Estimating the impacts of US tariffs on UK exports of single malt Scotch whisky
This discussion paper aims to estimate the impact of US tariffs on UK exports of single malt Scotch whisky between Q4 2019 – Q4 2020 using the novel synthetic control method.
Annex A – Donor Pool and Model Specifications
Donor Pool
The table below shows the countries in the donor pool. The choice was based on data availability and tariff. Any country with missing data for any of the predictors was discounted, as was any country with a change in tariff during Jan 2010 – Dec 2020.[55] Table A1 shows the countries in the final donor pool.
Tariff | Donor Pool (25+3 Control Countries) | |||
---|---|---|---|---|
No | Austria | Germany | Netherlands | South Africa |
Belgium | Hungary | New Zealand | Spain | |
Canada | Iceland | Norway | Sweden | |
Czechia | Ireland | Poland | Switzerland | |
Denmark | Italy | Portugal | ||
Finland | Japan | Slovakia | ||
France | Lithuania | Slovenia | ||
Yes (no change) | Australia (ad-valorem equivalent tariff of 5%) | |||
Chile (6%) | South Korea (20%) |
The total number of countries initially considered for the donor pool was 152. The majority of these (126) were indeed nations without any change in whisky tariff between 2010 – 2019. This was further narrowed down to 28 countries due to data availability of predictors (see Table A2).
These 28 countries accounted for 44.6% of total single malt export value in 2018, with the United States accounting for a further 26.4%. In total, this represented 71.0% of total exports in 2018; 16% came from countries not selected due to limited data availability, and the remaining 13% came from countries not selected due to changes in tariffs during 2010-2020 (or major changes in alcohol duty policies in the case of Latvia and Estonia).
Note that tariff data did not include tariffs introduced as a result of trade disputes, only those agreed upon as part of goods schedules in bilateral or multilateral trade negotiations (e.g. the 1994 WTO Uruguay Round or so-called '1980 procedures') . This means that even the United States was reported to have a 0% ad-valorem equivalent tariff for the HS6 category '220830 – Whiskies' in the WTO tariff data.
Data on tariffs introduced as a result of trade disputes is scarce. The trade disputes available on WTO's trade dispute gateway[56] was carefully examined for trade disputes where the respondent was either (i) the United Kingdom or (ii) the European Union (pre-2021), and retaliation was granted at any point between 2010-2020.[57] The only trade dispute that met these criteria was DS316, the Airbus-Boeing dispute which resulted in tariffs on single malt.
Countries | Value (£m) | Prop. (%) | |
---|---|---|---|
Donor pool | 28 | 583 | 44.6 |
United States | 1 | 344 | 26.4 |
Total selected | 29 | 927 | 71.0 |
Not selected due to data availability | 69 | 207 | 15.8 |
Top five export destinations: | |||
08 Singapore | 1 | 83 | 6.4 |
12 China | 1 | 32 | 2.5 |
18 United Arab Emirates | 1 | 17 | 1.3 |
19 India | 1 | 17 | 1.3 |
25 Hong Kong | 1 | 9 | 0.7 |
Not selected due to changing tariffs | 54 | 173 | 13.2 |
Top five export destinations: | |||
06 Taiwan | 1 | 101 | 7.7 |
11 Latvia (tax policy change) | 1 | 39 | 3.0 |
24 Mexico | 1 | 10 | 0.7 |
28 Israel | 1 | 7 | 0.5 |
30 Vietnam | 1 | 6 | 0.4 |
Total not selected | 123 | 379 | 29.0 |
Total exports | 152 | 1,306 | 100.0 |
The average monthly values of each predictor and export value in the pre-tariff period (Jan 2010 to Oct 2019) is shown below.
