Structural Changes the Danish Alcohol Market
Transcription
Structural Changes the Danish Alcohol Market
ANDERS MILH0J • Structural Changes In the Danish Alcohol Market This poper examines annual (1978-1995) and quarterly (1990- 1995) data on the Danish consumption of beer, wine and spirits, in order to identify changes in their respective market shares as aresult of the major tax cuts introduced for beer and wine in 1991 and 1992. It is immediately clear that the sales of spirits have decreased while the sales of wine have increased. The changes are too large to be explained exclusively on the basis of the reduced levelof border trade with Germany, the main motive for the tax reductions. The statistical analyses give a more detailed picture of the reasons for the decline in the market share of spirits in total Danish alcohol sales from 14 % in 1990 to 11 % in 1993. It is included that the main part of this fall, at least 2 %, is due to the tax-induced changes in price relations, while more than l % comes from a trend movement. The market share of beer has been relatively constant for a number of years. However, this is the net effect of a trend movement tending to reduce its market share and, on the other hand, the reduced price of beer relative to the price of spirits caused by the tax reductions, which has increased its market share. The market share of wine has been steadily increasing since the 1970s. There is no evidence of any dependencies on the prices of alcohol. This conclusion cannot, however, be sufficiently documented on the basis of the Danish data series, as it is impossible to distinguish between effects due to changing wine prices and effects due to changing beer prices, because of the parallel tax cuts for beer and wine. Key words: alcohol market, tax reduction, border trade, Denmark The Danish alcohol market in the 19905 Consurner taxes on beer and wine have been reduced twice in Denmark during the 1990s, i.e. on 1 July 1991 and on 1 October 1992. The reductions were quite drastic: approximately 10 % in 1991 and as much as 25 % on some low-price brands of beer in 1992. The tax reductions were motivated by the high levelof border trade with Germany. It was common knowledge that, before the tax cut s, no one in the southern parts of Jutland bought any wine and only very smal1 quantities of beer, while people living further away from the German border went on regular trips to Germany mostly to buy beer and wine. The Danish government tried to control this traffic by imposing various restrietions on how much people were allowed to bring into the country, but being incompatible with EU intentions, they had to be relaxed. The high levelof border trade was not only unacceptable for fiscal reasons, but there was the additional problem that while in Germany, Danish people not only bought alcohol but also traded in other commodities. As beer and wine were considered the main reason for border trading, the retailers in southern Jutland in particular called for a solution. Danish taxes on beer and wine and also on petrol were consequently reduced, helping to achive a significant reduction in the total amount ofborder trade. The tax reductions apply of course to the whole Nordisk Alkoholtidskrift Vol. 13, 1996: English Supplement -33- country, inc1uding the island of Zealand with the Copenhagen are a, where any significant border trade was impossible for geographical reasons. Two previous papers (Milhøj 1993 and 1995) studied the effect of the tax reductions on the total consumption of alcohol in the whole country as well as in various areas of the country. The overall conc1usion was that total alcohol consumption did not increase as a consequence of the reduced taxation, even though official statistics indicated an increase in alcohol sales because of the reduced level of border trade. The dramatic price cuts for beer and wine have, however, increased their sales at the expense of spirits, the price of which has remained constant. This price driven substitution between beer and wine, on the one hand, and spirits, on the other, is the subject of the present paper. and for this reason the calculation of sales volumes in 100 % alcohol is carried out in two steps. In the fust step the sales volumes measured in litres in drinking strength of the various brands are derived from the tax revenue. For the two years 1991 and 1992, such data on the volume in drinking strength as well as official data on sales in pure alcohol are available. Based on these figures, the average strength of beer for both years can then be calculated as 4.8 %; accordingly, the average strength of wine is 12.3 %. These strengths are then used