University of St. Gallen. Swiss Institute for Empirical Research, SEW-HSG. Varnbüelstrasse 14, 9000 St. Gallen

Transcription

1 University of St. Gallen Swiss Institute for Empirical Research, SEW-HSG Varnbüelstrasse 14, 9000 St. Gallen Increasing risk aversion: the consequences of the financial crisis for retirement behaviour in Switzerland February 2016 Monika Bütler Alma Ramsden

2 Abstract We analyse the effects of the latest financial and economic crisis on individual retirement behaviour using administrative data from a large Swiss pension provider. We find that the individual propensity to annuitize pension wealth increased after the crisis. The effect is predominantly driven by individuals who had worked, prior to retirement, in sectors that were more exposed to the crisis: a difference-in-differences estimation shows that the increase in the annuity rate of individuals in export-oriented sectors was around 4% larger, on average, than the annuity rate of individuals in non-export oriented sectors. Our results suggest that the crisis had an impact on individual risk aversion and that, as a consequence, individuals that were more exposed to the crisis value the longevity insurance more, on average.

3 Jel Classification: D81, D90, J26, G01 Keywords: Annuities, Risk Aversion, Financial Crisis, Switzerland

4 Contents 0.1 Introduction Related Literature Institutional Setting The Swiss Pension System Switzerland during and after the crisis Methodology Data and Descriptive Statistics Results Conclusion Tables and Figures

5 0.1 Introduction The latest financial and economic crisis had a significant impact on economies and societies globally. While the large economic downturn had a direct impact on asset positions and employment prospects, it might have also affected individuals behaviour in a more indirect manner: an exposure to a deep recession might change individuals retirement behaviour even in the absence of direct consequences because it makes them more risk averse. Risk aversion is one of the key elements in the decision to annuitize. Annuities insure against the financial consequences of longevity, i.e., the risk of outliving one s assets. The more risk averse an individual, the more valuable is the longevity insurance of an annuity (J. Brown, 2001). A number of papers have shown that macroeconomic shocks affect individual risk aversion (or perception of risk-taking behaviour) in future decisions. If a crisis pushes up risk aversion, annuity rates can be expected to go up, in particular among individuals who are more exposed to the crisis. We study this indirect impact by analysing whether the latest financial and economic crisis has affected individual annuitization decisions. In Switzerland, individuals on the brink of retirement are particularly interesting to study indirect effects of macroeconomic downturns as their direct exposure to them is low: stringent regulation not only protects accumulated pension wealth, but also prescribes the applicable annuity factor. Moreover, the choice between an annuity and a lump sum is one of the most important economic decisions in life. It usually involves a large sum of money (300,000 CHF, or about US$ 300,000 on average) and is largely irreversible. We use administrative data from a large Swiss insurance company which caters to small and medium-sized companies. We observe the actual retirement decision of 11,475 individuals between the years 2007 and We explore the effects of the crisis on risk aversion by analysing whether the crisis changed the individual propensity to annuitize pension wealth, in particular among those who had been working in the export sector prior to retirement: export-oriented sectors were much more affected by the crisis than other sectors due to their dependence on global economic developments. We find that annuity rates increased just after the crisis - this effect is driven solely by individuals in the export-oriented sectors. A difference-in-differences model uncovers the degree by which the annuity rates of individuals in export-oriented sectors (the treated ) increased compared to individuals in non-export oriented sectors (the non-treated ). The average increase in the annuity rate in the export-oriented sector was over 4% larger than in the non-export oriented sectors. By analyzing how the latest financial and economic crisis affected cash-out decisions at retirement, our paper contributes to the large literature on annuity demand. Moreover, it adds another aspect to the growing literature which explores the degree by which risk preferences and, as a consequence, individual choices are influenced by macroeconomic shocks. This is one of the first papers to explore how the latest financial and economic crisis has affected the elderly population, and annuitization behaviour, in particular. Understanding the demand for annuities and its relationship to macroeonomic events is relevant for policy makers and the design of employer-provided pensions. The paper proceeds as follows: Section 0.2 summarizes the related literature. Chapter 0.3 gives an overview on the institutional setting: the Swiss Pension System and Switzerland during the crisis. Section 0.4 describes the methodology and Section 0.5 gives an overview on the dataset and descriptive statistics. Section 0.6 presents and discusses the results. Section 0.7 summarizes and concludes. 2

