In this section, I present the empirical results. The first two subsections detail the results with respect to the effect of the inheritance shock on hospitalization estimated using Model 1 and Model 2 respectively. The estimates imply that the inheritance shock increases the likelihood of hospitalization for any cause by around 5%. In the third subsection, I compare the impact of the inheritance shock to the impact of the inheritance as such and show that these are largely similar. The results from several placebo tests are discussed in Section 7.4, and these suggest that the main estimates could be interpreted as causal effects. Further support for this is reported in the fifth subsection, where the analysis is restricted to inheritances from sudden deaths. Section 7.6 reports results with respect to the dynamics of responses. In Section 7.7, I show that the effect is more pronounced for women, the relatively old, those with low education, and those receiving relatively large inheritances. The following subsection shows that a non-trivial share of the effect in hospitalization could be attributed to higher incidences of symptoms and signs of disease and cancer. In the final subsection, I present results suggesting that the inheritance shock does not have any detectable effects on sick leave or on mortality.
The effect of the inheritance shock on hospitalization—Model 1
The DID estimates of the effect of the inheritance shock on hospitalization reported in this section have been obtained from versions of Model 1 estimated using OLS.Footnote 43 Given that hospitalization is binary, the estimates should be interpreted as the percentage point difference in the probability of the outcome between the treated and the controls. In connection to the regression estimates, I report the mean of the dependent variable (in brackets), in percent, for the post-inheritance period for the relevant control group. Dividing the DID estimate by this statistic gives the effect in percent. In each specification, the standard errors have been clustered at the individual level to account for correlation within the individual over time.
Column 1 in Table 4 reports the DID estimate obtained from Model 1 with only time controls. This is comparable to the naïve DID estimate implied by the statistics in Table 2. As expected, the estimate implies that the treated subjects have 0.2 percentage point higher probability of being hospitalized in the pre-inheritance period relative to the controls.
Column 2 reports the DID estimate from Model 1 with time and year controls. Similarly to the estimate in column 1, the estimate is positive and statistically significant (p < 0.05) but, notably, almost twice as large (0.43). This discrepancy suggests that the treated and the controls experience differential year trends and that year controls indeed are essential.Footnote 44
Column 3 reports the DID estimate from Model 1 augmented with a set of predetermined individual-level covariates that are standard in the literature, including gender, a second-order polynomial in age, marital status, presence of children, level of education, income, and wealth.Footnote 45 The model also keeps baseline health constant by controlling for the total number of hospitalization episodes over the 4 years preceding the inheritance. This approach has previously been proven useful when it comes to accounting for sources of otherwise uncontrolled heterogeneity (Adams et al. 2003; Gardner and Oswald 2007; Kim and Ruhm 2012). The results indicate that the inclusion of controls changes the point estimate of the wealth effect and the standard error only with respect to the third decimal.Footnote 46
Column 4 reports the DID estimate from the model specification with individual fixed effects. It should be noted that the estimate of the wealth effect is largely similar to the estimates in columns 2 and 3, implying that any bias from unobserved heterogeneity across individuals (e.g., differences in genetics, time preferences, or early childhood exposures) is small. Moreover, the slight decrease in the standard error, as compared to the previous specifications, suggests that the efficiency gain from this extension of the model is trivial.
Taken together, the results in Table 4 suggest that the inheritance shock leads to a 5% increase in the probability of hospitalization. Is this a large or a small effect? To get some perspective regarding this issue, I compute the cross-sectional relationship between age (in years) and hospitalization, as it is well known that age has a large impact on health. It turns out that the effect of the inheritance shock equals the impact of being about 4 years older, suggesting that the wealth effect is non-trivial. However, when I relate the wealth effect to the impact of education, another factor related to health status (Lleras-Muney 2005; Cutler and Lleras-Muney 2010), I find that having primary or lower secondary education, as compared to upper secondary or postgraduate education (i.e., the impact of having lower education), increases the probability of hospitalization by 18%. This suggests that the effect on health of a 7% increase in wealth should be considered comparably small.Footnote 47
The effect of the inheritance shock on hospitalization—model 2
Table 5 details the DID estimates obtained from Model 2. These estimates are reported in percent, and given the fact that the inheritance shock is in log, they provide the percentage point change in the probability of the outcome from a 1% inheritance shock.
