Chapter 3 Data
The data used in this paper are sourced from Statistics New Zealand's prototype Longitudinal Business Database (LBD), a firm-level database constructed from a range of administrative and survey sources linked to the Longitudinal Business Frame (see Fabling 2009). In particular, we draw on mandatory shipment-level filings of merchandise trade data provided by the New Zealand Customs Service. This data covers the period from April 2004 to December 2010.[6]
Our unit of observation is the firm-good-destination relationship. This detailed indexing minimises the extent to which results are affected by changes in the composition of traded goods, as we consider only price changes within a specified category of good by an individual seller. Goods are defined using the highly detailed ten-digit Harmonised System (HS10) classification[7] and unit values are calculated as the monthly free-on-board value in New Zealand dollars divided by the quantity exported:
where f,c,g,t index the firm, destination, good and month respectively. The reported invoice currency value is converted to New Zealand dollars using monthly exchange rate information from the International Monetary Fund's International Financial Statistics. Quantities are measured in standard units that are time-invariant and good-specific (eg, kilograms, litres, or counts).[8]
We observe almost 1.8 million price levels reported by 14,415 exporters.[9] These observations span 164 export destinations and 8,072 distinct goods, giving a comprehensive picture of New Zealand exporter behaviour.[10]
Changes in unit values are calculated across two time horizons. The short-run change is defined as the log difference of two consecutive unit values, adjusted for the number of months (Mt) between trades
The long-run change in unit value (ΔLRPfcgt) is defined across the lifetime of the good by taking the log difference between the first and last observed unit value within the relationship, following Gopinath et al. (2010), and again adjusting for the number of months between the first and last trade.
Fabling & Sanderson (2010) show that many export relationships at the firm-good-destination level are short-lived. Conversely, some firms may export only intermittently, leading to large gaps between consecutive trades. To prevent long-run ERPT estimates from being affected by short-lived relationships and vice versa, we place restrictions on the gap between trades to be included in each calculation. In order to be included in short-run calculations, consecutive trades must be no more than 6 months apart (ie, Mt≤ 5). Symmetrically, to be included in long-run calculations a relationship must span at least six months, when measured between the first and last observed trades. These restrictions lead us to drop 16.7 percent of ΔSRPfcgt observations, and 21.9 percent of ΔLRPfcgt observations.[11]
Over the analysis period, the majority of firms (55%) trade in only a single currency. Where firms trade in multiple currencies within a relationship in a month, we allocate the monthly observation to a predominant currency according to the share of (NZD-converted) trade value associated with that currency in the month.[12] We then drop ΔSR observations where consecutive unit values are in different predominant invoice currencies. ΔLR observations are dropped if any ΔSR in the relationship has been dropped. Having coded relationships to invoice currencies, we then distinguish between three invoice currency groups in subsequent analysis: producer currency (NZD), local currency (the currency of the destination country), and vehicle currencies (primarily the USD).
Alongside information on unit values, we make use of a range of firm- and product-level characteristics to examine heterogeneity in exchange rate responses. The choice and definition of these is discussed in more detail in section 5.
Notes
- [6] While earlier Customs data are available, April 2004 saw the introduction of mandatory electronic filing of Customs returns, including the comprehensive invoice currency information required for this analysis.
- [7] HS10 classifications are concorded over time by grouping together codes which merge or split at any time over the sample period.
- [8] For a small proportion of trade, quantity units are not defined by the classification system – primarily because the span of goods in the ten-digit code is not thought to be homogeneous enough to be covered by a single unit of measurement. In such cases we use the shipment weight to derive the volume measure or, where this is not possible, drop observations.Section 4.2 provides support for these product groups being sufficiently homogeneous within firms to be included in the analysis.
- [9] All firm counts have been random rounded base three, and relationship counts have been graduated random rounded (base 100 for counts over 1,000) in accordance with Statistics New Zealand confidentiality requirements.
- [10] Filing is mandatory for all shipments over NZD1,000 in value. We lose a small proportion of trade associated with destinations without published macroeconomic data.
- [11] Including all longer-term trades in the short-run analysis or restricting the long run to relationship lifetimes of one year or longer has no significant effect on the main estimates.
- [12] 96 percent of observed short-run price changes involve only a single currency of denomination, while the remaining four percent are allocated to a predominant currency. On average, the predominant currency accounts for 78 percent of the monthly trade value. Even at the 25th percentile, 65 percent of value is in the predominant currency, suggesting that this aggregation is unlikely to affect any results.