in the second step to transform sales volumes measured in drinking strength to sales volumes measured in 100 %alcohol. This procedure is used to derive quarterly data for the sales of the three types of alcohol meas ure d in 1 000 litres of pure alcohol. As part of the monthly Danish Consurner Price Index, the StatisticaI Bureau has given the prices for the three types of alcohol since 1978: Data The Danish StatisticaI Bureau has provided annual data on the sales of alcohol measured in 100 % alcohol since 1978. In this paper the variables are denoted as: BSALE Volume of beer sales in 1 000 litres 100 % alcohol WSALE Volume of wine sales in 1000 litres 100 % alcohol SSALE Volume of spirits sales in 1 000 litres 100 % alcohol. The total sales of alcohol measured in 100 % alcohol is then given as: TOTSALE =BSALE + WSALE + SSALE. These data series are based on data supplied by the Ministry of Taxation on tax revenue from the sale of alcohol. Data on tax revenue are, however, also available for a shorter time interval than one year. For wine, figures on tax revenue are available on a quarterly basis since the beginning of 1989; and for beer, monthly data are available since July 1990 and for spirits since May 1989. The Ministry of Taxation figures are based on sales of spirits in 100 % alcohol. For wine and beer, taxation is not directly based on alcohol content, BPRICE Index numbers for price of beer WPRICE Index numbers for the price of wine SPRICE Index numbers for the price of spirits. The consurner price indices, which are based on 1980 = 100, give the prices as they are met by the consurners, taxes inc1uded. Yearly figures for the prices are simply derived as averages of 12 monthlyvalues. Two parallel analyses are presented in this paper on the basis of these data. First, an analysis of yearly data is presented on the basis of the data for the 18 years from 1978 to 1995; then, an analysis is presented on the basis of 22 quarterly observations from the third quarter of 1990 to the fourth quarter of 1995. The first analysis of the yearly data aims to identify long-term effects in the alcohol market, while the purpose of the second analysis is to identify short-term effects on the alcohol market from the tax cuts on beer and wine in 1991 and 1992. There are, however, some critical remarks that need to be made on the quality of these data. The sales data for one specific month are derived on the basis of the monthly tax revenue, not on the actual sales during the month. However, the alcohol tax is not paid by the retailers but by the wholesalers, as the goods are delivered to the retailers. If the shops hold some stock of wine, for instance, the wine taxed in one period could actually be sold Nordisk Alkoholtidskrift Vol. 13, 1996: English Supplement -34- Figure 2. Market shares in 1978-1995 Figure 1. Sales of alcohol in 1 OOOlitres 100 % alcohol for 1978- 1995 Market shares 1000 1100% 60000 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 50000 40000 30000 B BB B B BBB BBBBBB BBB B 20000 10000 W"W"WYf WWWWWWWWWW wwWW S S S S S S S S S S S S S S 0T-~~~'-~~~~~~~~~~~ 1975 1980 1985 1990 1995 TT Total BB Beer WW Wine LBTOT =10g(BSALE/TOTSALE) LWTOT = 10g(WSALE/TOTSALE) LSTOT =10g(SSALE/TOTSALE). These variables are the logarithmically transformed market shares of the three types of alcohol in the total Danish alcohol market. In the models these variables are explained in a linear regression B BB BBBBBBBB BB BB WWWWWWwWWWW "WWYf S WWWW S S S S S S S S S S S S S S 0.0 'r--->--'-~"-'~~""-"'r-r~""-"'--'--,-""-",,,,,,,,,,,,,-,--,, 1975 1980 1985 1990 1995 SS Spirits during alater period. This problem is especially relevant in the periods around ChristInas and Easter, for which sales data could be misleading beeause alcohol taxed in March, for instance, eould in faet be sold in April. Another problem is that the eonsumer price index is based on "normal" priees from a sample of Danish retail shops. These normal prices do not take temporary priee reductions etc. into account. Both the sales and price series are aggregated, so that beers of various brands, strengths and prices are considered as one. However, it is clearly possible that the price driven substitution in consumption among the three types of alcohol is more pronounced for low-price brands of beer, wine and spirits, than for more expensive brands. The main subject of the present analysis is to study price driven substitution. For this reason the market shares of the three types of alcohol are the variables of interest. These variables are denoted as: BB B BB Beer WW WIDe SS Spirits model by the lag-transformed relative priees LOGPBW = 10g(BPRICE/WPRICE) LOGPBS = 