In line with what I found in the previous section, the specification with only time effects (column 1) generates a point estimate that is positive and statistically significant at the 10% level. Turning to column 2, it may be noted that the estimate from the specification with both time and year effects is larger in magnitude than the previous estimate, but also that it is statistically insignificant. The estimate as well as the standard error remains similar when I control for observable characteristics in column 3.
In column 4, however, we see that the DID estimate from the specification with individual fixed effects is positive and statistically significant (p < 0.05). The magnitude of the estimate implies that an inheritance shock of 1% increases the incidence of hospitalization by 0.03 percentage points, or 0.4% if compared to the post-period mean of the variable.
How does this finding compare to the finding in the previous section? Assuming that the effect is consistent across the two models, the estimate in column 4 suggests that a 7% increase in wealth (i.e., the average wealth increase due to the tax reform) increases the likelihood of hospitalization by 2.9% (7 * 0.4). While this response is somewhat lower than the one obtained from Model 1, it is nevertheless of the same order of magnitude. The level of significance is also similar across the two models (p < 0.05), thereby suggesting that there is no evident efficiency gain from exploiting the variation in shock size.
The conclusion that may be drawn from the results in Table 5 is that the positive response in hospitalization is increasing in relation to the size of the inheritance shock. This contrasts with the estimates of the gradient between wealth and health. In Table 12, in Appendix 2, I report the estimate of the inheritance shock alongside the estimate of the gradient (in the current study population) and this comparison shows that the 95% confidence intervals are far from overlapping, suggesting that the two estimates are statistically distinguishable from each other.
A comparison between the effect of the inheritance shock and the effect of inheritance on health
The results in the previous subsections show that the inheritance shock is associated with an increased likelihood of being hospitalized. A question that remains, however, is whether the estimated effect merely picks up the effect of additional inheritance.
I investigate this by controlling for the inheritance amount in the estimations of Model 1 and Model 2. Since I do not have information on inheritance for heirs of decedents who passed away after the tax repeal, I use the imputed inheritance (see Section 4.1). In Table 16 in Appendix 6, I report the results from estimations that control, separately, for the inheritance in level, log, as well as in the form of a third-order polynomial.Footnote 48 The estimates with respect to the inheritance shock are qualitatively and quantitatively similar to the main estimates, implying that the impact of the inheritance shock remains even when controlling for the underlying inheritance.
Despite the results reported previously, it is interesting to investigate if the impact of the inheritance shock could be interpreted as the effect of inheritance as such. Estimating the inheritance effect is inherently difficult, however, since there is no obvious counterfactual. People who lose a parent but do not receive an inheritance, as the parent lacked wealth, are likely to differ from those who receive an inheritance. Moreover, a comparison of health responses between inheritors and individuals who do not lose a parent (and thus do not receive an inheritance) may be confounded by responses in health that are due to mourning and grief.
I propose two different strategies for resolving these issues. The first strategy (Strategy 1) exploits the fact that children of married couples do not receive an inheritance when the first parent passes away, but instead a postponed right to inherit (see Section 4.1). Children who receive a postponed inheritance right during the study period are included in the data and I use them as counterfactuals to those who do receive an inheritance in order to estimate the effect of receiving an inheritance on hospitalization. The estimation is essentially a difference-in-difference estimator that compares the incidence of hospitalization before and after the inheritance between heirs in the main sample (net of heirs inheriting in 2005, to avoid confounding impacts of the tax reform) and heirs with a postponed right (who would have received inheritances of equal sizes as the heirs in the main sample had they inherited today).Footnote 49 The regression results are reported in Table 17 in Appendix 6 and these show that receiving an inheritance increases the likelihood of hospitalization by 1.7 percentage points, or compared to the baseline incidence, almost 25%.Footnote 50 This is five times the effect size of 5% found in the main analysis. Given that the average value of inheritances in the treatment group is 548,000 SEK, or eight times the average inheritance shock, the expected effect size, if the relationship between inheritance and hospitalization would be linear, is 40% (5% * 8) rather than 25%. However, while the result suggests that the effect of receiving an inheritance on hospitalization may be non-linear, the effect is in the same ballpark as the effect implied by the inheritance shock.
