Price elasticity of the demand for soft drinks, other sugar-sweetened beverages and energy dense food in Chile

Overweight and obesity have increased in less developed countries at a noticeable pace [1]. It is estimated that overweight and obesity caused 3.4 million deaths and 93.6 million DALYs in 2010 [2]. Latin American countries have been particularly affected by the nutritional transition, overweight and obesity have become a challenge. In particular, Chile ranked seventh in adult obesity among the organization for economic cooperation and development members [3]. According to the Chilean National Health Survey 2009–2010, overweight and obesity reached 64.4% in men and 64.3% in women [4]. In the same period, the prevalence of diabetes mellitus was estimated at 8.4% in men and 10.4% in women [4]. Among children aged 6–18 years, overweight and obesity prevalence is estimated at 45.5% [5].

Although there are multiple risk factors associated to overweight and obesity, the literature shows that sugar sweetened beverages (SSB) are a significant risk factor for chronic diseases [6‐9]. There is also evidence that intake of high energy dense and poorly nutritious processed and ultra-processed food is associated to obesity and metabolic syndrome [10, 11].

Chile is the second world’s largest consumer of caloric beverages, behind Mexico, with an estimated 180 l per capita consumption in 2014 [12]. In one decade, soda sales doubled and only 19.1% corresponded to calories-free soda [13]. Between 1987 and 2007, SSB consumption raised from 116 to 289 ml per capita per day and this increase was larger among poor households [14]. In the same period, in households from the metropolitan area of Santiago, household expenditures on soda, juices with added sugar and sparkling water -as a percent of total spending on food and beverages- increased from 4 to 8.4% [15].

Since the eighties, Chilean households have increased the proportion of expenditures in processed food with high energy density and added sugars and have reduced the consumption of traditional food. Percentage spending in processed food increased from 42 to 57%, and in the poorest households it went from 53 to 68% [14]. Consequently, processed and high-processed food represents 55.4% of all energy purchased by Chilean households in 2006–2007 [16]. Expenditures in food prepared outside the household increased from 12.1 to 20.1% between 1987 and 2007. Poor households show the largest increase [15].

Several fiscal policies to reduce the consumption of SSB have been recommended and implemented in different countries [17]. The World Health Organization (WHO) and the Pan-American Health Organization (PAHO) suggest implementing fiscal measures to discourage the consumption of foods and beverages that can harm health. For instance, the Plan of Action for the Prevention of Obesity in Children and Adolescents in the Americas, presented by the PAHO in 2014, advises taxing SSB and high-energy dense products in order to stop the increase in the prevalence of obesity in children and adolescents [18]. France, Hungary, Egypt, Finland, Mexico and the City of Berkeley in the USA have implemented taxes on SSB or high-energy dense foods. In Chile, there was a tax on sales and imports of beverages of 13% [13]. Since October 2014, the tax rate increased to 18% for non-alcoholic beverages, naturally or artificially flavored, that have content greater than 15 g of sugar per 240 ml or with an equivalent portion [19].

Despite the implementation of taxes in several countries, there is limited published scientific evidence on the effectiveness of fiscal measures in reducing the consumption except for Denmark [20] and Mexico [21]. In the absence of such evidence, price-elasticity estimates are a useful tool in achieving the objective of measuring the effect of a fiscal policy on consumption and simultaneously to forecast potential substitution effects. Some factors can modify the expected or estimated response to a tax such as the tax incidence, that is how much the tax is passed on to consumers, or increase in advertising campaigns or promotion of taxed products in stores. However, price elasticities provide useful estimates of the potential response to changes in prices.

There is a large variability in the SSB price elasticity estimates in the world. A systematic review on price-elasticity studies of non-alcoholic beverages and foods indicate that the price-elasticity ranges from −0.27 to −1.0 [22]. In general, SSB show larger absolute values of elasticities, that is, their demand is more sensitive to changes in price. A systematic review of studies performed in United States, France, Brazil and Mexico showed a mean price-elasticity of demand of SSB of −1.2 [23]. A recent study in Mexico estimated that the price elasticity of SSB was −1.16% and between −1.06 and −1.29 for soft drinks [24, 25] In Ecuador, the price-elasticity of SSB ranges between −1.17 and −1.33 depending on the socioeconomic group [26]. In contrast, studies on price-elasticity of high-energy dense foods are scarce. A study in Mexico found that the demand of this type of food is elastic [25]. The wide variability in the estimation of price elasticities is partially explained by proportion spent on SSB, the availability of substitutes for SSB in each country, the data used and the methods to derive elasticities.