10g(BPRICE/SPRICE) LOGPWS = 10g(WPRICE/SPRICE). A graphical analysis In this section diagrams for all variables used will be presented using both yearly and quarterly figures. As the quarterly sales data contain some seasonal variation, the series are seasonally adjusted by a simple applicatian of seasonal dummies, which are also used in the statistical analyses. Figure 1 shows the total sales of alcohol in Denmark from 1978 to 1995. As we can see, sales increase d from 1978 to 1983, declined somewhat in the late 1980s and has recently started to increase again. However, the movements in the total sales of alcohol are af only minor importance and could to someextent be explained by changes in the volume of border trade. Figure 1 also shows the sales of the three types of alcohol for the same period. The sale of wine has steadily increased, while spirits sales have declined each year since 1986 and the sale of beer has been more or less constant throughout the whole period. In spring 1985 there was a long strike at the breweries in Copenhagen. The effects of this strike Nordisk Alkoholtidskrift Vol. 13, 1996: English Supplement -35- Figure 3. Price of alcoholic beverages as indices with Figure 4. Price relation beer to wine for 1978-1995 with 1980 = 100 for 1978-1995 1980= l Price indices Price relation 200 180 160 140 120 100 80 60 40 20 1.10 1.05 1.00 B~'W3VJ3W BW BW 0.95 0.90 0.85 0.80 0.75 0T-~~~'-~~~-'~~~~~~~~ 1975 1980 BB Beer 1985 1990 WW Wine 1995 0.70 T-~~~-r--~~~-r--~~~,....,~~-.--.r 1975 1980 1985 1990 1995 SS Spirits BW Beer/Wine Figure 5. Price relation beer to spirits for 1978-1995 with Figure 6. Price relation wine to spirits for 1978- 1995 with 1980= l 1980= l Price relation Price relation 1.10 1.10 1.05 1.00 0.95 BS BS BSBS BS BS BS BS BS WS 1.05 1.00 BS BSBSBS 0.95 0.90 BS 0.90 BS 0.85 0.80 WS 0.85 0.80 BSBSBS 0.75 w~s ws jN9NgmYv~f't.ns ws wjY1'i' WS 0.75 0.70 T-~~~--r-~~~--'-~~~--,-~~~r-r 1975 1980 1985 1990 1995 0.70 T-~~~--r-~~~--,-~~~--,-~-'-~""" 1975 1980 1985 1990 1995 BS Beer/Spirits WS Wine/Spirits \ on beer sales are easily seen in the diagrams: in 1985 sales figures for beer were lower than expected, while the sale of wine in particular was increased. Figure 2 shows yearly series for the market shares of the three types of alcohol, that is the variables LBTOT, LWTOT and LSTOT before the logtransformation. Clearly, the market share of wine has steadily increased, while the figures for beer and spirits have decreased, for spirits especially in 1992 and 1993. Moreover, the effect of the 1985 bre- weries strike is obvious. Figure 3 presents yearly price series for the three types of alcohol. Of course, inflation produces a steady upward trend in the prices, but the effects of the frequent minor adjustments in alcohol taxation in 1982-1986 are also evident. The tax cuts for beer and wine are clearly seen, but apart from that the prices for all types of alcohol and especially beer have changed very little in recent years, with the rate of inflation in Denmark being very low in the 1990s. A similar picture emerges from Figures Nordisk Alkoholtidskrift Vol. 13, 1996: English Supplement -36- Figure 7. Sales of alcoholic beverages in 1 OOOlitres Figure 8. Price of alcoholic beverages as indices with 100 % alcohol for the quarters 1990.3-1995.4 1980 = 100 for the quarters 1990.3-1995.4 1000 l 100% 14000 Price indices T T T T 12000 8000 B B B 200 T B 180 160 BB BB BB B 6000 T T T 10000 T B 2000 W~W BBBB BBBBB 140 W 4000 WWW SSSSSSSSSSSSS WWWWWWWWWWWWW BBBBBBBBBBBBB 120 S O 100 90Ql 91Ql 92Ql TT Total BB Beer 93Ql 94Ql WW Wine 95Ql 96Ql 90Ql SS Spirits 4-6, which present the relative prices, that is LOGPBW, LOGPBS and LOGPWS without the log-transformation. As the series starts from the base value 1980 = 1 we can see that the price of beer has decreased relative to both wine and spirits. However, the price relation between beer and wine has been almost constant, whereas the price of spirits has increased considerably compared with the prices of both beer and wine due to the tax reductions in the 1990s. This very parallel development of beer and wine prices gives rise to some technical problems in the analyses because of multicollinearity. Figure 7 gives seasonally adjusted as well as actual observed quarterly data for sales of the three types of alcohol and the total sales of alcohol. Figure 8 similarly presents the actual quarterly price series, which show no seasonal variation. The reductions in the prices of beer and wine relative to spirits caused by the tax cuts in 1991 and 1992 are clearly seen. The tax cut in 1991 has reduced the price of beer more so than the price of wine, while the reduction in 1992 did not change the beer / wine price ratio. 