6 0.2 Related Literature In Switzerland, like in all other countries, the demand for annuities is much lower than predicted by theoretical models such as the seminal work by Yaari (1965). This has caused a surge in the economic literature trying to explain the demand for annuities, most of which is theoretical. Among the most prominent explanations for the low demand for annuities resulting from this literature are bequest motives (J. Brown, 2001; Lockwood, 2012), uncertain medical expenditures (Peijnenburg et al., 2014), health (Sinclair and Smetters, 2004), annuity prices (Chalmers and Reuter, 2012), means-tested government benefits (Bütler et al., 2011) and insurance through marriage (J. Brown, 2001; J. R. Brown and Poterba, 2000). More recent research analyses the the demand for annuities empirically. Inkmann et al. (2007), for example, study the demand for cash-out decisions for private annuities (which are usually on top of employer-based annuities) using micro data from the English Longitudinal Study of Ageing (ELSA). 1 Only 6% of the respondents choose the annuity, a rate much lower than the annuitization propensity in our sample. Inkmann et al. (2007) find that education, life expectancy and wealth are significant predictors of the choice to annuitize whereas (mandated) pension income and bequest motives decrease annuity demand. However, private annuity decisions are very different from annuitization decisions in employer-provided, mandatory pension plans. The latter usually constitutes a large, if not the largest, part of income in old age while the former are mostly top-ups. Direct comparison between private and public annuitization decisions is thus difficult. Hagen (2015) investigates the demand for annuities in mandated plans. He uses data from Sweden s second largest pension fund together with administrative tax data. Individuals can either annuitize or withdraw their pension wealth as a fixed-term payout over 5, 10, 15, or 20 years. Hagen (2015) finds that people with little pension wealth are more likely to withdraw their wealth as a fixed-term payout than to annuitize, while payout choices are fairly constant across the income distribution (except for people at the very top of the income distribution). In line with previous research, the price of an annuity, induced by differences in taxation, is a significant predictor of the choice to annuitize. Hagen (2015) also provides evidence that individuals form expectations about their longevity and choose payout options accordingly. Demand for annuities needs not be stable over time. If there is correlation between macroeconomic events and risk aversion, or risk-taking behaviour, individuals that are directly or indirectly affected by a crisis should value the insurance provided by an annuity relatively more and consequently annuitize a larger share of their pension wealth. In their seminal paper, Malmendier and Nagel (2011) investigate whether individual experiences of macroeconomic shocks affect risk taking. The impact of macroeconomic events on individual risk preferences are analysed by means of the Survey of Consumer Finances from 1964 to The dataset includes individuals who experienced the 1930s Great Depression as teenagers or adults. Risk preference is measured by survey responses to questions about willingness to take risk, stock market participation, proportion of liquid assets invested by individuals in stocks or mutual funds and proportion of liquid assets other than stocks invested by individuals in bonds. Malmendier and Nagel (2011) find that individuals who experienced high stock market returns are less risk averse, more likely to participate in the stock market and invest more into risky assets, while individuals who experienced high inflation invest fewer of their non-stock liquid assets in bonds. They also find that respondents are somewhat more influenced by more recent events than by more distant events. Guiso et al. (2013) find that quantitative measures of risk aversion (the certainty equivalent required by an investor to give up the risky prospect) as well as qualitative measures of risk 1 Defined in the interview as follows: Annuity income is when you make a lump sum payment to a financial institution and in return they give you a regular income for the rest of your life (Inkmann et al., 2007). 3

7 aversion (individuals willingness to trade-off risk and return) have increased during the most recent financial and economic crisis. The analysis is based on a survey of a large Italian bank. Risk aversion did not only increase among individuals who experienced losses due to the crisis, but among all respondents. To test the hypothesis that respondents were not financially, but rather emotionally affected by the crisis, the authors use an experiment in which risk aversion is analysed after subjects have watched a scary movie (i.e. a movie which induces fear). Guiso et al. (2013) find that the behaviour of investors is indeed consistent with a fear-based model, emotions thus are as important a determinant of risk aversion as standard factors such as wealth. The finding by Guiso et al. (2013) that subjective willingness to take risk is lower in times of recession is confirmed in a recent paper by Cohn et al. (2015). Cohn et al. (2015) conduct a controlled experiment in which the willingness to take financial risk is measured. Participants are financial professionals who are randomly exposed to either a boom or a bust scenario and consequently asked to decide how much to invest in a risky asset with a positive expected return, and how much to invest in a risk-free account with a zero interest rate. Their results show that on average, subjects invest less in a risky asset in the bust condition than in the boom condition. Cohn et al. (2015) then expose participants to different levels of fear and find that respondents exposed to low levels of fear take higher risks, on average, than respondents exposed to high levels of fear. Weber et al. (2012) surveyed online brokerage customers from the UK in 3-month intervals between 2008 and Individuals were questioned about their risk expectations, risk attitudes and hypothetical risk taking behaviour. Investments into risky assets dropped in their sample between September 2008 and March 2009, but increased again in June At the same time, self reported risk attitudes remained fairly stable. The authors conclude that that changes in risk taking during the financial crisis are the result of changes in subjective feelings about market risk rather than changes in individual risk attitude. Taken together, the literature provides strong evidence for effects of macroeconomic shocks on risk aversion. However, little is known how macroeconomic shocks, and latest financial and economic crisis in particular, affected the elderly population. Chai et al. (2011) develop a life-cycle model to explore how elderly households adjust consumption and saving patterns, annuitization plans, work hours, and retirement behaviour in response to macroeconomic shocks. Their model predicts that, in the short run, individuals alter their asset allocation by reducing their equity exposure in favor of risk-free bonds. Consumption is reduced, work hours are expanded and retirement postponed due to the crisis. In the long run, the demand for private annuities decreases as investors invest more into liquid assets. Most of the predictions by Chai et al. (2011) are confirmed empirically by Hurd and Rohwedder (2010) who analyse the effects of the financial crisis on the population aged 55 and above using longitudinal data from the Health and Retirement Study: individuals near retirement work longer than before the crisis due to their decrease in assets. The latter is driven by a fall in house prices, also causing an increase in housing debt. Hurd and Rohwedder (2010) further find that elderly households reduce spending and anticipate to bequeath less, on average, due to these financial losses and also because of increased pessimism. In sum, little is known about how annuity rates in employer-provided pension schemes are affected by macroeconomic downturns. To our knowledge, there is no paper which analyses empirically the effects of recessions on individual annuitization decisions in employer-provided pension schemes. We contribute to the literature by empirically analysing the effects of the latest financial and economic crisis on individual risk aversion reflected by annuitization choices. 4