The objective of this study is to estimate the price elasticity of demand for soft drinks, other SSB and high-energy dense foods in urban areas in Chile. Our paper adds to the existing literature the estimation of a linear approximate almost ideal demand system (LA/AIDS) and a sensitivity analysis using two additional models: a quadratic almost ideal demand system (QUAIDS) and one-equation two-part model used in the literature to derive price elasticities to test of the robustness of our findings. Results from the study could be used in the re-design and evaluation of current fiscal policies related to food and beverages consumption, and the potential reduction in the prevalence of overweight and obesity in Chile.

Table 1 presents the proportion of households with positive purchases among the included categories in the demand system. Plain water shows the lowest percentage of households with positive expenditure. In contrast, 77.2% of Chilean households report spending in soft drinks. There is also a high prevalence of households with positive expenditures in sweet snacks (Table 1).

Table 2 shows the average unit values of beverages, snacks and desserts reported by the households with positive purchases, in 2011–2012 Chilean pesos. Beverages are cheaper by kilogram than snacks and desserts. Among beverages, soft drinks are more expensive than milk, coffee, tea and infusions, and plain water. Socio-demographic characteristics of the households show that in Gran Santiago 58% are headed by men and 62% in regional capitals. There is no difference in the age of the head of the household, or in household size across zones. However, in Gran Santiago the percentage of heads of the households with graduate level of education seems slightly higher than in the rest of regional capital cities (Table 3).

Table 4 shows the results of the own and cross price elasticities with respect to soft drinks using LA/AIDS. The left half of the table shows the results of the models that include the IMR. The price-elasticity of all the eight categories is elastic. The price-elasticity of soft drinks is −1.37 implying that a 10% increase in price would be followed by a decrease of 13.7% in the amount consumed, which shows an elastic demand. The most extreme case is plain water, with a price-elasticity of −3.20. This implies that the demand of plain water is very sensitive to changes in price. The estimations of cross-price elasticities show that the degree of substitution of soft drinks with plain water is higher compared to other beverages and high-energy dense foods. A price increase in soft drinks is also associated with a higher quantity consumed of milk, coffee, tea and infusions, other sweetened beverages, sugar and desserts. On the right side of the table, the results of the model estimated without including the IMR. The estimates are robust, with no great differences with respect to the models that include the IMR.

Figure 1 shows the price-elasticity of soft drinks by income quintile. As we can see, the population is fragmented into two major groups. Although all quintiles present an elastic demand to soft drinks, the first and second quintiles show a greater price-elasticity, whilst the third, fourth and fifth income quintiles are less sensitive to changes in prices.

Table 5 displays the results of the QUAIDS model. Unfortunately, it is not possible to get standard errors or p-values of price-elasticities using the QUAIDS Stata command. However, the results using this model are quite similar than those of the LA/AIDS model. The estimated price-elasticity for soft drinks is also −1.37 and the cross-price elasticities show the same sign and magnitude when compared to results from LA/AIDS model.

Lastly, we derived from the two-part model the price-elasticity of participation (−0.19) and the price-elasticity of intensity calculated on households with a positive consumption of soft drinks (−1.05). Hence, the total price-elasticity using this two-part model is of −1.24, slightly lower that the estimated using the LA/AIDS and QUAIDS.

We estimated a demand system for beverages and high-energy dense foods using a cross-sectional survey in Chile. We found that the demand for SSB in Chile is elastic (−1.37 for soft drinks and −1.67 for other SSB). Likewise, we provided evidence that an increase in soft drinks’ prices could lead to increases in the demand of other goods, such as plain water, milk, coffee, teas or other SSB. The sensitivity analysis shows very similar results, price-elasticity of soft drinks ranges from −1.24 to −1.37.

To our knowledge, this is the first paper that estimates price elasticities of the demand for soft drinks in Chile. Our findings are similar to other studies. Cabrera found an average price-elasticity of SSB of −1.29 in several countries [23]. Our estimate for soft drinks is also in the range of other Latin American countries, such as Mexico or Ecuador (−1.06 and −1.20, respectively) [24, 26]. The variability could be due to different definitions of SSB, different statistical models and also by the particular nature of demand in each country. Regarding income level, the studies by Colchero [24] and by Paraje [26] agree that population in lower income groups are more responsive to changes in price, as we found in the Chilean case.

Cross price-elasticities allow us to classify the rest of categories as substitutes and complements. Since their cross price-elasticity is positive, our study suggests that milk, coffee, tea, mate and other infusions, plain water, other SSB, sugar and honey and desserts show a substitute behavior towards soft drinks. For instance, regarding water, our preferred model shows that a 10% increase in price would be associated to an increase of 6.3% of plain water. However, own price-elasticity of water is high, perhaps because of the low proportion of households that report positive expenditures on this item and the high quality of private water supply systems in Chile. Cross price-elasticities of milk, tea, coffee, other infusions and other SSB are smaller, which means that substitution for these products is less clear than water’s.