91Ql 92Ql BB Beer 93Ql WW Wine 94Ql 95Ql 96Ql SS Spirits The statisticai model The basic statisticai model presented here will be used for both yearly and quarterly data. The model is constructed in such a way that it is particularly well suited for studying the effects of relative price changes of wine, beer and spirits on their market shares. To this end linear regression models are used with the market shares LBTOT etc. as response variables and the relative prices LOGPBS etc. as explanatory variables. As demonstrated by the graphical analyses, the market share of wine has been steadily increasing over the past years, while the market share of spirits has been decreasing. It is possibie then that market shares change for other reasons than relative price changes. It seems reasonable to suggest, for instance, that Danish people have acquired a stronger taste for wine partly because of changes in eating habits and partly because of their more frequent travel in southern Europe. In the statisticai models such a steady development can easily be described by a trend variable, that is an explanatory variable, TREND, which simply equals the observation number. With both relative prices and the trend as expla- Nordisk Alkoholtidskrift Vol. 13, 1996: English Supplement -37- natory variables, the basic model for the market share of spirits, for instance, is written as: LSTOT = ~INT + ~BsLOGPBS + ~wsLOGPWS + Models for the market share af beer The estimated basic model for the market share of beer is: ~TRENDTREND. In this model the market share of spirits rather than the actual sales volume of spirits is used as the response variable. Therefore changes in the total sales of alcohol have no effect on the response variable as long as the sales of the three types of alcohol change proportionally. Many factors that could possibly influence the total sales of alcohol such as income, other taxes, prices for other goods e.g. housing, will under this condition have no influence on the market shares. Under these circumstances changes in the population size could also be excluded from the model. However, if the effects of external factors on the three types of alcohol are not proportional, the effects are probably included in the trend. For instance, changes in drinking habits caused by older generations preferring beer or akvavit, and younger people preferring wine, could be included in the trend. FiHed models for yearly data This section describes the results of estimations of parameters from the various versions of the basic model, as fitted models using the estimated parameters. Some technical comments are also made. Finally, the concluding section is devoted to a more intuitive discussion of the results. In all the models presented, the actual numerical value of the intercept is of no direct interest and is hence omitted. Moreover, as some of the estimated parameters are insignificant, significant values using a two-sided test at a 5 % level are identified by giving the numbers in bold face. The fitted relations are checked by the usual control procedures for regression models. As is usual for time series data, all the models present a good fit with high values of R2. For time series data it is especially important to note that the models contain no autocorrelation in the residuals, as the hypotheses of no autocorrelation are accepted by Durbin-Watson tests. LBTOT = !NT - 0.04 LOGPBW - 0.05 LOGPBS 0.009 TREND In this model only the parameter representing the trend is significant. To further study the effect of relative price changes, models with only one relative price were fitted, but the price ratios were still found to be insignificant, when tested at a 5 % level. If both price ratios are excluded from the model, only the trend remains LBTOT = INT - 0.008 TREND, a relation that, measured by R2, is nearly as good as the model also using the price ratios as explanatory variables. In summary, the development in the market share of beer is dominated by the downward sloping trend, and the price ratios are of no importance. The data clearly indicate that the relative beer I wine price ratio has no significance with regard to the market share of beer. This is because the beer I wine price ratio in Denmark has been more or less constant during the period concerned, and therefore it is impossible to study the effect of changes in this price ratio on the market share of beer, The effect of the beer I spirits price ratio is hardly significant, but an effect of the beer I spirits price ratio is consistent with the effect of the beer I spirits price ratio for the market share of spirits, as