8 0.3 Institutional Setting The Swiss Pension System The Swiss pension system consists of three pillars. The first and the second pillar are the core of Switzerland s three pillar pension system. The first pillar is a pay-as-you-go, universal system which aims to provide a subsistence level of income to all retirees. Benefits depend on the number of years contributed and, to a limited degree, on average earned income. For retirees whose total income in old age is not enough to cover basic needs, means-tested supplemental benefits are paid in addition. 2 The third pillar constitutes a voluntary, private pension plan. The statutory retirement age is 65 for men and 64 for women. The second pillar, which is the focus of our analysis, is an occupational, fully funded pension scheme which is mandatory for all employees whose annual income exceeds a pre-defined threshold (CHF 24,675 in 2015). Its goal is to maintain pre-retirement living standards. An employer can choose from different organizational structures for his occupational pension plan. These range from setting up a completely autonomous pension fund to outsourcing the scheme entirely to an insurance company. The latter is relatively common, particularly for small and medium sized companies. Most pension plans are based on defined contributions. Upon retiring, workers may withdraw the accumulated retirement capital as a monthly lifelong annuity, a lump sum, or a mix of the two options. People who receive full disability insurance cannot opt for the lump sum, and at most a partial lump sum if they claim partial disability insurance. Annuities are strictly proportional to accumulated retirement assets: the pension wealth W is translated into a yearly nominal annuity A using conversion rate γ, hence A = γw. The conversion rate depends on the age of retirement but not on marital status although an annuity also includes survival benefits for spouses and child benefits for dependent children. The law stipulates a minimum conversion rate for the mandatory part, which is currently 6.8%. 3 In the super-mandatory part, pension insurers are free to set conversion rates. Nonetheless, conversion rates have stayed remarkably stable even in this less regulated fraction of pension wealth. This is reflected by figure 1 which shows the average conversion rates for both the mandatory and the super-mandatory part from 2002 to 2014: 2 By guaranteeing a minimum income in retirement, means-tested benefits create incentives to cash out second pillar pension wealth. Bütler et al. (2013) find that means-tested benefits are indeed associated with a decrease in demand for annuities, especially for individuals at the lower end of the wealth-distribution. 3 The amount of insured income above the lower threshold (CHF 24,675 in 2015) and below the upper threshold (CHF 84,600 in 2015) is called the mandatory component, and income above the upper threshold is called the super-mandatory component. All pension providers are required by law to insure the mandatory share. They are free to offer insurance for the super-mandatory share, however most do because the second pillar is considered an integral part in attracting well-educated workers in Switzerland s tight labour market. 5

9 Figure 1: Average conversion rates (mandatory and super-mandatory part) for leading Swiss pension funds, Source: Swisscanto (2014) The accumulation and decumulation phase of occupational pensions are organised by the same provider. It is possible to withdraw the accumulated balances to buy an annuity in the unregulated market but such a strategy would never be optimal as the conversion rates in unregulated markets are well below conversion rates in the regulated second pillar Switzerland during and after the crisis Although not as severely affected by the latest financial and economic crisis as other countries, Switzerland did not remain untroubled by its events. Global downturns such as the European debt crisis have a direct impact on a small open economy like Switzerland s. Moreover the Swiss Frank became a safe haven during the crisis and consequently appreciated rapidly. The export sector and tourism industry suffered the most from this strengthening of currency. As a consequence, to aid these industries, the Swiss National Bank introduced a minimum exchange rate to the Euro in September According to the president of the Swiss National Bank, the introduction of the minimum exchange rate of CHF 1.20 against the euro represented a drastic measure against an excessive strengthening of the Swiss franc that was threatening price stability and the Swiss economy (Jordan, 2013). Switzerland experienced a recession right after the collapse of Lehman brothers in late 2008 (Indergand et al., 2013; Feuz, 2012) of which the export-oriented industries suffered the most (BFS, 2012b). In 2009, Switzerland experienced a decline in GDP of 1.9%, its greatest decline in GDP since the 1970s oil crisis (Bundesamt für Statistik, 2011). About 60% of Swiss firms indicated being affected by the aftermath of the crisis, with up to half of all export-oriented firms indicating being severly affected by the crisis. As of 2010, the export industry started to recover from its dip. Yet, despite its recovery, it suffered major setbacks in the years following the crisis and did not get back to its prior levels (see Straumann, 2013, for a discussion) Figure 2 shows the evolution of the change in the value of exports and imports between 2007 and 2013: 6

