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

6 Results

6.1 Overview

A wide range of model specifications were explored, many of which resulted in similar results. In general, the specifications using quarterly data provided a better pre-tariff fit than those using monthly data, which is why the quarterly results are presented as the primary sets of results and the focus of Sections 6.2, 6.3, and 6.4. Monthly export data may simply be too volatile (or vary too much between countries in the donor pool) to be able to accurately mimic exports to the US.

Frequency of data was not the only variation between specifications – value, quantity, or price as a dependent variable were all explored. Different lagged dependent variables were explored as predictors, as well as the interpolation and extrapolation of mid-year population estimates and disaggregation of quarterly consumption data (in the case of the monthly specifications).

This section shows an overview of results for the quarterly and monthly model specifications with a 'first' lagged dependent variable (i.e. the previous quarter or month) and no population interpolation or consumption disaggregation. Here, results are presented in per-capita terms. Aggregated impacts are presented in Section 0 – these show total tariff impacts over the course of the Q4 2019 – Q4 2020 post-tariff period.

Results for other lagged model specifications are shown in the Annex (see e.g. Table A4 for a list of all specifications explored, and Table D1 and Table D2 for results for these specifications).

6.1.1 Quarterly data

Results using quarterly export data are shown below, in Table 4. These show the average quarterly impact of the tariff on per-capita exports of single malt to the US, in terms of proportion of export value, quantity, and price. Confidence sets and p-values are also provided through the use of placebo tests (see Section 3.2.1).

Table 4. Results using a quarterly first-lagged specification

Dependent variable	Estimated quarterly tariff effect	p-value	26/29 confidence set (~90%)
	Proportion (%)		Lower (%)	Upper (%)
Value (£/cap)	-18.3**	0.069	-100.0	-4.7
Quantity (mLPA/cap)	-10.3**	0.069	-19.6	-9.5
Price (£/LPA)	-10.2	0.345	-64.7	+17.9

Without population interpolation. Key: *** p < 2/29; ** p < 3/29; * p < 4/29.

The confidence sets^[44] above suggest a strictly negative impact on both value and quantity, with the estimated impacts on export value and quantity given by [-100.0%, -4.7%] and [-19.6%, -9.5%], respectively. The estimated impact on average export price is non-significant and could be negative or positive as shown by the [-64.7%, +17.9%] confidence set.

6.1.2 Monthly data

Monthly specifications were also explored (using monthly trade data and monthly predictors where possible). Table 5 below shows the estimated per-capita impact of the tariff for the first-lagged monthly specification, along with 90% confidence sets.

Table 5. Results using a monthly first-lagged specification

Dependent variable	Estimated monthly tariff effect	p-value	26/29 confidence set (~90%)
	Proportion (%)		Lower (%)	Upper (%)
Value (£/cap)	-17.4	0.207	-87.0	+1.2
Quantity (LPA/cap)	-10.7	0.448	-34.9	+16.8
Price (£/LPA)	-11.9	0.414	-56.7	+3.8

Without population interpolation or consumption disaggregation

The results above suggest a less clear picture than those obtained by using quarterly data: here, we are no longer confident that the impact on export value or quantity was different from zero – despite similar point estimates. The impact on average export price is similarly non-significant. A stated earlier, this may be because monthly export data is too volatile (including to countries with zero or near-zero export values during some months).

6.2 Results by Quarter

6.2.1 Value

The synthetic control and the observed UK-US export values are shown in Figure 9. The 'gaps' between the actual and synthetic values before and after the introduction of the tariffs can be compared, where the difference in the average gaps would provide a (point) estimate of the impact of the tariff. This is shown in Table 6 (this uses the same first-lag quarterly specification as above).

The point estimates suggest that, across the post treatment period, the average value of per-capita export was roughly 18.3% less than what it otherwise would have been without a tariff (calculated from the log tariff impact^[45]).

