movements, conditional on the use of a vehicle currency.[21] Δe_(p/l)t and Δe_(p/v)t are positively correlated (coefficient of 0.774), but not perfectly so. Thus, when we estimate equation 5.3 (column 4 of Table 6 ), we see a stronger ERPT coefficient on the bilateral exchange rate between the producer and vehicle currencies - implying a consistent story across all currency groups, that the main driver of short-run unit value fluctuations is nominal stickiness in the contract currency.

Table 7 : Exchange rate correlations, conditional on vehicle currency usage

	Δe(p/l)t	Δe(p/v)t	Δe(v/l)t
Δe(p/l)t	1.000
Δe(p/v)t	0.774	1.000
Δe(v/l)t	0.244	-0.426	1.000

Δe_(x/y)tis the change in the exchange rate betweencurrency groups x and y. Currency groups areproducer (p), local (l) and vehicle (v).

Having established the importance of controlling for invoice currency, we now turn our attention to the question of whether ERPT differs systematically with characteristics of the firm or the exported product. Berman et al. (2012) find consistent and significant differences in pass-through behaviour between high- and low-productivity firms, with high-productivity exporters absorbing a greater share of exchange rate changes into their margins than less productive firms.

We revisit the Berman et al. (2012) findings by generating binary indicators for high export performance or other characteristics δ¹ which we then interact with the currency dummies, giving a total of six distinct exchange rate coefficients (two performance groups by three invoice currency groups)

Firm characteristics are mainly based on lagged firm-level export data,[22] and reflect various elements of export performance, diversity, and/or potential hedges. Firstly, on the basis that total export value is correlated with firm performance (Eaton et al. 2011), we compare relationships according to whether the firm has relatively high or low export revenue. Additional measures of export performance includes lags of the number of destinations, discrete goods exported, and currencies in which exports have been invoiced. We also consider the degree of diversity in lagged export receipts, as measured by the reciprocal of the Herfindahl-Hirschman index of concentration for destinations, goods and currencies (ie, "high" group firms have more diverse trade). As well as being correlated with export performance, use of a diverse mix of currencies may also provide firms with a form of natural hedge if exchange rate fluctuations are imperfectly correlated across currencies. For each measure we generate a binary indicator of whether the firm is above or below the currency group-specific median of the performance metric.[23]

Heterogeneity in ERPT is also considered for three further characteristics: whether the firm has ever explicitly hedged their export exchange rate risk, whether the export is a differentiated good according to Rauch (1999),[24] and whether the firm is under foreign ownership (FDI). If firms use hedging to insulate themselves from exchange rate shocks, we might expect to see less price adjustment in the foreign contract currency (ie, higher βs) among firms with hedging experience. Alternatively, explicit hedging may indicate a firm is particularly sensitive to NZD-denominated price fluctuations, leading it to adjust prices more quickly than other firms (yielding lower βs).

If differentiated goods face a lower elasticity of demand than commodities, we might expect to see higher βs among "commodity" exporters as they attempt to stabilise prices with respect to their competition. Finally, if some part of the exports of foreign-owned firms are destined for their foreign parent and the firms are able to take advantage of transfer pricing, or alternatively if membership of multinational organisations provides a degree of implicit hedging, we might expect to see higher βs among foreign-owned firms.

Table 8 shows the results of this analysis, including tests of whether the invoice currency coefficients differ by characteristics. In contrast to Berman et al. (2012), we see almost no evidence of differences in ERPT associated with firm performance. Only three pairs of coefficients differ at the five percent level, all among trades denominated in the producer currency (NZD), where the high group shows no response to the currency change (β_producer¹ insignificantly different from zero) and the low group displays a positive response (β_producer⁰).

Table 8 : Short-run ERPT by invoice currency group and firm characteristics

	Total exports	Number of countries	fx rates	goods	Herfindahl-1 countries	fx rates	goods	Prior hedging	Differentiated goods	FDI
βproducer0	0.137**† [0.034]	0.105** [0.033]	0.106** [0.031]	0.099** [0.031]	0.106** [0.034]	0.145**† [0.033]	0.074* [0.031]	0.157**‡ [0.036]	0.113** [0.036]	0.109** [0.028]
βproducer1	0.045† [0.032]	0.070* [0.033]	0.065 [0.035]	0.073* [0.035]	0.072* [0.032]	0.034† [0.032]	0.104** [0.035]	0.037‡ [0.030]	0.069* [0.030]	0.039 [0.042]
βlocal0	0.885** [0.046]	0.938** [0.046]	0.856** [0.046]	0.888** [0.042]	0.890** [0.046]	0.863** [0.045]	0.896** [0.042]	0.867** [0.078]	0.901** [0.045]	0.935** [0.035]
βlocal1	0.913** [0.040]	0.871** [0.041]	0.934** [0.040]	0.913** [0.043]	0.908** [0.040]	0.930** [0.041]	0.905** [0.044]	0.906** [0.033]	0.899** [0.041]	0.804** [0.059]
βvehicle0	0.688** [0.044]	0.720** [0.043]	0.677** [0.040]	0.709** [0.042]	0.724** [0.044]	0.694** [0.043]	0.695** [0.043]	0.836** [0.110]	0.698** [0.038]	0.743** [0.036]
βvehicle1	0.723** [0.041]	0.694** [0.042]	0.745** [0.046]	0.704** [0.043]	0.691** [0.042]	0.719** [0.042]	0.718** [0.043]	0.696** [0.031]	0.721** [0.049]	0.620** [0.056]
N(ΔSR)	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100
Within R2	0.014	0.014	0.014	0.014	0.014	0.014	0.014	0.014	0.014	0.014