10 Figure 2: Change in value of exports and imports, 2007 to 2013 Notes: Own calculations based on data from the Swiss Federal Customs Authority. Individuals working in export-oriented sectors were clearly more exposed to the adverse effects of the crisis. Many firms in the machine building, electrical and metal industry reduced working hours in the first years of the crisis to compensate financial losses (SECO, 2013). The number of firms receiving government compensation for reduced working hours rose from 815 in 2008 to 8,870 in 2009, corresponding to an increase in the number of persons concerned from 2,086 to 67,398 (SECO, 2013). While unemployment rates increased in all sectors after the crisis, the export-oriented sectors were disproportionally affected. Figure 3 shows the evolution of number of unemployed for six export-oriented industries. Among the highest revenue generating export-oriented industries in Switzerland are the chemical production and pharmaceuticals (generating 85 Billion Swiss Francs and 41% of all exports in 2014), the production of precision instruments, watches and jewellery (generating 47 Billion Swiss Francs and 22.6% of all exports in 2014) and the production of machines, appliances and electronics (generating 33 Billion Swiss Francs and 16% of all exports in 2014) (EZV, 2014). The food, textile and production of synthetic materials industries are also largely export-oriented. All these industries experienced an increase in number of unemployed persons after the onset of the crisis and in some of these industries, unemployment rates remained high for many years after the crisis. 7

11 Figure 3: Evolution of number of unemployed in 6 export-oriented industries, Notes: Own calculations based on data from the Swiss Federal Statistics Office. Y-scales differ. Red line denotes collapse of Lehman Brothers. As in other countries, the crisis led to a sharp decrease in interest rates for both short-term and long-term interest rates (see figure 4). This is of particular importance in this setting since interest rates affect the decision to annuitize: for a given annuity conversion rate, a decrease in interest rates should increase the demand for annuities because lower interest rates increase the money s worth of an annuity and make alternative investments of pension wealth less attractive. With our methodological approach - a difference-in-differences estimator - we are able to disentangle risk aversion effects from interest rate effects, as explained in the next section. Figure 4: Evolution of interest rates of Swiss bonds with different maturities, 2007 to Note: Own calculations based on data from the Swiss National Bank. Red line denotes collapse of Lehman Brothers. 8

12 0.4 Methodology We define the collapse of Lehman brothers in September 2008 as the start of the crisis in Switzerland to analyse the effects of the crisis on retirement behaviour. The use of this date is motivated by the fact that the collapse of Lehman brothers not only marked the beginning of large-scale public awareness of the crisis but was also the onset of the recession in Switzerland (see Section 0.3.2). As outlined in chapter 0.3.2, export-oriented sectors were more severely affected by the crisis than other sectors. As a consequence, we expect individuals working in the export sector to be much more exposed to the impacts of the crisis. While accumulated pension wealth remained unchanged due to stringent regulation (no direct impact), exposure to the crisis through, e.g., high unemployment rates in those sectors might affect individual risk aversion and thus the propensity to annuitize (indirect impact). 4 In a first step, we analyse the effect of the crisis on the elderly population using the full sample in our dataset. We implement a simple before-after estimator to investigate annuity rates before and after the crisis, i.e. the mean difference between treatment (after) and control (before) annurate = Y 1 Y 0 : Y t = β 0 + β 1 P ost t + β 2 X t + ξ t (1) where Y t is the annuity rate in time t, P ost is a dummy for the onset of the crisis, X t are individual-specific control variables and ξ t is the error term. The outcome variable is defined as the ratio of pension wealth withdrawn as an annuity to total pension wealth and thus corresponds to a number between 0 and 1 (with bunching at 0 and 1): Y = Amount of pension wealth withdrawn as annuity Total pension wealth The drawback of the before-after estimator is that it attributes all differences in outcomes relative to pre-treatment level to the analysed treatment (Heckman and Smith, 1999). This can be mitigated to a certain degree as we can control for economy-wide differences such as unemployment rates (available on regional level 5 ) and interest rates in the regression. In a second step, we investigate whether there are differential effects of the crisis on the annuity rate of the elderly according to their exposure to the crisis. In this difference-in-differences model, the treatment group consists of individuals who had worked in the export-oriented sectors prior to retirement, while the control group consists of individuals working in the non-export oriented sectors at the time of retirement. We define the export sector as the sector which comprises the following industries: the textile and food industry, the chemical production and pharmaceuticals industry, the synthetic materials production industry and the electronic goods (2) 4 We rule out the potential channel of a reduction of trust in institutions. A decrease in trust in institutions caused by the crisis might make people believe that they are better off withdrawing the lump sum. The only empirical evidence on this issue (to our knowledge) is provided by Roth et al. (2013). Roth et al. (2013) find that trust in institutions decreased only moderately during the crisis among EU countries and is driven mainly by the decline in trust in 4 countries (Spain, Greece, Portugal and Ireland). They also find no decline in trust in the national government or national parliament in European countries. Swiss data on federal votes which contains variables on trust in institutions reveals that average trust in institutions is actually statistically significantly larger in the years after the crisis than in the years before the crisis. Lastly, pensions funds in Switzerland recovered quickly from their poor performance in 2008 and performed relatively well in the years following the crisis (Swisscanto, 2014). Thus, although a priori an effect of the crisis on annuitization rates through this second channel cannot be ruled out, it seems very unlikely that the crisis had significant impacts on people s trust in institutions. 5 These regions are the Italian part, the area around Lake Geneva (the French part), Zürich, North-Western Switzerland, Central Switzerland, Eastern Switzerland and the Midland. 9