Table 6. Comparing export value between actual and synthetic US

Period		Log Exports per Capita			Exports per Capita			Tariff Impact
		Natural log of £/capita			£/capita
		Actual	Synth.	Gap	Actual	Synth.	Gap	%
Pre-Tariff (Q1 2010 – Q3 2019):
	Average	-1.762	-1.770	+0.008	0.18	0.18	+0.00	..
Post-Tariff (Q4 2019 – Q4 2020):
	Q4 2019	-1.238	-1.152	-0.086	0.29	0.32	-0.03	..
	Q1 2020	-1.688	-1.593	-0.095	0.18	0.20	-0.02	..
	Q2 2020	-1.823	-1.866	+0.043	0.16	0.15	+0.01	..
	Q3 2020	-1.422	-1.141	-0.281	0.24	0.32	-0.08	..
	Q4 2020	-1.653	-1.105	-0.548	0.19	0.33	-0.14	..
	Average	-1.565	-1.372	-0.194	0.21	0.26	-0.05	..
Tariff Impact (Post-Tariff less Pre-Tariff):
	Average	0.197	0.399	-0.202	0.03	0.09	-0.06	-18.3

Showing rounded values

It is apparent from both Figure 9 and Table 6 that large gaps are seen during Q4 2020, and that Q2 2020 saw a marginally positive gap. These quarterly gaps do not take into account the average pre-tariff fit and should therefore not be interpreted as quarterly tariff impact estimates. Additionally, any quarterly tariff impact estimates would have their own associated p-values and confidence sets which are not presented here (see Annex C – Inference and Confidence Sets. Instead, confidence sets for the average quarterly tariff impact are presented in Section 6.1.1.

6.2.2 Quantity

All outputs reported for value above are available for quantity as well. Table 7 below shows that the tariff resulted in an estimated -10.3% change in the average quantity of single malt exported per quarter. Table 4 shows that this result is statistically significant (meaning if we replicate the synthetic control for other countries, we find very few similar estimated impacts of 'placebo' tariffs).

Table 7. Comparing export quantity between actual and synthetic US

Period		Log Exports per Capita			Exports per Capita			Tariff Impact
		Natural log of LPA/capita			mLPA/capita
		Actual	Synth.	Gap	Actual	Synth.	Gap	%
Pre-Tariff (Q1 2010 – Q3 2019):
	Average	-5.430	-5.433	+0.003	4.50	4.47	+0.03	..
Post-Tariff (Q4 2019 – Q4 2020):
	Q4 2019	-5.094	-5.175	+0.081	6.13	5.66	+0.48	..
	Q1 2020	-5.387	-5.344	-0.043	4.58	4.78	-0.20	..
	Q2 2020	-5.549	-5.495	-0.054	3.89	4.11	-0.22	..
	Q3 2020	-5.155	-4.980	-0.174	5.77	6.87	-1.10	..
	Q4 2020	-5.302	-4.961	-0.341	4.98	7.01	-2.02	..
	Average	-5.297	-5.191	-0.106	5.07	5.68	-0.61	..
Tariff Impact (Post-Tariff less Pre-Tariff):
	Average	0.133	0.242	-0.109	0.57	1.21	-0.64	-10.3

Showing rounded values

As with Table 6, the gaps presented in Table 7 should not be interpreted as tariff impact estimates in their own right. However, we can see a similar pattern where Q4 2020 is the quarter where the synthetic control and US deviate the most.

6.2.3 Price

Table 8 shows that the tariff resulted in an estimated -10.2% change in the average price of single malt exported. Additionally, the (negative) gaps are the largest in Q1 2020 and Q4 2020. However, as shown in Table 4, the estimated tariff impact is not statistically significant (meaning that if we replicate the synthetic control for other countries in our donor pool, we find that more than 10% of our total sample of countries show equal or larger estimated impacts of 'placebo' tariffs).