The second strategy (Strategy 2) exploits variation in timing of heirs who received an inheritance during the period of 2003–2004.Footnote 51 More specifically, I compare the before-after change in hospitalization of the cohort of heirs who inherited in 2003 with the same before-after change among heirs in the 2004 cohort, who are identical, except that they inherit 1 year later.Footnote 52 The counterfactual, in other words, is inheriting the next year rather than this year.Footnote 53 The regression results are displayed in the second column in Table 17 in Appendix 6, and here we see that they are similar to those from the previous strategy in that the inheritance effect is positive and statistically significant. However, the response is somewhat larger, 37%, and, subsequently, more in line with the expected effect size of 40% implied by a linear relationship.
In sum, the results in this section suggest that the effect of the inheritance shock is largely similar to the effect of an inheritance as such. This is comforting as inheritances represent one of the most common increases in wealth that people experience in life.
In this section, I present results of several placebo tests designed to establish whether the main estimates represent a causal relationship, and not just a spurious correlation.
I begin by estimating Model 1 on the BTT sample (Placebo test I).Footnote 54 An insignificant response in hospitalization, or at least a DID estimate that is smaller in magnitude than the corresponding estimate for the main sample, should be considered as supporting the causal interpretation of the main estimates. The results are reported in the first column of Table 6 and these show that the coefficient estimate on the treatment indicator is positive but smaller than the main estimate and also statistically insignificant, suggesting that the repeal of the inheritance tax does not have an impact on the incidence of hospitalization of heirs receiving an inheritance below the tax threshold.
The second placebo test (II) tests for an impact of the reform among children who received a postponed right to inherit during the period of 2003–2005 (see Section 7.3 for a discussion). As these children did not receive an inheritance during this period, they should not be affected by the tax repeal. To test for this, I estimate Model 1 with the treatment indicator indicating children who received a postponed right to inherit following the tax repeal (i.e., in 2005). I restrict my focus to those who would have received inheritances of equal sizes as the heirs in the main sample had they inherited today. The results of this test are reported in column 2 and they show that the coefficient estimate on the indicator is negative and statistically insignificant, suggesting that the tax repeal does not offset any confounding effects.
The third placebo test (III) tests for differential responses in hospitalization between heirs in the main sample inheriting in 2003 and 2004 (i.e., before the reform). Since both of these cohorts were unaffected by the tax repeal, we should not expect to find a differential response in the outcome in the years following the inheritance. The test is carried out by estimating Model 1 with the treatment indicator indicating whether the individual inherits in 2004, as compared to in 2003. The coefficient on the treatment indicator could thus be interpreted as the effect of the reform in the counterfactual case that was implemented 1 year before the actual implementation. The results are presented in column 3 and here we see that the coefficient estimate on the indicator is negative (as opposed to positive in the main analysis) and statistically insignificant at conventional levels.
The results in this section can be seen as supportive evidence that the main estimates of the effect of the inheritance shock on hospitalization represent a causal relationship.
Additional test of the exogeneity of the inheritance shock using sudden deaths
The causal interpretation of the main estimates reported in Sections 7.1 and 7.2 hinges on the premise that the inheritance shock is exogenous. The results from the placebo tests reported in Section 7.4, as well as the results in Section 6, suggest that this is indeed the case and that the tax repeal approximates the ideal experiment of random assignment of additional inheritance fairly well. However, to further investigate the credibility of this assumption, I use data from the Cause of Death Register to identify children of parents who passed away suddenly and test for whether responses in this group differ from those in the main analysis. Adding the restriction that the death should have been sudden may be thought of as adding an additional natural experiment on top of the existing one, further strengthening the exogeneity of the inheritance shock. For instance, focusing on sudden deaths is likely to effectively account for any remaining unobservable differences between the treated and the controls that may be due to differences in estate planning among the decedents who passed away before and after the tax repeal. The classification of sudden deaths (natural and unnatural) follows the classification in Andersen and Nielsen (2011).Footnote 55 Of the children in the main sample, 17% have parents who passed away suddenly and I use these children in estimations of Model 1 and Model 2. The results are reported in Table 18 in Appendix 7. The estimate in the first column, generated by Model 1, is statistically significant and positive, implying that the inheritance shock increases the likelihood of hospitalization. Whereas the effect is somewhat larger in size than the main estimate, it is of the same order of magnitude.Footnote 56 The same goes for the estimate obtained from Model 2 (reported in column 2), but with the difference that it falls just above the 10% statistical significance level (p = 0.137).