Regressions include unreported HS4-destination fixed effects and macroeconomic variables as outlined in the main text. Standard errors in brackets (** ; * denotessignificance at the 1%; 5% level respectively). ‡; † signify that low-high coefficient pairs are significantly different from each other at the 1%; 5% level respectively.

To relate these results to those of Berman et al. (2012), we re-estimate equation 5.4 constraining β coefficients to be the same across invoice currencies

These estimates (Table 9 ) follow the pattern observed by Berman et al. (2012). High-performance firms show a significantly higher degree of exchange rate absorption (ie, lower ERPT from the importer's perspective) across most binary measures.

Table 9 : Short-run ERPT by firm characteristics only

	Total exports	Number of countries	fx rates	goods	Herfindahl-1 countries	fx rates	goods	Prior hedging	Differentiated goods	FDI
β0	0.379**‡ [0.024]	0.377**‡ [0.023]	0.307**‡ [0.023]	0.468** [0.022]	0.400**‡ [0.024]	0.307**‡ [0.023]	0.452** [0.022]	0.324**‡ [0.031]	0.516**‡ [0.023]	0.512**‡ [0.019]
β1	0.540**‡ [0.021]	0.555**‡ [0.021]	0.630**‡ [0.022]	0.480** [0.023]	0.529**‡ [0.021]	0.612**‡ [0.021]	0.498** [0.023]	0.525**‡ [0.018]	0.433**‡ [0.022]	0.382**‡ [0.029]
N(ΔSR)	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100	1,029,100
Within R2	0.014	0.014	0.014	0.014	0.014	0.014	0.014	0.014	0.014	0.014

Regressions include unreported HS4-destination fixed effects and macroeconomic variables as outlined in the main text. Standard errors in brackets (** denotes significanceat the 1% level). ‡ signifies that low-high coefficient pairs are significantly different from each other at the 1% level.

Table 10 provides the bridge between tables 8 and 9, comparing the distribution of invoice currency for each "high" performance group with the full population distribution (bottom row). Firms with high export receipts, exporting to multiple countries, exporting in a range of different currencies and with experience of exchange rate hedging - the same characteristics which are associated with high βs in Table 9 - tend to be over-represented in vehicle and local currency use, and under-represented in producer currency use. In contrast, where the "high" group includes an above average share of producer currency invoicing - foreign-owned firms and those trading in differentiated goods - the relationship is reversed, with β⁰ higher than β¹. That is, "good" exporters tend to absorb a greater proportion of exchange rate movements in the short run, and this difference is related to their greater tendency to trade in foreign currencies, rather than because of within-currency group differences in ERPT.

While binary performance indicators provide a blunt test of the relationship between firm characteristics and ERPT, figure 2 provides an indication of the extent to which this relationship differs across the performance distribution, plotting βs estimated separately for each decile of lagged export value. These coefficients show the same basic patterns reported for the binary groups in tables 8 and 9. When invoice currency is ignored (panel D), we see mild evidence of a positive relationship between exporter size and the degree of ERPT, driven in part by significantly higher β in the top decile of export receipts. However, within currencies, this relationship disappears, with no clear correlation between export value and βs.

Table 10 : Short-run invoice currency shares for "high" characteristic groups

	Producer	Local	Vehicle
Total exports	0.402	0.276	0.322
Number of:
countries	0.410	0.274	0.315
fx rates	0.293	0.377	0.330
goods	0.560	0.246	0.194
Herfindahl-1:
countries	0.444	0.248	0.308
fx rates	0.343	0.394	0.263
goods	0.546	0.244	0.211
Prior hedging	0.436	0.299	0.265
Differentiated goods	0.587	0.263	0.150
FDI	0.592	0.215	0.193
Overall	0.543	0.255	0.202

Figure 2 - Short-run ERPT by export decile

β coefficients estimated separately for each decile of lagged export value, via expanded versions of equations 5.4 (panels A-C) and 5.5 (panel D). Deciles are calculatedacross the full sample, rather than within currency groups. Vertical lines represent 95 percent confidence intervals centered on point estimates. The solid horizontal lineshows average β estimates from Table 6 (column 3 for panels A-C and column 1 for panel D).