13 and electronic equipment production industry (see also Section for an overview). Detailed information on the sectors in which the individuals worked prior to retirement allows us to unambiguously assign individuals to treatment and control groups. Since the occurrence of the crisis is a truly exogenous shock, selection into treatment is not an issue in this setting. Of course, selection into work industry is not completely exogenous however the difference-in-differences estimation takes care of such differences between groups because they are time-constant. A more crucial issue is whether the common trend assumption (i.e., the assumption that annuity rates of treatment and conrol would have followed a similar trend in the absence of the crisis) is fulfilled in this setting: in section 0.6 we describe some approaches to investigate the common trend assumption. Unfortunately, due to data limitations, the options to do so remain limited. The difference-in-differences estimator using cross-sectional data can be written as follows: Y t = α + δt reated t + λp ost t + γ 1 (T reated P ost) t + φx t + η t (3) where T reated t is a dummy for the treatment group (having worked in the export sector prior to retirement), P ost t = 1 if the individual retires after the onset of the crisis and 0 otherwise, (T reated P ost) t is the interaction term i.e. the Difference-in-differences coefficient of interest and η t is the error term. We control for covariates X t in the regression: age, gender, marital status, wealth, income and the present value of the annuity. As long as the differences in covariates between treatment and control group are constant over time, they will be captured by the difference-in-differences estimator and are not problematic as long as there is strict exogeneity (no influence by treatment) (Lechner, 2011). However, there is conclusive evidence in the literature for a strong correlation between individual characteristics and the decision to annuitize (see Bütler and Teppa, 2007; Chalmers and Reuter, 2012; J. Brown, 2001, e.g.) hence, including covariates may increase the precision of the estimator. For similar reasons, we control for interest rates in our estimations as the sharp decrease in interest rates in the aftermath of the crisis made the annuity more valuable relative to the lump sum. The dummies P ost t in both equations require further explanation: until the end of 2010, the insurance company which provided the data made individuals choose between the annuity and the lump sum 12 months prior to retirement. Consequently, decisions made in October 2008, for example, are observed in the data in October Hence, the effect of the start of the crisis is observed twelve months after the collapse of Lehman brothers. The cut-off date thus needs to be shifted from October 2008 to October As of 2011, decisions made by individuals are immediately observable in the data as they reflect choices made when they enter retirement. 0.5 Data and Descriptive Statistics We use administrative records from a large Swiss insurance company which provides pension plans to small- and medium-sized companies in the private sector. Small- and medium-sized companies constitute 99 percent of all companies in Switzerland and account for about two thirds of employment in Switzerland (BFS, 2012a). 6 The data contains information on retirement choice (annuity, lump sum, or a combination of the two), from which we derive the outcome variable, annurate. It also contains the following individual characteristics: date of birth (birth), age at retirement (age), date of retirement (date), gender (sex), marital status (married), total pension wealth at time of retirement (wealth), income in the year before retirement (income), and whether the individual receives a disability insurance (disability). We further know in which sector the individual worked before retirement (noga, defined by so-called Noga codes). The latter information is used to construct the treatment and control groups. We exclude individuals 6 The formal definition of a small- or medium-sized company provided by the Swiss federal statistical office is that it should not contain more than 249 employees. 10

14 who receive full disability insurance because they are severely restricted in their choice to take the lump sum (see Section 0.3.1; descriptives statistics for the full sample are in the Appendix in Table 8). Table 1 provides summary statistics for the observations which are used for the analysis: Table 1: Descriptive statistics (I) Variables N mean s.d. min max Covariates Sex 11, Married 11, Age at retirement 11, Wealth 11, , , ,428, Income 11,475 78,793 61, ,085, Annuity after retirement 11,475 9,207 14, , Export 11, PV 11, Outcome variable Annuity rate 11, Notes: Export=1 if individual works in export-oriented sector, 0 otherwise. Sex=1 if female, 0 if male. Married=1 if married, 0 otherwise (divorced, widowed, single). Wealth and income are in Swiss Francs, income refers to last income in the year before retirement. Individuals that receive full disability insurance are excluded. Our sample is fairly representative of the Swiss population and corresponds closely to other papers which have used data from Swiss pension providers: the average annuity rate in this sample over all years is This is almost equivalent to in Bütler et al. (2013), also using data from stand-alone Swiss pension funds. Average age at retirement is only slightly higher than in Bütler et al. (2013) and Bütler and Teppa (2007), where the average age is 63.9 years and 61.8 years, respectively. The pension balance corresponds closely to the national average which is a little less than CHF 300,000. The share of women in this dataset is lower than the share of men and corresponds roughly to the national average of second pillar recipients, which is 0.41 (BFS, 2013). For every individual in the dataset we calculate the present value of an annual income stream of 1 after retirement (P V ). This variable captures changes in the yield curve which represent investment opportunities if the lump sum is taken (see also J. Brown (2001) for a discussion on present value calculations). We calculate this present value for a 65 year old man until the end of his life. We use nominal yields on Swiss treasury bonds with maturities of 1, 2, 3, 4, 5, 6, 7, 8, 10, 20 and 30 years to calculate the expected nominal short rate in each future period (see figure 4). Life expectancy is calculated using data from mortality tables created by the Swiss Federal Statistics Office (BFS). P V reveals the importance of interest rates in individual annuity demand: the lower the interest rate, the higher the value of the annuity relative to the lump sum. With a stable conversion factor, this should lead to a higher annuity rate. Table 10 in the Appendix shows that interest rates on Swiss bonds are always significantly negative related to the annuity rate. It is therefore crucial to include P V in our estimations. Table 2 shows that individual characteristics before and after the crisis for treated and control are fairly similar. The crisis has not led to large changes in wealth or income among retirees in the sample, nor to early or delayed retirement: average age and income are almost the same in all years, average wealth increased slightly however not during the years affected most by the crisis. 11

