RDP 2016-05: Trade Invoicing Currency and First-stage Exchange Rate Pass-through 1. Introduction

The depreciation of the Australian dollar over the past few years has renewed interest in understanding the pass-through of exchange rate changes to consumer prices (e.g. RBA 2014). Exchange rate pass-through can be divided into two stages: first-stage pass-through from exchange rate changes to across-the-docks Australian dollar import prices; and second-stage pass-through from import prices to consumer prices. Recent evidence using aggregate time series data has found that first-stage pass-through in Australia is around 80 per cent, much of which occurs immediately (Chung, Kohler and Lewis 2011). We investigate whether and how the currency in which goods imports are invoiced affects first-stage pass-through.

For homogenous goods traded in perfectly competitive markets (e.g. commodities), deviations from the law of one price are small, and trade invoicing currency is irrelevant: first-stage pass-through is immediate.^[1] But for differentiated goods (e.g. manufactures), prices are typically sticky in their currency of invoice, and thus currency invoicing mechanically affects short-run pass-through: it is 100 per cent for foreign currency-invoiced trade and zero for local currency-invoiced trade. Local currency invoicing dampens the short-run impact of exchange rate changes on consumer prices and reduces the expenditure-switching role of a flexible exchange rate (Devereux and Engel 2003).

While currency invoicing mechanically affects pass-through over the period for which prices are fixed, pass-through is the same for local and foreign currency-invoiced trade when prices are changed in standard models. This is because the invoicing currency is typically assumed to be exogenous, or unrelated to desired pass-through. However, Gopinath, Itskhoki and Rigobon (2010) estimate that long-run pass-through of exchange rate changes into US dollar import prices is only 17 per cent for imports invoiced in US dollars. Using price-level data, they show that even following a price change pass-through for US dollar-denominated goods imports is just 25 per cent. In contrast, for non-US dollar-invoiced goods, Gopinath et al (2010) estimate long-run pass-through to be 98 per cent. This implies that pricing-to-market behaviour is important, resulting in mark-up variation and persistent deviations from the law of one price.

Gopinath et al (2010) develop a model of endogenous currency choice that can explain why their estimates show long-run pass-through to be related to invoicing currency. In their model, firms choose whether to invoice in the local or the producer currency based on the magnitude of their desired pass-through over the duration that prices are fixed. Firms with low desired pass-through choose to invoice in local currency, while those with high desired pass-through will tend to invoice in producer currency. Critically, firms with low desired long-run pass-through are likely to also have low desired short-run pass-through, and so choose to invoice in local currency. Similarly, firms with high desired long-run pass-through are likely to also prefer high desired short-run pass-through, and so invoice in foreign currency. Thus, trade invoicing currency may serve as a sufficient statistic for desired long-run pass-through of exchange rate changes into import prices.

With endogenous choice of trade invoicing currency, the proposition that local currency pricing generates inefficient price dispersion over the duration that prices are fixed is weakened, because firms choosing to invoice in local currency are those that would absorb a large share of exchange rate changes in mark-ups even in a flexible price world. Desired pass-through is likely to be low for overseas exporters competing against local producers whose costs may vary relatively little with the exchange rate. Local currency invoicing mimics desired mark-up variation in response to exchange rate changes.

Existing Australian work either ignores the role of trade invoicing currency, or has implicitly assumed that invoicing currency is unrelated to firms' desired pass-through, in which case long-run pass-through is unrelated to invoicing currency. There are important differences between the structure of trade invoicing in Australia and the United States, suggesting that the findings of Gopinath et al (2010) may not apply. Firstly, in the United States around 90 per cent of imports are invoiced in local currency (US dollars) whereas in Australia only around 30 per cent of imports are invoiced in local currency (Australian dollars). Secondly, third-country invoicing is prominent in Australia but not in the United States: over 50 per cent of Australia's imports are invoiced in US dollars, despite only around 11 per cent of imports arriving from the United States.

Despite differences in the structure of invoicing between Australia and the United States, our results are consistent with Gopinath et al (2010). We estimate first-stage pass-through for foreign currency-invoiced goods to be immediate and complete, while pass-through for Australian dollar-invoiced imports is initially close to zero and is estimated to be only about 14 per cent after two years. The confidence intervals around our estimates are wide, but we can reject there being more than 50 per cent pass-through for Australian dollar-invoiced goods at a 95 per cent confidence level. These findings indicate that pass-through over a two-year period is close to bimodal, being approximately complete for foreign currency-invoiced trade and close to zero for Australian dollar-invoiced trade.

Our findings contribute to the Australian literature along several dimensions. First, the low degree of exchange rate pass-through for Australian dollar-invoiced trade provides evidence of large and persistent deviations from the law of one price. The aggregate import price series is the denominator in the terms of trade, so the less than one-for-one response of aggregate import prices to exchange rate changes implies that exchange rate changes caused by factors such as the stance of monetary policy can have a long-lived effect on the goods terms of trade. Second, our findings indicate that short- and long-run exchange rate pass-through is similar. Our findings of highly persistent deviations from the law of one price imply a very slow speed of adjustment in aggregate error correction models (ECMs) of import prices. Third, we show that invoicing currency serves as a sufficient statistic for firms' desired pass-through. Thus, variation in the share of Australian dollar-invoiced imports can be used to infer time-varying exchange rate pass-through to aggregate import prices.

Our paper relates most closely to Gopinath et al (2010), but also fits into a burgeoning literature studying variable mark-ups and international relative prices; see, for example, Knetter (1993) and Atkeson and Burstein (2008). This literature builds on seminal works by Dornbusch (1987) and Krugman (1987), who document the existence of pricing-to-market behaviour and identify imperfect competition and dynamic pricing considerations as explanations for incomplete pass-through. Our work also contributes to a large body of Australian evidence on first- and second-stage exchange rate pass-through, although to the best of our knowledge none of the previous work has investigated the role of currency invoicing. The most recent work is by Chung et al (2011), who estimate first-stage pass-through to be around 80 per cent, most of which occurs immediately.^[2]

The remainder of the paper proceeds as follows: Section 2 outlines the data we use; Section 3 explains our regression framework and reports our main findings; Section 4 investigates whether an invoice-share-weighted exchange rate index is preferable to a trade-weighted exchange rate index for modelling import prices; Section 5 discusses implications of our findings; and Section 6 concludes.

Footnotes

See, for example, Rauch's (1999) classification of goods into differentiated, reference-priced and those traded on organised exchanges. [1]

Previous research also includes work by Dwyer, Kent and Pease (1993) and Dwyer and Lam (1994). They use an ECM and find that first-stage pass-through is around 50 per cent in the quarter in which the depreciation occurs, and largely complete within a year; since the ECM they use imposes the law of one price, first-stage pass-through is complete in the long run. [2]