Export Value |
Local Private Final Cons. | UK Private Final Cons. | Population (a) |
Interest Rate | Local-GBP Exchange Rate | Avg. Alcohol Consumption 2010-18 |
Distance Between Capitals | ||
---|---|---|---|---|---|---|---|---|---|
Unit | £ per capita | £000's per capita | £000's per capita | Millions | % | Local currency / £ | Litres per cap | 000's km | |
Frequency | M | Q | Q | A | M | M | .. | .. | |
US | 0.06 | 6.5 | 4.7 | 319.4 | 2.42 | 1.48 | 8.8 | 5.9 | |
1 | Australia | 0.06 | 5.7 | 4.7 | 23.6 | 3.32 | 1.74 | 9.9 | 17.0 |
2 | Austria | 0.04 | 4.4 | 4.7 | 8.6 | 1.51 | 1.20 | 12.3 | 1.2 |
3 | Belgium | 0.08 | 4.0 | 4.7 | 11.2 | 1.82 | 1.20 | 10.0 | 0.3 |
4 | Canada | 0.09 | 4.7 | 4.7 | 35.6 | 2.09 | 1.70 | 8.2 | 5.4 |
5 | Chile | 0.00 | 1.6 | 4.7 | 17.9 | 4.91 | 852.87 | 7.7 | 11.7 |
6 | Czechia | 0.01 | 1.7 | 4.7 | 10.6 | 1.97 | 31.27 | 11.6 | 1.0 |
7 | Denmark | 0.09 | 4.7 | 4.7 | 5.7 | 1.21 | 8.93 | 9.6 | 1.0 |
8 | Finland | 0.04 | 4.4 | 4.7 | 5.5 | 1.38 | 1.20 | 8.9 | 1.8 |
9 | France | 0.18 | 3.8 | 4.7 | 66.2 | 1.61 | 1.20 | 11.9 | 0.3 |
10 | Germany | 0.07 | 4.1 | 4.7 | 81.7 | 1.09 | 1.20 | 11.1 | 0.9 |
11 | Hungary | 0.00 | 1.2 | 4.7 | 9.9 | 4.91 | 361.91 | 10.9 | 1.5 |
12 | Iceland | 0.05 | 5.0 | 4.7 | 0.3 | 5.68 | 176.32 | 7.2 | 1.9 |
13 | Ireland | 0.01 | 3.9 | 4.7 | 4.7 | 3.22 | 1.20 | 11.1 | 0.5 |
14 | Italy | 0.03 | 3.5 | 4.7 | 60.2 | 3.23 | 1.20 | 7.3 | 1.4 |
15 | Japan | 0.01 | 4.0 | 4.7 | 127.1 | 0.47 | 148.16 | 7.2 | 9.6 |
16 | S. Korea | 0.01 | 2.4 | 4.7 | 50.8 | 2.96 | 1,652.65 | 8.8 | 8.9 |
17 | Lithuania | 0.02 | 1.7 | 4.7 | 2.9 | 2.58 | 1.20 | 13.6 | 1.7 |
18 | Netherlands | 0.18 | 3.9 | 4.7 | 16.9 | 1.36 | 1.20 | 8.6 | 0.4 |
19 | New Z. | 0.04 | 4.0 | 4.7 | 4.6 | 3.66 | 1.98 | 9.1 | 19.1 |
20 | Norway | 0.04 | 6.2 | 4.7 | 5.1 | 2.19 | 10.33 | 6.2 | 1.2 |
21 | Poland | 0.01 | 1.4 | 4.7 | 38.4 | 3.93 | 5.04 | 10.5 | 1.5 |
22 | Portugal | 0.02 | 2.4 | 4.7 | 10.4 | 4.82 | 1.20 | 10.4 | 1.6 |
23 | Slovakia | 0.00 | 1.7 | 4.7 | 5.4 | 2.19 | 1.20 | 10.0 | 1.3 |
24 | Slovenia | 0.01 | 2.2 | 4.7 | 2.1 | 2.92 | 1.20 | 10.5 | 1.2 |
25 | South Africa | 0.02 | 0.7 | 4.7 | 54.5 | 8.54 | 16.12 | 7.3 | 9.0 |
26 | Spain | 0.04 | 2.9 | 4.7 | 46.6 | 3.00 | 1.20 | 10.0 | 1.3 |
27 | Sweden | 0.14 | 4.4 | 4.7 | 9.8 | 1.37 | 11.27 | 7.2 | 1.4 |
28 | Switzerland | 0.13 | 7.6 | 4.7 | 8.2 | 0.46 | 1.41 | 9.6 | 0.7 |
(a) Population is not a predictor in itself – instead, it was used to convert export value (or quantity) and private final consumption into per-capita values. These predictors were subsequently logged.