15 This confirms that, contrary to other countries, individuals on the brink of retirement were not directly affected by the crisis in Switzerland - mainly due to the regulatory environment. Figure 6 in the Appendix shows averages of second pillar pension wealth, income in the year before retirement and age at retirement over time. Results from t-tests show that these variables are not statistically significalty different over time. 12

16 Table 2: Descriptive statistics (II) Before Crisis After Crisis Treated Control Treated Control Mean Min Max Mean Min Max Mean Min Max Mean Min Max 13 Married (0.47) (0.45) (0.47) (0.45) Age (1.58) (1.23) (1.75) (1.46) Sex (0.49) (0.43) (0.49) (0.42) Employment (17.99) (12.76) (22.34) (17.3) Wealth 275, ,267, , ,209, , ,428, , ,486,000 (330,271) (190,052) (413,132) (231,916) Income 76, ,600 76, ,941 79, ,085,000 79, ,200 (58,679) (39,868) (67,439) (49,306) N 1, ,544 2,654 Notes: Standard deviations in parenthesis. Significant differences are denoted by *( 10%), **( 5%), ***( 1%). Sex=1 if female, 0 if male; married=1 if married, 0 otherwise (divorced, widowed, single). Employment denotes employment in % before retirement. Income is annual income in the year before retirement, age is age at time of retirement.

17 Figure 5 shows the evolution of the annuity rate between 2007 and 2015 for the treatment and the control group. Recall that the average annuity rate prior to 2011 reflects decisions made 12 months earlier whereas the average annuity rate as of 2011 reflects decisions made at the time of entry into retirement. Decisions made at the end of 2008 and in early 2009 are thus visible in the graph as of 2010, where indeed we see an increase in the annuity rate for both treated and control - this increase is steeper for the treatment group than for the control group. For the treatment group, there is a 15 percentage point increase in the annuity rate in the first half of The annuity rate keeps increasing after the crisis and at a steeper rate for the treatment group than for the control group. Overall, this shows that the annuity rate increased in the years following the crisis, and more so for the individuals working in the export-oriented sector than for other individuals. Figure 5: Annual annuitization rate, treated and control (excluding individuals receiving full disability insurance) 0.6 Results Table 3 shows the difference in the average annuity rate before and after the crisis. The difference is positive and significant for the full-sample (column (I)). The effect of the crisis on the annuity rate becomes insignificant when the variable P V is included: interest rates capture most of the impact of the crisis on the annuity rate. Splitting the sample by export- and non-export sector reveals that the difference in the annuity rate is only significant for the export sector (column (V) vs. column (III)): hence, the positive effect observed in the full sample seems to be driven by this group alone. As before, the effect becomes insignificant for both groups when including the present value of the annuity as an explanatory variable. Results in table 3 confirm previous findings of a strong correlation between individual characteristics and the decision to annuitize (see also table 9 in the Appendix). While there is no significant effect of being married on taking the annuity per se, married women are significantly more likely to take the lump sum than married men and singles. This result has been found in previous studies and can be explained by an implicit insurance of married women through 14