Table 8. Comparing export price between actual and synthetic US

Period	Log Price			Exports			Tariff Impact
	Natural log of £/LPA			£/LPA
	Actual	Synth.	Gap	Actual	Synth.	Gap	%
Pre-Tariff (Q1 2010 – Q3 2019):
Average	3.668	3.656	0.012	39.734	38.988	0.745	..
Post-Tariff (Q4 2019 – Q4 2020):
Q4 2019	3.856	3.839	0.017	47.28	46.48	0.80	..
Q1 2020	3.698	3.882	-0.183	40.38	48.50	-8.11	..
Q2 2020	3.726	3.764	-0.038	41.51	43.12	-1.61	..
Q3 2020	3.732	3.828	-0.096	41.77	45.97	-4.20	..
Q4 2020	3.648	3.825	-0.177	38.40	45.82	-7.42	..
Average	3.732	3.827	-0.095	41.868	45.977	-4.110	..
Tariff Impact (Post-Tariff less Pre-Tariff):
Average	0.064	0.172	-0.108	2.134	6.989	-4.855	-10.2

Showing rounded values

6.3 Sensitivity Analysis

6.3.1 Removing countries

The three countries with the most weight in the synthetic control were removed from the donor pool, one by one. Results for these estimations, using a quarterly first-lag specification^[46] without population interpolation, are reported in Table 9. This is similar in some ways to the sensitivity analysis conducted by Abadie et al (2015). The resulting confidence intervals for these estimations use a donor pool of 27 countries (as opposed to 28). This means that a 26/29 confidence set cannot be constructed any longer – instead, a 25/28 confidence set is used, where 25/28 is now closest to 90%.

Clearly, the results are sensitive to the removal of the top three countries with the most weight in the synthetic control – particularly for quantity, which relies heavily on Canada's time series. This perhaps suggests that the initial choice of donor pool is important, and that increasing the size of the donor pool further could be beneficial (while limiting it to countries that are similar to the US in some way).

Table 9. Results when removing countries from the donor pool

Dependent variable	Estimated quarterly tariff effect	p-value	26/29 (no removal) or 25/28 confidence sets(a)
			Lower	Upper
	%		%	%
Value (£/cap)	-18.3**	0.069	-100.0(b)	-4.7
Removed from donor pool (and original weight):
1 Canada (25%)	-20.0**	0.071	-66.4	-7.9
2 France (23%)	-20.0**	0.071	-100.0(b)	-3.3
3 Australia (16%)	-13.4*	0.107	-100.0(b)	-1.2
Quantity (LPA/cap)	-10.3**	0.069	-19.6	-9.5
Removed from donor pool (and original weight):
1 Canada (56%)	-13.6*	0.107	-35.5	-2.4
2 S. Korea (15%)	-16.2*	0.107	-41.2	-0.7
3 Australia (10%)	-8.2**	0.071	(c)	(c)
Price (£/LPA)	-10.2	0.345	-64.7	+17.9
Removed from donor pool (and original weight):
1 Switzerland (38%)	-12.9	0.321	-83.9	+22.9
2 Australia (14%)	-10.6	0.321	-71.7	+18.3
3 Lithuania (12%)	-11.1	0.321	-74.2	+18.8

Showing results for a quarterly first-lag specification. Key:(a) the 25/28 confidence set was the closest to 90% given a donor pool of 27 countries (plus the US); (b) truncated to -100%; (c) no 90% confidence set was obtained due to low power (however, for Australia's quantity, a 26/28 confidence set was given by [-13.7%, -11.7%]); *** p < 2/29 or 2/28; ** p < 3/29 or 3/28; * p < 4/29 or 4/28.

6.3.2 Covid-19

Although the synthetic control method tries to account for post-tariff fluctuations experienced in the US and the control states, it is only an assumption that these variations are identical. Covid-19, a major contributor to the dampened demand for whisky and many other food and drink exports during 2020, may not have impacted our synthetic control and the US in a similar way.

• Comparing new confirmed Covid-19 cases in the US to cases in our weighted average^[47] suggests cases were particularly high in the US during November and December 2020. If this did have a major impact on demand for whisky not reflected in our synthetic control estimate, the tariff impact would be overestimated.
• The Oxford COVID-19 Government Response Tracker^[48], however, suggests this rise in cases did not result in a major deviation in government response between the US and our weighted average. Nonetheless, importers and consumers of Scotch whisky may have been more cautious regardless of state-level restrictions.

The comparison between the US and our synthetic control for both of these metrics is shown in Figure 11. The country weights used were those of the quarterly first-lag synthetic control described in Section 6.1.1.