Model Specifications
Data Transformation
The following data transformations and combinations were explored, each with either no lagged dependent variable, a first-lag dependent variable, a seasonal-lag dependent variable, or both a first- and seasonal-lag as a predictor (along with other predictors) – making for a total of 48 monthly specifications and 24 quarterly specifications (where only the first-lagged specifications with no consumption disaggregation or population interpolation are reported in the final report).
The price dependent variable was constructed by taking the monthly export value to a given country and dividing it by the monthly export quantity. For cases where the quantity was reported as zero, the next-smallest non-zero quantity for that country between 2010-2020 was used. Where both value and quantity were zero, a missing value was generated. None of the 29 countries had one of these missing values.
Dependent variable* |
Cons-umption(a) | Pop-ulation(b) | RMSPE(c) (prior to tariff introduction) | |||
---|---|---|---|---|---|---|
No Lag | First Lag | Seasonal Lag | First & S. Lag | |||
Value (£) | Q | A | 0.23836 | 0.19924 | 0.20298 | 0.19917 |
Value (£) | Q | IM | 0.23853 | 0.19918 | 0.20294 | 0.19912 |
Value (£) | DM | A | 0.23836 | 0.19924 | 0.20298 | 0.19917 |
Value (£) | DM | IM | 0.23853 | 0.19918 | 0.20294 | 0.19912 |
Quantity (LPA) | Q | A | 0.22948 | 0.16004 | 0.16304 | 0.16004 |
Quantity (LPA) | Q | IM | 0.22944 | 0.16007 | 0.16306 | 0.16007 |
Quantity (LPA) | DM | A | 0.22948 | 0.16004 | 0.16304 | 0.16004 |
Quantity (LPA) | DM | IM | 0.22944 | 0.16007 | 0.16306 | 0.16007 |
Price (£/LPA) | Q | A | 0.14120 | 0.12383 | 0.12499 | 0.12383 |
Price (£/LPA) | Q | IM | 0.14120 | 0.12383 | 0.12499 | 0.12383 |
Price (£/LPA) | DM | A | 0.14120 | 0.12383 | 0.12499 | 0.12383 |
Price (£/LPA) | DM | IM | 0.14120 | 0.12383 | 0.12499 | 0.12383 |
(a) Dependent variable was logged per capita for value/quantity, logged for price. (b) Key: Q = Quarterly; DM = Disaggregated into Monthly; A = Annual (mid-year); IM = Interpolated to Monthly. (c) Root Mean Squared Prediction Error (comparisons between different dependent variables should not be made). The model(s) including a seasonal lagged dependent variable did not include a monthly dummy. Rounded to five decimal places.
Dependent variable* |
Cons-umption(a) | Pop-ulation(b) | RMSPE(c) (prior to tariff introduction) | |||
---|---|---|---|---|---|---|
No Lag | First Lag | Seasonal Lag | First & S. Lag | |||
Value (£) | Q | A | 0.17703 | 0.11943 | 0.12009 | 0.11943 |
Value (£) | Q | IM | 0.17711 | 0.11923 | 0.11995 | 0.11923 |
Quantity (LPA) | Q | A | 0.14457 | 0.06657 | 0.06709 | 0.06656 |
Quantity (LPA) | Q | IM | 0.14434 | 0.06657 | 0.06710 | 0.06656 |
Price (£/LPA) | Q | A | 0.10643 | 0.09301 | 0.09453 | 0.09301 |
Price (£/LPA) | Q | IM | 0.10643 | 0.09301 | 0.09453 | 0.09301 |
(a)(b) See Table A4 for key; (b) The model(s) including a seasonal lagged dependent variable did not include monthly dummy, other specifications did. Rounded to five decimal places.
Interpolation and Disaggregation
Population interpolation and consumption disaggregation, as shown above, do not majorly affect the fit of the synthetic control (and where they do, this affects the post-tariff fit the same way and therefore the tariff impact estimates are similar). Figure A1 below shows this for the United States – quarterly consumption disaggregated into monthly values, and annual mid-year estimates interpolated to a monthly frequency. Population interpolation was also explored for quarterly specifications.
Contact
Email: agric.stats@gov.scot
There is a problem
Thanks for your feedback