seen below. Models for sales af wine When the basic model is fitted to the market share of wine, only the trend variable is significant, while neither the relative beer I wine price nor the relative wine I spirits price give a significant contribution to the model: LWTOT = !NT - 0.08 LOGPBW - 0.14 LOGPWS + 0.029 TREND. This is still true when only one of the price ratios is Nordisk Alkoholtidskrift Vol. 13, 1996: English Supplement -38- used as an explanatory variable in the model. Hence it is conduded that the market share of wine only depends on the trend: LWTOT =!NT + 0.031 TREND. This upward trend is highly significant, giving a sharply increasing market share for wine in Denmark, probably due to changing habits, e.g. dietary habits, and is not caused by changing price relations. feds of changes in the winel spirits price ratio. However, when comparing the R2-values, it seems that the price ratio of beer to spirits is of most importance. To sum up, the coefficients to LOGPBS and LOGPWS add up to one in all relations involving the relative prices. In all models the negative trend is significant, which means that the effeds of changes in the price relations could be distinguished from the slowly decreasing market share of spirits due to changing drinking habits caused by other reasons. Models for the sales af spirits Analysis of quarterly data For the market share of spirits the trend is of some importance, but the price ratios also enter the model. Using both the beer I spirits and the winel spirits price ratios and the trend variable, we find: LSTOT = INT + 0.97 LOGPBS + 0.10 LOGPWS0.010 TREND. Although it seems on the basis of this relation that only the beer I spirits price ratio and not the winel spirits price ratio is of importance, this condusion could be astatistical artifad due to multicollinearity. When only one of the relative prices is used as an explanatory variable, we find: LSTOT = !NT + 1.05 LOGPWS - 0.017 TREND In the application of the basic model to quarterly data, seasonal dummies are induded in order to model the seasonal variation. This provides a description of the systematic seasonal variation in drinking habits, but as this is of no interest in the present study, the estimated values for the coefficients to the seasonal dummies are not reported here. All the models pass the usual graphical and numerical tests. It is particularly noteworthy that no autocorrelation is found. In the same way as in the analyses of yearly data series, trends are induded in all relations, but as quarterly observations are only available for a period of three and a half years, it is impossible to give a precise estimate of the trend. and Models for the market share af beer LSTOT = INT + 1.04 LOGPBS - 0.009 TREND. As both coefficients to the relative prices are now individually significant, we see that these relations are significantly betler than the model which only indudes the trend: LSTOT = !NT - 0.024 TREND. The condusion is that the price ratios of beer to spirits and wine to spirits enter the relation with a coefficient dose to one, if only a single price ratio is applied. However, as the two price ratios show a parallel development during the period under study, it is impossible to distinguish the effeds of changes in the beer I spirits price ratio from the ef- The fitted basic model for the market share of beer is: LBTOT = INT + 0.04 LOGPBW - 0.14 LOGPBS 0.005 TREND + seasonal dummies. In this relation the price ratios are of no significance, but if the trend is exduded price ratios become significant. However, all models with the trend variable exduded give R2-values smaller than the model using only the trend: LBTOT = INT - 0.003 TREND + seasonal dummies, and hence this model is preferred. Nordisk Alkoholtidskrift Vol. 13, 1996: English Supplement -39- Models for the market share af wine The fitted version of the basic model is: LWTOT = INT - 0.94 LOGPBW - 0.18 LOGPWS +0.009 TREND + seasonal dummies. Using various combinations of the explanatory variables, we find that models including the trend are superior to all other models, and in summary the final model using only the trend seems most reasonable: LWTOT mies. = INT When these relations are compared with regard to R2, it is obvious that the trend is unimportant, but equally obvious that at least one of the price ratios should be included in the model. However, it is not clear which price ratio, beer to spirits or wine to spirits, should be used in the model due to the multicollinearity, as the price ratios shift in parallel. As for the yearly data series, it is clear that the coefficients to LOGPBS and LOGPWS add up to one in all relations. Moreover, it