2. Background of the Spanish Christmas lottery draw

The Spanish Christmas lottery can be considered a unique lottery since 75% of Spaniards play it and the sales revenue reaches 0.3% of the Spanish GDP (Bagues & Esteve-Volart, Reference Bagues and Esteve-Volart2016). It takes place every year on December 22 and, except for some multi-state lotto games in the United States (e.g., Powerball and Mega Millions), it is the largest lottery draw in the world, offering a jackpot of 688 million euros in the 2020 edition. Unlike other lottery prizes, this prize, which is the total amount of the jackpot prize corresponding to all the issued tickets with the winning number, is usually spread among many people: up to 1,720 winners can win 400,000 euros each. If all the tickets are sold, the total issuance amounts to 3,440 million euros, of which around 2,408 million are allocated to prizes (70% payout rate).

The Spanish Christmas lottery is truly a national event in which society is involved in terms of individual purchases and syndicate play (Garvía, Reference Garvía2007). Lines of people waiting to buy a ticket from “lucky stores” (Guryan & Kearney, Reference Guryan and Kearney2008) are becoming more and more common. It should be noted that the tickets may become available as early as July each year.

Although the operator introduces changes to the characteristics of the draw with each new edition, the lottery is based on tickets with five-digit numbers ranging from 00,000 to 99,999, just like any regular draw of the Spanish National Lottery. The numbers allocated to each outlet are randomly determined (Bermejo et al., Reference Bermejo, Ferreira, Wolfenzon and Zambrana2020).

Due to the enormous popularity of the game, each set of numbers on each ticket is sold multiple times in several so-called “series”; for example, the 2020 edition had 172 series for each of the 100,000 tickets. Between the series, there is no difference in prize money. That is, the prize on a five-digit number is paid for that number in every unit. Therefore, the series number is merely an administrative one. The price of a whole ticket is 200 euros for each “serie.” Because this may be too expensive for many players, the tickets are usually sold as tenths, making the price of a tenth just 20 euros.

The jackpot prize amounts to 4 million euros for each “serie” of the winning number, or 400,000 euros for each tenth, which is equivalent to a return of 20,000 euros of prize money for every euro staked. Therefore, a winning ticket (20 euros) may receive more than the average wealth of a Spanish household (274,000 euros). There is also a second prize of 1.25 million euros (125,000 euros for each tenth) and a third prize of 500,000 euros (50,000 for each tenth). In addition, there are many other small prize categories. Due to data availability limitations, the focus of this study is on the first (jackpot) prize.

Finally, it should be mentioned that since 2013, the portion of a prize exceeding 2,500 euros is susceptible to a tax rate of 20%, whereas prizes of 2,500 euros or less are tax-exempt. In all cases, the prizes are paid immediately through a unique payment.

3. Data and model specification

Information on the variables related to household structure decisions (the numbers of marriages, divorces, and births – per 1,000 people) at the province level is collected by year for each of the 52 provinces of Spain between 2000 and 2018. The sample period is limited by the availability of Christmas lottery prize data. Each prize is expected to have enough variability in its temporal and geographical dimensions to allow us to distinguish between temporal and geographical effects.

To further describe the dependent variables, Fig. A1 in the appendix describes the evolution of the marriage, divorce, and birth rates over the period 2000–2018. It can be seen that the marriage rate has been in continuous decline in Spain in recent years. Besides the generalization of cohabitation as an alternative to marriage (Hiekel & Castro-Martín, Reference Hiekel and Castro-Martín2014), this pattern can also be explained by economic conditions. In particular, the recent deterioration of the Spanish labor market may be contributing to delays in marriage decision-making (De la Rica & Iza, Reference De la Rica and Iza2005; González-Val & Marcén, Reference González-Val and Marcén2018; Gutiérrez-Domènech, Reference Gutiérrez-Domènech2008). According to the Spanish Statistical Office, in 2018, the unemployment rate in Spain was 15.25%; at the end of the analysis period, Spain had the second-highest unemployment rate among EU countries. In this context, Bellido and Marcén (Reference Bellido and Marcén2021) concluded that marriages exhibit a pro-cyclical behavior in Europe. Even though a rise in the divorce rate can be observed in the early 2000s, the divorce rate seems to exhibit a steady flat trend in recent years. In this respect, Bernardi and Martínez-Pastor (Reference Bernardi and Martínez-Pastor2011) discussed how divorce risk factors have changed over time in Spain; González-Val and Marcén (Reference González-Val and Marcén2018, Reference González-Val and Marcén2020) found different effects of the gender unemployment rate on the divorce rate.