We acknowledge several limitations in our study, which come primarily from the data. First, we were not able to split the category of soft drinks into non-diet or diet soft drinks, which would have been valuable in order to isolate the possible effect of a tax on soft drinks with added sugar from drinks without calories. Nevertheless, only around one fifth of soft drink consumption in Chile is on calories-free soda [13]. Since our data is cross-sectional, it is also likely that purchases of food and beverages included in our analysis are under-reported, because it excludes expenditures for consumption outside the household and does not consider purchases of all the households members.

Information on geographic location size, or a cluster variable to a more disaggregated level than zone was not available in the data set. Since the Family Budget Survey did not include information on prices that households face, we rather computed the unit values by dividing the amount of reported expenditure on the analyzed food and beverage category by the quantity purchased in kilograms. As noted by Deaton, this introduces two types of biases. First, there are measurement errors that come from both the numerator and denominator of the unit value calculation [37, 38]. Second, a quality effect is also present: households’ characteristics such as size or income affect unit values, because the products are also chosen by their quality. Deaton proposes a system of two equations for each category, which based on certain conditions, allows estimating price and budget elasticities while controlling for differences in the quality of the goods and measurement errors. In spite of its theoretical advantages, we were not able to estimate this Deaton’s model because the analyzed households must belong to small enough geographic units (such as villages) to support the assumption that the prices they face are the same. Deaton’s method assumes that variations in unit values in the same geographic unit are due to differences in the quality of goods. Therefore, to implement his method, information on small geographic location is crucial. We lacked this, and then tried to create artificial clusters based on income and the two zones available in the data set. Nonetheless, we got implausible price-elasticities estimates that we think were caused by the artificial nature of the clusters that we created and the small number of them. Furthermore, the FBS was collected along almost 1 year. Prices could have varied during this period. Regrettably, from the public microdata, it is not possible to know each household’s day of interview, so monetary variables are inflation adjusted. However, Chile presented a low inflation rate between 2011 and 2012 of 3.3 and 3.0%, respectively. Likewise, seasonality of purchases was not considered which could bias our price-elasticity estimations if the pattern of beverages consumption varies according to weather changes.

We also recognize the potential endogeneity due to omitted variables if the unit values are correlated with unobservable variables that influence demand. Ideally, we would have desired to at least aggregate unit values at a geographic level but the data set has only two broad zones. We could have also used predicted unit values from a first stage by calculating the expected unit value using ordinary least squares or generalized linear models adjusting for socio-demographic variables. Unfortunately, this imposed us a trade-off, since doing so introduced collinearity with the variables included in the demand system. Instead, in households without reported unit value we imputed using the averaged unit values at zone level and preferred to use a set of demographic variables to model the decision to purchase the items and so include the IMR into the equations of the demand system.

Despite these data limitations, the LA/AIDS with IMR is our preferred model for several reasons. First, it allows to test the conditions of homogeneity and symmetry through linear restrictions on fixed parameters [28]. The system also produces low standard errors when the number of commodities is greater than six [39]. It also allowed us to estimate cross price-elasticities with standard errors. In addition, the LA/AIDS results are somewhat similar than those produced by the QUAIDS and the two-part model. This gives us confidence that the price-elasticity that we estimated is robust.

The type of tax on sweetened beverages produces different outcomes. A specific tax (a fixed amount of money per physical unit of product) has several advantages. It is easier to administer and provides more stable fiscal revenue than ad valorem tax (a percentage of the value of the product) [40]. In addition, it reduces the gap between expensive and cheap brands [41]. The specific tax should at least be indexed to inflation to avoid that it dilutes over time. It is also feasible to implement a mixed tax system, where specific and ad valorem taxes coexist. Ad valorem taxes have the advantage that they are automatically adjusted for inflation. In any case, since a tax and its subsequent price increase implies a money transfer from consumers to government, careful assessment of health and economic impacts should be done. In one hand, it is possible that reduction in consumption produces health gains, through weight loss, reduced risk of metabolic syndrome and other desirable effects on health. These consequences are to be seen in the medium and long term and only after a significant increase in prices. Low taxes on sweetened beverages have little or no impact on body weight for instance [23, 42]. Substitution to other caloric beverages is also likely and the reduction of calories from soft drinks could be offset by increases of calories in the other products. Additional fiscal revenue should be returned to consumers to reduce potential regressivity effects of the tax, such as providing public drinking fountains, and education programs to reduce the information asymmetry.