Taken together, the similarities between the estimates reported in this section and the main estimates may be viewed as further support for the empirical strategy and the causal interpretation of the main estimates.
Dynamics of responses
The results in Table 4 and Table 5 give us no sense of the dynamics of the wealth effect—whether it accelerates or stabilizes over time. To explore these dynamics, I estimate Model 1 and Model 2 with leads and lags of the treatment. More specifically, I include interactions between the treatment variable (discrete or continuous) and time dummies for each of years before the inheritance, the year of the receipt, and for each of the subsequent years. The results, reported in Table 7, show that the coefficient estimates on the lead indicators from both models are statistically insignificant. This is comforting, as it suggests that the parallel trend assumption is indeed satisfied. As for the pattern of the lag structure, it shows that the difference in probability of hospitalization between the treated and the controls increases sharply at the time of the inheritance receipt. This could be viewed as additional support for the causal interpretation of the wealth effect. It should be noted, however, that the implied effect varies across the years and that it is only statistically significant for the second, third (only Model 2), and fifth years after the inheritance. In Table 19 in Appendix 8, I report results from the three placebo tests (described in Section 7.4) using the dynamic version of Model 1. Except for one lead indicator in the first test, the lead and lag estimates are statistically insignificant, a finding that further strengthens the belief in the main estimates.
In this section, I test for how the wealth effect varies with respect to socioeconomic characteristics.
Table 20 in Appendix 9 displays estimates of heterogeneous effects with respect to age, gender, and education, obtained from Model 1 (upper panel) and Model 2 (lower panel) with time, year, and individual fixed effects, in addition to a second-order polynomial in age.
The results show that the effect is markedly larger for old heirs (above the mean age of 53 years) than for young heirs (below the mean age). This finding corresponds with previous studies (see, for example, Lindahl 2005). Moreover, I obtain imprecisely measured DID estimates of the wealth effect when the population is limited to the working-age individuals (between 16 and 65 years) and the post-inheritance period is restricted to 4 years (to be comparable with the estimates with respect to sick leave). Although this finding may be a consequence of the shorter time period, it accords with the previous finding that the response in hospitalization is primarily driven by the relatively old.
The estimates from Model 1 suggest that the effect is primarily driven by women and not by men, whereas the estimates from Model 2 suggest that differences in responses across the sexes are negligible.Footnote 57 Regarding education, the wealth effect is positive and statistically significant for heirs with primary or lower secondary education and negative and statistically insignificant for heirs with upper secondary or postgraduate education. This finding is in line with previous research documenting that highly educated individuals have more knowledge about, and are better at avoiding and managing, harmful health effects than their peers with lower levels of education (Goldman and Smith 2002).
Models of health production, as well as previous empirical studies, suggest that the wealth effect should be increasing in relation to the size of the shock. When I explore this in more detail, by estimating the wealth effect separately for heirs receiving an inheritance within the first, second, third, and fourth quartiles of the distribution using the two models, I find that the effect is solely driven by heirs in the third quartile of the distribution (see columns 1–4 in Table 21, Appendix 9). While the lack of an effect of the shock for the two lowest quartiles may be a consequence of the shock being too small to have any implications, it is more difficult to explain why the largest shocks have no impact. Wealthy heirs, as compared to poor, tend to receive larger inheritances in absolute terms; however, as a fraction of their wealth, the inheritances are typically relatively small (Wolff 2002; Elinder et al. 2016). To investigate whether the inheritance shock matters more for the relatively poor than for the relatively wealthy, I test for heterogeneous responses across the distribution of the inheritance scaled by initial wealth.Footnote 58 The results show that the effect is only evident for heirs in the third quartile of the distribution (see columns 5–8 in Table 21, Appendix 9). The lack of effect for the two bottom quartiles may again be explained by the insignificant importance of the shock. However, that the effect is only evident for the third and not for the fourth quartile is somewhat puzzling, but may indicate diminishing returns to wealth.