seems that the price ratio of beer to spirits is to some extent more important than the price ratio of wine to spirits. +0.012 TREND + seasonal dumConclusion It seems that the market share of wine has increased because of changing drinking habits and not because of changing price relations. Models for the market share af spirits The relations for the market share of spirits present some multicollinearity problems, in this case because of the parallel behaviour of the beer / spirits and the wine/ spirits price ratio. The fitted basic model is: LSTOT = INT +2.54 LOGPBS - 1.54 LOGPWS + 0.000 TREND + seasonal dummies, in which only the coefficient to the beer / spirits price ratio is significant. The negative sign of the coefficient to the wine/ spirits price ratio is unexpected, but as the coefficient of the beer / spirits price ratio is correspondingly high, this is probably due to multicollinearity. If the trend is excluded, the model becomes: LSTOT = INT +2.65 LOGPBS - 1.62 LOGPWS + seasonal dummies. Even this relation suffers from multicollinearity, which is easily seen when models with only one price ratio as an explanatory variable are fitted LSTOT = INT + 1.14 LOGPBS + seasonal dummies, LSTOT = INT + 1.20 LOGPWS + seasonal dummies. Interpretation af the models for quarterly and yearly data In all relations based on yearly data, the trend makes a significant contribution to the model fit. For the sales of beer, the trend parameter around the value ~TREND = - 0.008 corresponds to a redudion in the market share of almost 1 % per year. As the ratio BSALE/TOTSALE is approximately 60 %, this implies a redudion of the ratio to 59.4 % within the following year. For the sales of wine, the trend parameter ~TREND = 0.030 corresponds to ayearly increase of 3 % in the market share of wine, which in view of the actual ratio of around 30 % gives an increase to 30.9 %within the next year. For spirits a downward trend parameter ~TREND is estimated within the range of between - 0.020 and -0.010, depending on whether the price ratios are included in the models or not. As the market share of spirits is approximately 10 %, this makes a reduction to between 9.8 % and 9.9 % within a year. In the relations base d on quarterly data, the trend estimates involve much greater uncertainty, partly because of multicollinearity and partly because data is only available since 1990. The estimated parameter values of around ~TREND = - 0.004 for the quarterly market share of beer corresponds to a parameter value for yearly data approximately four times higher, that is ~TREND = - 0.016. This is an even greater reduction in the beer market share than is found using yearly data for the total period Nordisk Alkoholtidskrift Vol. 13, 1996: English Supplement -40- T able l. Comparison of relative alcohol prices in 1991 and 1993 Year Beer/wine LOGPBW price ratio 1991 1993 0.94 0.93 Beerl LOG PBS spirits price ratio - 0.061 - 0.070 Winel LOGPWS spirits price ratio -0.049 -0.251 0.95 0.78 1.01 0.84 0.012 - 0.180 Table 2. Comparison of market shares for the three types of alcohol in 1990 and 1993 Year market share of beer 1990 1993 0.59 0.57 LBTOT - 0.52 - 0.55 market share of wine LWTOT market share of spirits 0.27 0.31 - 1.31 -1.16 0.14 0.11 1978 - 1993. Similarly, the trend parameter f3 TREND = 0.012 for the market share of wine gives a more pronounced increasing market share than is established in the analysis of the whole period. One possible interpretation of this is that the trend has accelerated in recent years. For spirits the trend did not appear significant. The fitted models for both quarterly and yearly data point at a significant contribution of variations in the beer / spirits or wine / spirits price ratios on the market share of spirits with coefficients f3BS and f3 ws' which sum to one. This implies, for instance, that a 10 % reduction in the relative beer / spirits and wine / spirits prices causes a decrease of 10 % in the market share of spirits, which, if the market share is initially at 10 %, corresponds to a reduction to 9 %. However, because of multicollinearity, the effeds of the two price ratios cannot be separated. It is of course self-contradictory that a price dri-: ven substitution is found for the market share of spirits but not for the market shares of beer or wine. This, however, is easily explained by weaknesses of the data, as some of the substitution in disfavour of spirits is in favour of the sales of beer and some in favour of the sales of wine, so that the largest substitution is found for spirits. The minor LSTOT -2.00 -2.20 effects of changing prices upon the market shares of beer and wine then become insignificant, and