18 Table 3: Effect of the crisis on the annuity rate: before-after estimator 15 Dependent Variable: Annuity Rate Full sample Non-export Export (I) (II) (III) (IV) (V) (VI) Post *** *** *** *** (0.0111) (0.0175) (0.0133) (0.0209) (0.0199) (0.0318) Sex *** *** 0.108*** 0.107*** (0.0163) (0.0162) (0.0190) (0.0189) (0.0330) (0.0330) Married (0.0126) (0.0126) (0.0155) (0.0155) (0.0214) (0.0214) Female married *** *** *** *** *** *** (0.0198) (0.0198) (0.0230) (0.0230) (0.0404) (0.0404) Wealth 1.63e-07*** 1.61e-07*** 1.35e-07*** 1.33e-07*** 4.11e-07*** 4.12e-07*** (1.69e-08) (1.69e-08) (1.80e-08) (1.80e-08) (5.07e-08) (5.06e-08) Income -4.19e-07*** -4.25e-07*** -2.78e-07** -2.84e-07** -1.22e-06*** -1.23e-06*** (9.98e-08) (9.97e-08) (1.11e-07) (1.11e-07) (2.37e-07) (2.36e-07) Age 0.691*** 0.725*** 0.721*** 0.754*** 0.396* 0.438* (0.0951) (0.0952) (0.105) (0.105) (0.229) (0.229) Age sq *** *** *** *** * * ( ) ( ) ( ) ( ) ( ) ( ) PV *** *** *** ( ) ( ) ( ) Constant *** *** *** *** * ** (3.039) (3.045) (3.362) (3.371) (7.297) (7.304) Observations 11,475 11,475 8,135 8,135 3,340 3,340 R-squared Notes: Standard errors in parenthesis. Level of significance is denoted by *( 10%), **( 5%), ***( 1%). P ost = Before-after estimate. Age refers to age at retirement, P V =expected discounted value of an annuity after retirement calculated for a man at age 65, from time of retirement until expected end of life (calculated with survival probabilites). Sex=1 if female, 0 if male. Married=1 if married, zero otherwise.

19 the annuity of their husbands. Their pension wealth (which, for this generation of retirees, is usually lower than their husbands ) is cashed-out to be used as spending money or to pay off a mortgage. Moreover, cashing-out married women s pension savings is a way to split the tax bill and get around tax progressivity.older retirees choose the annuity more often, on average, but this effect is decreasing with increasing age as indicated by the positive and highly significant coefficient on age and the negative and significant coefficient on age sq. The coefficients for income from the last year before retirement and pension wealth are highly significant but have opposite signs: income is negatively related to the annuity rate whereas wealth is positively related to the annuity rate. This is likely the result of non-linear income over the life-time: individuals that have accumulated a large pension wealth reduce their income - e.g. through reduced working hours - at the end of their lifetime. Table 4: Effects of crisis on annuitization decision Outcome variable is annuity rate (I) (II) (III) (IV) Did estimate *** *** *** *** (0.0124) (0.0125) (0.0124) (0.0139) Treated * (0.0161) (0.0157) (0.0160) (0.0154) Post *** (0.0124) (0.0124) (0.0158) (0.0144) Sex (0.0212) (0.0214) (0.0207) Married *** *** *** (0.0169) (0.0172) (0.0189) Wealth 1.68e-07*** 1.66e-07*** 1.65e-07*** (5.17e-08) (5.13e-08) (5.07e-08) Income -3.98e-07** -4.05e-07** -4.53e-07** (1.74e-07) (1.72e-07) (1.86e-07) Age at retirement 0.698*** 0.733*** 0.635*** (0.107) (0.109) (0.0948) Age squared *** *** *** ( ) ( ) ( ) PV *** ( ) Unempl. Rate ( ) Constant 0.427*** *** *** *** (0.0161) (3.489) (3.580) (3.119) Observations 11,475 11,475 11,475 11,475 R-squared Notes: Heteroskedasticity-robust standard errors in parenthesis. Level of significance is denoted by *( 10%), **( 5%), ***( 1%). T reated = 1 if individuals work in export sector, 0 otherwise. Age refers to age at retirement, P V = expected discounted value of an annuity after retirement calculated for a man at age 65, from time of retirement until expected end of life (calculated with survival probabilites). Married=1 if married, zero otherwise. The difference-in-differences model is estimated for different specifications: without covari- 16

20 ates (I), with covariates but without including P V, which captures the interest rate structure (II), including P V (III) and controlling for unemployment (IV) (see table 4). Results show that, on average, the annuity rate was between 4 and 5 percentage points higher in the treatment group than in the control group after the financial crisis. Given a mean annuity rate of about 0.45 (or 45%), this effect is sizeable. Controlling for the present value, covariates and the unemployment rate does not change the results: the difference-in-differences estimator remains similar in size and highly significant. The distribution of the outcome variable has two mass points at at zero and one as a large fraction of individuals in the sample chooses either only the lump sum or only the annuity. The resulting loss in efficiency can be taken care of by computing heteroskedasticity-robust standard errors. Results of a Breusch-Pagan test for heteroskedasticity support the use of heteroskedasticity-robust standard errors. We perform several robustness checks. First, we use as an outcome variable a binary indicator for choosing the annuity or the lump sum (i.e. excluding the choice of the combination ). This leaves the results virtually unchanged (see table 5). Second, we re-define the cut-off date to be October 2008: because of the announcement period of 6 months, we should not observe any effect right after the actual start of the crisis. Columns (I) to (IV) in table 6 show that this is indeed the case, the coefficient is negative and insignificant. We also perform a placebo difference-in-differences using only the year 2007 and 2009 and defining January 2008 as the cut-off date. Again the effect is not significant (see columns (V) to (VIII)). In a last step, we investigate some of the industries which make up the export sector separately: we define the textile industry, the machines production, electronics production, chemical production, synthetics production, electronic equipment production industries and pharmaceutical industry as treated and leave the control group unchanged (i.e., all non-export industries). Analysis of other industries is difficult due to insufficient data. The difference-in-differences estimate is positive and significant for all industries but production of electronic equipment and the pharmaceutical industry (see table 7). The former is the only sector which experienced a steady decline in unemployment rates even during the crisis (see Figure 3) hence people working in this sector were in that regard less exposed to the crisis. The latter recovered very quickly from its dip in 2009, and was the main driver of export growth just after the crisis due to its focus on research and its geographic diversification. On the contrast, export-oriented industries that rely heavily on labour as input factor and cater mainly to European countries were hit much harder (CS, 2014). 17