Regarding births, the declining trend in fertility (and thus births) over the last decade has been particularly evident in southern European countries. These countries were strongly affected by the financial crisis of 2007–2008 and present high rates of unemployment and precariousness (Matysiak et al., Reference Matysiak, Sobotka and Vignoli2021). In this regard, Bellido and Marcén (Reference Bellido and Marcén2019) suggested that economic uncertainty and unfavorable labor market conditions might have negatively impacted fertility rates in 30 European countries from 1993 to 2013.

As the main covariate, an indicator of whether there has been a random wealth shock in a province is created based on information from the Christmas lottery. Figure A2 in the appendix shows the number of provinces that won at least a portion of the jackpot prize and the corresponding amount. The winning provinces are identified by the location of the sales points that sold the jackpot-winning tickets since most prizes are collected in the province where the tickets were sold (Bermejo et al., Reference Bermejo, Ferreira, Wolfenzon and Zambrana2020). This is a seemingly strong assumption since lottery players may acquire tickets from other provinces either by exchanging or sharing tickets with relatives or people in their social networks or by purchasing lottery tickets outside their province of residence, among others. Bagues and Esteve-Volart (Reference Bagues and Esteve-Volart2016) empirically supported such an assumption, concluding that Christmas lottery winners tend to live in the same province where the lottery tickets were sold.

Based on the information collected, a dummy variable for a jackpot-winning province is created and coded as one where all the jackpot-winning tickets are sold in that province: the only-winning province. Accordingly, up to 1,720 individuals in an only-winning province are expected to collect 400,000 euros each; on aggregate, such a wealth shock exceeds 5% of the median GDP of the Spanish provinces in 2018. Because Christmas lottery players often play in a syndicate (Garvía, Reference Garvía2007), sharing lottery tickets, this figure should be considered as a lower bound of the number of individuals receiving lottery winnings. As reported later on in the paper, a further analysis is performed where the ratio of the collected jackpot share for each province to that province's GDP is used as the main explanatory variable (discussed in the robustness checks section of the paper).

Table 1 shows relevant figures for the dependent variables (nuptiality rate, divorces and separations rate, and birth rate) in the only-winning provinces for the treatment year and four years later. The nuptiality rate consistently decreases in all these provinces while both the divorces and separations rate and the birth rate show more variability, with some provinces experiencing decreases and others experiencing increases over the four-year period.

Table 1.

Only-winning provinces figures at year t and year t + 4

* Values for 2020 should be interpreted with caution due to the effects of COVID-19.

In addition, for an only-winning province, exogenous good economic conditions are generated that may produce a temporary increase in the happiness of that province's inhabitants, including non-lottery players. In turn, this can affect household structure decisions, even when the number of lottery winners is limited to a small number of people.

The available information has a panel data structure, which allows us to control for changing socio-economic and demographic conditions among provinces and for the effects of unobserved variables. Accordingly, macroeconomic and social information for the selected period is gathered from multiple public access databases to control for the characteristics of provinces that can affect household decision-making. This gives the following covariates for inclusion in the model. The real GDP, measured in euros with the purchasing power of 2016, controls for the income effect. The ratio of the foreign population controls for the population's social structure and tries to capture social trends, such as the effect of the migrant population's evolution. The ratio of the population with higher education works as a proxy for human capital, while the unemployment rate by gender allows for a comparison of the effects of male and female unemployment on family structure decisions. Under the life cycle hypothesis (Modigliani & Brumberg, Reference Modigliani and Brumberg1954) a relationship may exist between household savings and their feelings of an upcoming marriage, divorce, or birth. In line with previous studies, such as González and Özcan (Reference González and Özcan2013) and Voena (Reference Voena2015), the amount of money allocated to fixed-term deposits is used as a proxy for household savings; the year fixed effects control for possible time-dependent effects. Moreover, a dummy variable controlling for the Express Divorce Law in Spain, Law 15/2005, is included in the corresponding model specification.

Table A1 in the appendix reports the sources of the considered variables, while basic descriptive statistics of the variables are shown in Table 2. The data sample used in the empirical part of the study comprises 988 observations from 2000 to 2018.

Table 2.

Summary statistics for the key variables

* Information for Ceuta in the year 2015 is not available.

^† All winning tickets are sold in a unique province.

To check the randomness of winning provinces, a panel probit model is estimated in which the variable to be explained is the probability of a province being an only-winning province. The explanatory variables include the following macroeconomic variables: GDP, the ratio of the population with higher education, and the unemployment rate. The estimates in Table 2A in the appendix show that province characteristics do not predict the probability of an only-winning province.