they are to some extent considered as part of the trend in the fitted models. Effects af tax reductians In June 1991, just before the tax cut on beer and wine on 1 July 1991, the consumer price index was 160 for beer, 170 for wine and 168 for spirits. Two years later, in June 1993, the figures were 137, 147 and 176, respectively, indicating that the price of beer had declined by 14.4 %, the price of wine by 13.5 %, and that the price of spirits had increased by 4.8 %. Table 1 gives the relative priees as well as the actual values of the logarithmically transformed priee ratios. It appears that the beer / wine price ratio has been almost unaffected by the tax reductions, while spirits compared with both beer and wine have become much more expensive. Total sales of alcohol, measured in 1 000 litres of 100 % alcohol, have inereased by only 2.5 % from 49 643 in 1990 to 50 879 in 1993, but the market shares of all three types of alcohol have ehanged considerably, as we ean see from Table 2 above. The sales volume of beer measured in 100 % aleo- Nordisk Alkoholtidskrift Vol. 13, 1996: English Supplement -41- hol is almost unchanged at 29 479 in 1990 and 29236 in 1993, but the market shares have changed so that the sales of wine have increased from 13 435 to 16 018 and the sales of spirits dropped from 6 729 to 5 620. The sales data are presented in Table 2 with 1990 as a reference year, as this was the last calendar year before the first tax reduction, and moreover relative prices were steady from 1990 to June 1991. In the relations for the market share of spirits, both the price ratios beer to spirits and wine to spirits are of potential importance. If both parameters are estimated simultaneously, the estimates add up to one, but if one of the parameters f3 BS and f3 ws is set to zero, the remaining parameter is estimated as dose to one. The variables LOGPBS and LOGPWS have declined by 0.20 and 0.17, respeetively, and hence the model implies a 17-20 % reduction of LSTOT, the logarithmically transformed market share of spirits. As the market share of spirits in 1990 was 14 %, the reduction in LSTOT corresponds to a reduction in the market share to below 12 % caused by the tax reductions for wine and beer. It is remarkable that this reduetion in the market share of spirits due to changes in the relative prices is found besides the downward sloping trend, which considered over a three year period gives a further reduction of approximately 0.7 %in the market share of spirits, which added to the reduction caused by the changing price relations has led to a further reduetion in the market share to 11 %. During this three-year period the market share of beer has been reduced, while the market share of wine has increased because of the trend movement, However, when the fitted parameters for the trends are considered, we find that the market shares of beer and wine are greater than explained by the trend, because of the substitution caused by the changed price relations. The tax cuts for beer and wine, together with the cuts for petrol, were intended to reduce the level of border trade with Germany. The data series here show that border trade has in faet been reduced. Sales of beer in southern Jutland have increased to some extent, while sales of wine have increased more sharply in southern Jutland than in other parts of Denmark. As noted ab ove, total sales of alcohol in Denmark have increased by 2.5 % from 1990 to 1993, mainly because of increasing wine sales and probably folIowing the reduced levelof the border trade. This means that even constant sales of spirits would cut back the market share of spirits, but the reduetion in the market share for this reason is small, only from 14 %to 14/1.025 = 13.66 %. As the decrease in the market share of spirits to 11 % is more dramatic, it cannot be solely explained by increasing sales of other types of alcohol, caused by reductions in border trade. Moreover, the reduced levelof legal border trade with beer and wine mayaIso have reduced the illegal border trade with spirits. Sales figures (albeit undocumented) indicate that sales of spirits have increased in southern Jutland, implying that the (partly illegal) private import of spirits has been reduced. REFERENCES Milhøj, A.: Virkningen af afgiftsnedsættelser på øl- og vins alget i Danmark. Nordisk Alkoholtidskrift 10 (1993): 6, 319 - 329 Milhøj, A.: Udviklingen i danske familiers øl- og vinindkøb 1990 -1993. Nordisk Alkoholtidskrift 12 (1995): 4, 181-193. The article is a revised version of an article previously published in Danish in Nordisk Alkoholtidskrift 2/1996. Nordisk Alkoholtidskrift Vol. 13, 1996: English Supplement -42-