Explaining the wealth effect on hospitalization
Taken together, the results in the previous sections suggest that the inheritance shock leads to an increase in hospitalization. At a first glance, this finding suggests that increased wealth has detrimental effects on health. One should, however, keep in mind that hospitalization does not inform us about the reasons for the hospital admission. To place this issue in perspective, I therefore continue and test for heterogeneous responses across the diagnoses reported in connection with the hospital admissions. In this section, I report regression results for the effect of the inheritance shock on the diagnosis indicators detailed in Section 4.
Column 1 in Table 22 in Appendix 10 displays the discrete DID estimates obtained from Model 1 (with time, year, and individual fixed effects) estimated on the main sample. It is noticeable that there are only two outcomes for which the DID estimate is statistically significant (p < 0.05): neoplasms and symptoms and signs. The estimate with respect to neoplasms implies that the inheritance shock leads to a nearly 14% increase in the probability of the outcome, whereas for symptoms and signs, the coefficient implies an increase of 16%. Taken together, the two effects explain around 70% of the effect in hospitalization. The fact that there is no significant response in any other variable (neither in the single diagnosis variables nor in the variable others) suggests that the wealth effect on hospitalization is operating solely through symptoms and signs and neoplasms. Moreover, estimates obtained from the three placebo tests, described in Section 7.4, are statistically insignificant with respect to both diagnoses, suggesting that the main estimates are causal (see Table 23 in Appendix 10). In column 2, I report the DID estimates from Model 2 estimated on the main sample. These show a similar pattern, in terms of sign and level of significance, to the estimates from Model 1. In accordance with the previous comparison of the estimates from the two models, they are of the same order of magnitude in percentage terms (if evaluated relative to the average inheritance shock).
What do the responses in symptoms and signs and neoplasms tell us about the mechanisms through which wealth affects hospitalization?
The variable symptoms and signs, as indicated by the name, captures symptoms and signs of disease (e.g., irregular heart rate, shortness of breath, fever, senility, general feeling of illness), as well as unusual findings during medical examinations (e.g., blood and urine samples).Footnote 59 Given that the condition has resulted in a hospital admission, the response in the variable may on the one hand imply that the inheritance shock leads to worse health, and potentially more so if we had studied the effects over a longer period of time.Footnote 60 On the other hand, the response could be interpreted as if the shock has resulted in people being more prone to seek care for health irregularities, possibly to reduce the likelihood of more severe conditions in the future. This is in line with previous studies, which document that economic circumstances are positively associated with disease prevention (see, for example, Cawley and Ruhm, 2011).
Regarding neoplasm, it contains diagnoses of cancers at different stages of development (i.e., benign, potentially malignant, and malignant tumors).Footnote 61 It is difficult to give an analytical explanation for why the inheritance shock causes an increase in the likelihood of cancer, especially since it is commonly considered an equal opportunity disease (Smith 2004). Although lifestyle factors such as smoking and drinking, which are reported to be positively related to improved wealth (Apouey and Clark 2015; Kim and Ruhm 2012), are linked to many types of cancers (e.g., lung, head and neck, pancreatic, liver, colon, gastric, see, for example, Kushi et al. 2012), it seems unlikely that an increase in these risk factors would manifest into a higher cancer incidence within a period of only 6 years. If the inheritance shock has caused people to smoke and drink more, we should rather expect to find responses in diagnoses that are more immediately related to these risk factors, such as injuries (e.g., alcohol poisoning), mental and behavioral disorders, diseases in the digestive system (e.g., liver cirrhosis), respiratory diseases (e.g., chronic obstructive lung disease), and circulatory diseases (e.g., coronary heart disease and stroke) (World Health Organization, WHO 2002). Moreover, previous studies report that improved wealth leads to reduced obesity (Lindahl 2005; Kim and Ruhm 2012) and improved mental well-being (Gardner and Oswald 2007; Apouey and Clark 2015). But, if the inheritance shock exploited here has led to reduced obesity or improved mental well-being, we should, if anything, expect to find a reduction in cancer incidences, and not an increase (Kushi et al. 2012; Chida and Steptoe 2008).