Both static and dynamic approaches are considered for the analysis of the wealth shock. The empirical study is based on a linear panel event-study following Freyaldenhoven et al. (Reference Freyaldenhoven, Hansen, Pérez and Shapiro2021). Since the Christmas lottery draw takes place on December 22 every year, we consider that the treatment takes place the following year after receiving the jackpot. The static effects are estimated as follows:

(1)$$y_{it} = \alpha _i + \delta _t + q_{it}\gamma + \beta z_{it} + C_{it} + \varepsilon _{it}$$

where Y _it is the outcome variable (marriage rate, divorces and separations rate, and birth rate) for province i in year t, α _i is the province fixed effect, δ _t is the year fixed effect, and q _it is a vector of the covariates that could affect the outcome variable. C _it represents a confound that may be correlated with the positive wealth shock. Z _it is a vector of dummy variables for the only-winning province which takes the value 1 when all the winning tickets have been sold in a unique province in the year t. The key parameter of interest is β; it measures the static effect of the treatment.

The analyzed wealth shock exhibits variation in treatment timing and may involve dynamic effects. This design allows for tracking the observed units before the lottery win and catching pre- and post-wealth shock (treatment) trends, thereby providing a representation of the event's causal impact. The dynamic effects are estimated as follows:

(2)$$y_{it} = \alpha _i + \delta _t + q_{it}\gamma + \mathop \sum \limits_{m = {-}G}^M \beta _mz_{i, t-m} + C_{it} + \varepsilon _{it}.$$

Here, the outcome variable at year t + 1 might be affected by the jackpot prize at most M ≥ 0 periods before t + 1 and at most G ≥ 0 periods after t + 1. Then, the term $\sum\nolimits_{m = {-}G}^M {\beta _mz_{i, t-m}}$ could be interpreted as the effect of the wealth shock during the G periods after treatment. The value of both G and M is assumed to be 4 since the trend during the four years before the treatment is also considered a control. Accordingly, the analysis excludes the periods 2000–2006 and 2016–2018. In this case, the term $\sum\nolimits_{m = {-}G}^M {\beta _mz_{i, t-m}}$ measures the year-by-year effects four years before and after the lottery winning. The effects for the endpoint years – those occurring more than four years before and after the wealth shock – are measured through two aggregate coefficients: one preceding and another following the shock.

This model, as a generalized extension of “difference-in-differences” or two-way fixed-effect models, allows for dynamic lags and estimating the event of interest while also controlling for fixed factors by area (province) and time (year) (Clarke & Tapia-Schythe, Reference Clarke and Tapia-Schythe2021).

4. Empirical analysis and results

4.1. Static analysis

Table 3 shows the results of the static analysis. The results reveal the positive impacts of lottery winnings on the birth and divorce rates. However, the nuptiality ratio seems to be unaffected by the economic shock: while the coefficient shows a positive direction, it does not reach statistical significance.

Table 3.

A linear panel event-study design

Static analysis results.

Clustered standard errors at province level in brackets*, **, and *** denote significance at the 10%, 5%, and 1% level, respectively.

4.2. Dynamic analysis

The estimates from the dynamic approach are shown in Table 4; the parameters summarizing the magnitude of the dynamic effects are plotted in Figs 1–3.

Figure 1.

Event-study plot (nuptiality, marriages rate).

Figure 2.

Event-study plot (divorces and separations rate).

Figure 3.

Event-study plot (birth rate).

Table 4.

A linear panel event-study design

Dynamic analysis results.

*, **, and *** denote significance at the 10%, 5%, and 1% level, respectively.

The results show a significant effect of the wealth shock variable on the three dependent variables. Indeed, divorces and separations seem to increase during the second and third years after the shock. On average, marital breaks increase by 0.373 and 0.544 per 1,000 inhabitants in the second and third years, respectively. Concerning the birth and nuptiality rates, a more lasting effect is observed in the years following the wealth shock, with both rates gradually increasing over time.

4.4. Robustness checks

In the 2004 draw, all El Gordo lottery tickets that won the jackpot were sold in the town of Sort (Lleida) in a single outlet (La Bruixa d'Or). The total prize, 390 million euros, was spread over various regions: more than half of the tickets were sold online and the rest were distributed to lottery players in Mallorca, Madrid, and Jaen. The lottery outlet in Lleida is famous for having sold the winning tickets of the Christmas lottery jackpot on three occasions: 2003, 2004, and 2007. Table 5 reports the estimated coefficients from the dynamic event-study removing the province of Lleida from the sample. The results remain largely unchanged and are in line with those discussed above.

Table 5.

A linear panel event-study design

Dynamic analysis results excluding Lleida.

