Estimating the impacts of US tariffs on UK exports of single malt Scotch whisky

4 Data Selection

4.1 Sources

The monthly value (in GBP) and quantity (in litres of pure alcohol) of single malt exports for January 2010 – December 2020 are obtained from HMRC's Overseas Trade Statistics^[34]. The OECD database was used to obtain quarterly nominal private consumption expenditure for Q1 2010 – Q4 2020, monthly exchange rates and monthly long-term interest rates for January 2010 – December 2020, as well as annual per-capita alcohol consumption between 2010-2018^[35]. The distance between London and partner countries' capitals was obtained from CEPII^[36] and serves as a proxy for freight and time-associated costs. Annual population data between 2010-2020 was obtained from OECD while tariff data was obtained from WTO.

Some variables, including export value and quantity, were available at a monthly frequency, while two predictors were not: quarterly private consumption expenditure and annual population. Population itself is not used as a predictor (see below); it is used to transform other variables into per capita values.

For this reason, both quarterly and monthly specifications were explored. In each of these, population interpolation to a quarterly or monthly frequency was also explored (see Table A4 and Table A5 in the Annex). Additionally, disaggregation of quarterly consumption was also explored in the case of the monthly specification.

4.2 Variable Selection and Transformation

4.2.1 Dependent variable(s)

• Export value and quantity: Since the synthetic control will be a weighted average of the donor pool (in terms of the dependent variable), export values and quantities were scaled before analysis – this consisted of using per capita export values/quantities as opposed to total values/quantities.

These values and quantities were then logged due to periods of large variances (mostly in countries with smaller populations, who in some months import small amounts of single malt). Months with export values/quantities of £0 or 0 LPA for a given country were imputed by the smallest monthly value in that country's time series.^[37]

4.2.2 Population

• Population: This was not used as a predictor or dependent variable but rather to transform values into per-capita values where possible. Annual population data consisted of historical data (2010-2018) and projections (2019-2020). This data is based on mid-year population estimates – for the sake of this analysis, these are assumed to be in June. Linear extrapolation (forwards and backwards) to cover January 2010 to December 2020 was explored (to avoid step-changes having disproportionally large impacts on results) but this was found to have no major impact on final results.

4.2.3 Independent variables (predictors)

• Private consumption: nominal final private consumption data was available from OECD on a quarterly basis (non-seasonally adjusted). Consumption data was chosen over GDP since this was deemed to have a more 'direct' impact on exports or imports of a final consumption good such as single malt.

Consumptionvalues were transformed into per-capita consumption values using the population variable mentioned above and transformed into GBP by using the local-USD and USD-GBP exchange rates.

In the case of the monthly specification, disaggregation of quarterly values into monthly values was also explored using the tempdisaggpackage in R, described in more detail in Sax and Steiner (2013).^[38] This disaggregation, like the interpolation of population values, was also not found to have a major impact on final estimates. ^[39]

• Exchange rate: values expressed in a local currency were transformed to GBP using the monthly average nominal exchange rate (obtained from the OECD). This was done in two stages: transforming local currency to USD, and then transforming USD to GBP (there was no comprehensive database of GBP to local currencies on the OECD database). The local-GBP exchange rate itself was also used as a predictor.
• Alcohol consumption: alcohol consumption data was only available on an annual basis from 2010 to 2018 (litres per capita per year). This was averaged over 2010-18, resulting in a time-invariant predictor.
• The distance between London and the export destination's capital was also used as a time-invariant predictor.
• Four different lagged dependent variable specifications were explored: no lag, the first lag of one month (or one quarter), a seasonal lag of twelve months (or four quarters), or both the first and seasonal lags. Results were similar in most lagged specifications, with the best pre-tariff fit most commonly found in the first-lag specification.
• A monthly or quarterly dummy was included in the no-lag and first-lag specifications to account for any seasonality in single malt exports not accounted for by seasonality in consumption.

Comparisons of final model results with disaggregated consumption data and interpolated population data (for the per capita variables) are included in the Annex.

4.3 Donor Pool Selection

4.3.1 Tariffs and alcohol tax policy changes

The collection of countries considered in this analysis consists of OECD nations without a change in tariff on any whisky imports from the UK, including Scotch whiskies, and full data availability. Sixty-nine countries (out of 152) had no change in tariffs during the pre-treatment period.^[40] Thirty of these nations had full data availability in both the outcome and predictor variables (the US also had this).

Estonia and Latvia were both removed from these thirty countries because exports to those nations were likely affected by major changes in alcohol taxes in Estonia. This is especially evident when looking at single malt export value percapita (Figure 5).^[41]

This left 28 countries in our donor pool which all had (a) no change in the whisky tariff in the pre-treatment period and (b) full data availability. This donor pool plus the United States accounted for 71% of single malt exports in 2018.^[42]

lines and Canada constitute the final donor pool. Figure 7 shows the same using quarterly data, where the similarity in trends and seasonality between the US and the donor pool becomes more apparent. The synthetic control can draw on the full range of the donor pool to more closely approximate exports to the US (as opposed to a simple average).

4.3.2 Covid-19

Covid-19, a major contributor to the dampened demand for food and drink exports during 2020, may not have impacted our donor pool and the US in a similar way.

Comparing Covid-19 cases in the US to our donor pool average suggests cases were particularly high in the US during late 2020. However, the Oxford COVID-19 Government Response Tracker^[43] suggests that the US' government response to Covid-19 was within the range covered by our donor pool and exceeding the donor pool average from May to December 2020.