One possible though speculative explanation for the positive response in neoplasm is instead that the inheritance shock has led to more health care visits in general, as indicated by the results with respect to symptoms and signs, and that cancer, which would otherwise have remained undiagnosed or been diagnosed later, is detected and possibly treated earlier. Thus, the higher incidence of hospitalization does not necessarily mean that the inheritance shock has detrimental effects on health, but rather that it results in more preventative actions against future morbidity.
To investigate the plausibility of this explanation, I estimate the dynamics of the response in the two diagnoses using Model 1 and Model 2 augmented with leads and lags of the treatment, as in Section 7.6. This would be seen as supporting the explanation if the response in symptoms and signs precedes the response in neoplasm. The results from this exercise are reported in Table 24 in Appendix 10 and they show that the responses in the two diagnoses generally occur simultaneously (in the second, third, and fourth year following the inheritance). The exception is for Model 1, where there is a response in symptoms and signs already in the year of the receipt and no similar response in neoplasm. Although this finding could be viewed as supporting the explanation, it should be noted that there is a response in symptoms and signs (but not in neoplasms) also in the year prior to the receipt (t = −1) and, subsequently, that one should be careful not to infer too much in this regard.
This pre-inheritance shock response in symptoms and signs also suggests that one should be careful in drawing the conclusion that the relationship between the diagnosis and the inheritance shock is causal. However, in Table 25, I report estimates for the dynamic impact of the inheritance shock on symptoms and signs for heirs whose parents passed away suddenly (and for whom the exogeneity assumption is likely to be more plausible), and these show that there is no pre-inheritance shock response suggesting that the relationship, in this sub-population, is causal. While this finding speaks in favor of the explanation that symptoms and signs is responsible for the response in hospitalization, one should interpret it cautiously as the underlying analysis is based on a subset of the main study population.
The effect of the inheritance shock on sick leave and mortality
In this section, I complement the previous analyses by investigating responses in outcomes capturing health events that are both less and more severe than those resulting in hospital admissions. More specifically, I estimate the effect of the inheritance shock on sick leave (less severe) and mortality (more severe).
Table 26 in Appendix 11 reports DID estimates from Model 1 and Model 2 with respect to sick leave. The models are estimated on heirs of working age in the main sample over a period of 10 years before and 4 years after the inheritance receipt. A comparison of the estimates from the models estimated with and without year fixed effects indicates that the treated and the controls experience differential year trends in the outcome. This is in line with what I found for hospitalization. In this case, however, the DID estimates from the preferred specifications of the models are all statistically insignificant, implying that the inheritance shock does not have any evident effect on the likelihood of sick leave.Footnote 62 It should be noticed that I cannot rule out the possibility that the inheritance shock has consequences for health events captured by sick leave for heirs who are younger than 16 and older than 65.Footnote 63 However, the fact that the wealth effect with respect to both sick leave and hospitalization is statistically insignificant for the working-age population lends additional support for the conclusion that the inheritance shock generated by the tax repeal has limited health consequences for the majority of the affected population.Footnote 64
The impact of the inheritance shock on mortality is estimated by comparing the difference in the probability of dying over the post-inheritance period between treated and the controls in the main sample with the similar difference for the BTT sample (see Section 5).Footnote 65 The regression results with respect to each of the six mortality indicators (i.e., Mortality1,..., Mortality6) are presented in Table 27, Appendix 11.Footnote 66 Neither the difference estimates (for any of the two samples) nor any of the DID estimates (which accounts for biases from time-invariant differences and year trends) are statistically significant at conventional levels. These results suggest that the inheritance shock has no detectable effect on mortality within any year over a period of 6 years after it occurs.Footnote 67