6.3.3 Scotch Whisky Sales

The same IWSR data as shown in Table 2 was used to construct a change in (Scotch) whisky sales for the synthetic control (using the same quantity first-lag specification's country weights as above for volume, and the value first-lag specification's country weights for value).

These results suggest that the synthetic control saw an increase in Scotch whisky sales between 2019 and 2020, both in terms of volume (+1.6%) and value (+5.4%). This is in contrast the to the US' decline in Scotch whisky sales, shown in Table 10 below. These results are somewhat in line with expectations, suggesting that Scotch whisky sales in our combination of countries fared better than in the US. Similarly, sales of all whisky saw increases in both the US and our weighted average, although the increase seen in the latter was more muted.

Table 10. Scotch whisky sales in the United States and the synthetic control, 2019-20

		United States		Synthetic Control
		Volume(a)	Value(b)	Volume(a)	Value(b)
		millions of 9-litre cases	$ billions	millions of 9-litre cases	$ billions
Scotch whisky	2019	8.51	3.47	9.42	6.89
	2020	8.38	3.36	9.57	7.26
	Growth (%)	-1.5	-3.1	+1.6	+5.4
All whisky	2019	74.77	19.90	36.86	12.45
	2020	78.40	21.16	37.09	13.01
	Growth (%)	+4.9	+6.4	+0.6	+4.5

Source: IWSR Drinks Market Analysis Ltd. (via the Scotch Whisky Association), RESAS calculations. (a) Using the first-lag quarterly quantity specification's country weights; (b) Using the first-lag quarterly value specification's country weights.

Note however that this is sales of all Scotch whisky, not just single malt. Other varieties, e.g. blended Scotch whiskies, will make up a large proportion of sales. Additionally, data on sales value from the IWSR includes sales tax, duty, and other mark-ups, while the export value obtained from HMRC excludes these. The IWSR data therefore also includes the tariff duty paid from October 2019 onwards.^[49]

The data includes both on- and off-trade sales, with the US showing a high proportion of off-trade Scotch whisky sales (82.2%) compared to the global average (78.1%) in 2019. Splitting the Scotch whisky sales volume data into on- and off-trade parts, we note that the increase in off-trade sales between 2019 and 2020 was more pronounced in the synthetic control (+20.9%) than in the US (+6.3%), while the decrease in on-trade sales was less pronounced (-49.0% in the synthetic control versus -55.5% in the US).

6.4 Aggregated Results

The estimated per-capita tariff impacts on value and quantity shown in Section 6.1 can be aggregated using US population data.

Using the original proportional impacts and 90% confidence sets, we can obtain approximate tariff impacts over the entire Q4 2019 – Q4 2020 post-tariff period. As shown in Table 11, the quarterly tariff impact of 18.3% in value and 10.3% in quantity translates to a quarterly reduction of £18.5m or 213.6 thousand LPA per quarter. When totalled over the five observed post-tariff quarters, this amounted to £92.4 million or 1,068 thousand LPA. The wide confidence set for value means the total impact could be as small as £23.7m, while quantity's more narrow confidence set suggests a minimum impact of 985,700 litres of pure alcohol.

Table 11. Estimated total tariff impact on single malt export value and quantity

Unit	Estimated tariff impact	Average quarterly impact		Total impact (Avg. x 5)
	%	£/cap	£m	£m
Value	-18.3**	-0.06	-18.5	-92.4
90% conf. set	[-100.0, -4.7]	[-0.31, -0.01]	[-101.3, -4.7]	[-506.3, -23.7]
	%	mLPA/cap	000's LPA	000's LPA
Quantity	-10.3**	-0.64	-213.6	-1,068.0
90% conf. set	[-19.6, -9.5]	[-1.22, -0.59]	[-405.8, -197.1]	[-2,028.8, -985.7]

Showing results for the quarterly first-lag specifications without population interpolation. Key: *** p < 2/29; ** p < 3/29; * p < 4/29.

When looking at the quantity exported, we could transform 'litres of pure alcohol' to quantities of single malt by assuming an alcohol content of 40% (by volume). This would lead to an estimated tariff impact of 2.7 million litres of single malt, with the 90% confidence interval given by [-5.1 million litres, -2.5 million litres].