*, **, and *** denote significance at the 10%, 5%, and 1% level, respectively.

The purchase of lottery tickets by households could be inclined toward summer vacations in coastal provinces. To check this, the event-study model (equation (2)) was re-estimated excluding the coastal provinces from the sample. The results (Table 6) are in line with those previously obtained. Another recent phenomenon to be considered is the increasing possibility of buying Christmas lottery tickets through the Internet. However, as discussed by Bagues and Esteve-Volart (Reference Bagues and Esteve-Volart2016), only around 2% of all tickets are purchased online.

Table 6.

A linear panel event-study design

Dynamic analysis results excluding coastal provinces.

*, **, and *** denote significance at the 10%, 5%, and 1% level, respectively.

As an additional robustness check and to provide more informative results, a two-way fixed effects panel data are estimated. In this model, following Bagues and Esteve-Volart (Reference Bagues and Esteve-Volart2016), the indicators of only-winning provinces used in the previous models are replaced with the ratio of the collected jackpot share for each province to that province's GDP. Thus, all the provinces that have obtained at least a portion of El Gordo are now considered:

(3)$$Y_{it} = \alpha _i + \gamma _t + \beta ^{DID}Jackpot_{it-k} + \delta X_{it} + \varepsilon _{it}, \;$$

where Y _it is the analyzed outcome variable (marriage rate, divorces and separations rate, and birth rate) for province i in year t. The terms α _i and γ _t represent the province and year fixed effects, respectively. X _it is the set of controls at the province level; $\varepsilon _{it}$ is a random error term. The variable Jackpot _it is defined as the ratio of the effectively collected jackpot share for each province to the corresponding province's GDP in the year t − k, where k takes the value 1, 2, 3, or 4. To account for non-linear effects, the squared terms for this variable are considered.

The estimates in Table 7 are in line with those obtained from the static analysis. A positive effect of the jackpot share received by a certain winning province on both the divorces and separations rate and the birth rate in that province is noted years after the lottery winning. Thus, an increase of 1% in the jackpot/GDP ratio causes a 0.006 increase in divorces and separations per 1,000 inhabitants. Concerning having children, a similar but greater effect on the birth rate is observed: a 1% increase in the jackpot/GDP ratio causes a 0.013 increase in births per 1,000 inhabitants. In both cases, according to the shape of the estimated relationship, an additional increase in the jackpot/GDP ratio will result in a decreasing marginal effect due to the negative coefficient of the quadratic wealth shock variable. No effect on the number of marriages is apparent.

Table 7.

A more informative specification

Clustered standard errors at province level in brackets.

Panel data estimates (TWFE model). Lleida and Madrid are not included.

*, **, and *** denote significance at the 10, 5, and 1% level, respectively.

5. Concluding remarks

The decisions of households in developed countries are much more complex today than in past decades. For this reason, the analysis of household decision-making and behavior and the study of the variables that affect household decisions have attracted interest in recent years. This paper presents an empirical exercise that aims to estimate the effect of a significant positive variation in household wealth and/or the surrounding economic conditions on decisions that go beyond mere consumption expenditure and that could affect the household's structure.

Several previous studies have explored the influence of exogenous (unearned) random wealth shocks on various household decisions using lottery winnings as a proxy for such variations in wealth or environmental economic conditions. In this paper, the Christmas draw of the Spanish National Lottery is used as a natural experiment to explore the effect of wealth shocks on household structure decisions. The focus is on exploring whether the fact that the Christmas draw's winning tickets are sold in a particular province affects the number of marriages or divorces and the number of births in that province. The aggregate size of such an exogenous random shock, totaling 700 million euros, is equivalent to 5.28% of the median GDP of Spanish provinces in 2018. Additionally, winning the lottery may generate exogenous good economic conditions in the province, thereby affecting people's overall happiness and decision-making.

Both static and dynamic linear panel event studies were conducted to consider dynamic lags and estimate the event of interest. The results from both studies indicate that a positive unexpected exogenous (unearned) wealth shock is linked to an increase in the number of births and divorces in a province. However, no conclusive results are found regarding the number of marriages. Furthermore, the magnitude of the unexpected impact seems to be conditioned by the amount of the effectively collected prize. Overall, the evidence is consistent with couples getting divorced and having children when their province's economic conditions change due to an unexpected increase in wealth.

From the perspective of policy implications, these results can provide valuable insights for policymakers to increase their accuracy when identifying the regions in which to implement social policies. Given the positive effect found on births, a possible policy implication would be to validate measures that encourage the birth rate by offering monetary support (exogenous) and supporting the decision to have children. Moreover, the statistical data on household structure decision-making have broader implications for population research.