21 Table 5: Effects of crisis on annuitization decision Outcome variable is binary choice annuity or capital (excl. mixed option) (I) (II) (III) (IV) Did estimate *** *** *** (0.0260) (0.0128) (0.0127) (0.0144) Treated *** ** ** (0.0232) (0.0163) (0.0166) (0.0157) Post ** *** (0.0177) (0.0128) (0.0143) (0.0153) Sex (0.0225) (0.0226) (0.0213) Married ** ** ** (0.0205) (0.0208) (0.0230) Wealth 1.85e-07*** 1.83e-07*** 1.82e-07*** (5.63e-08) (5.61e-08) (5.43e-08) Income -3.92e-07** -4.00e-07** -4.56e-07** (1.73e-07) (1.72e-07) (1.96e-07) Age at retirement 0.654*** 0.688*** 0.593*** (0.110) (0.112) (0.0986) Age squared *** *** *** ( ) ( ) ( ) PV *** ( ) Unempl. Rate * ( ) Constant 0.348*** *** *** *** (0.0192) (3.541) (3.612) (3.202) Observations 9,571 9,571 9,571 9,571 R-squared Notes: Heteroskedasticity-robust standard errors in parenthesis. Level of significance is denoted by *( 10%), **( 5%), ***( 1%). T reated = 1 if individuals work in export sector, 0 otherwise. Age refers to age at retirement, P V =expected discounted value of an annuity after retirement calculated for a man at age 65, from time of retirement until expected end of life (calculated with survival probabilites). Married=1 if married, zero otherwise. Number of observations is smaller in (IV) because unemployment rate not available for

22 Table 6: Effects of crisis on annuity rate, placebo difference-in-differences 19 Outcome variable is annuity rate Cut-off is October 2008* Cut-off is January 2008 I II III IV V VI VII VIII DiD estimate e (0.0273) (0.0271) (0.0270) (0.0286) (0.0398) (0.0396) (0.0397) (0.0412) Treated (0.0251) (0.0250) (0.0249) (0.0258) (0.0337) (0.0337) (0.0337) (0.0351) Post * *** (0.0151) (0.0150) (0.0206) (0.0160) (0.0217) (0.0217) (0.0256) (0.0225) Constant 0.420*** *** *** *** 0.415*** *** *** *** (0.0140) (3.050) (3.058) (3.636) (0.0184) (7.440) (7.444) (7.713) Observations 11,475 11,475 11,475 11,475 3,357 3,357 3,357 3,357 R-squared Covariates no yes yes yes no yes yes yes PV no no yes yes no no yes yes Unempl. rate no no no yes no no no yes Notes: Heteroskedasticity-robust standard errors in parenthesis. Level of significance is denoted by *( 10%), **( 5%), ***( 1%). T reated = 1 if individual work in the export-oriented sector, 0 otherwise. *This analysis ignores the announcement period.

23 Table 7: Effects of crisis on annuity rate, separate regression for 6 export-oriented industries 20 Outcome variable is annuity rate Export-oriented industries: Textiles Machines and electronics Chemicals Synthetics Electr. equipment Pharmaceuticals I II III IV V VI DiD estimate 0.744*** 0.125** 0.105** * (0.192) (0.0537) (0.0522) (0.0404) (0.0411) (0.136) Treated *** ** *** (0.0796) (0.0467) (0.0436) (0.0353) (0.0353) (0.113) Post (0.0214) (0.0210) (0.0211) (0.0209) (0.0208) (0.0214) Constant *** *** *** *** *** *** (0.273) (0.267) (0.268) (0.266) (0.262) (0.272) Observations 6,722 7,113 7,085 7,300 7,408 6,769 R-squared Covariates yes yes yes yes yes yes PV yes yes yes yes yes yes Notes: Standard errors in parenthesis. Level of significance is denoted by *( 10%), **( 5%), ***( 1%). T reated = 1 if individual works in one of the export-oriented sectors, 0 if it works in a non-export industry. P V =expected discounted value of an annuity after retirement calculated for a man at age 65, from time of retirement until expected end of life (calculated with survival probabilites). Treatment groups consist of different export-oriented industries, control group are non-export industries (excluding all export industries which are not the treatment group).

24 0.7 Conclusion This paper investigates how the latest financial and economic crisis affected the demand for annuities. As annuities provide insurance against the financial consequences of longevity, more risk averse individuals can be expected to annuitize a larger fraction of their pension wealth despite not being directly affected. Experiencing a crisis might have an impact on risk aversion, pushing up annuity rates. We investigate this by looking at individuals who had a greater exposure to the crisis impact, but whose assets were not directly affected. We use observed annuitization decisions of a sample of retirees obtained from a large Swiss insurance company. The advantage of our institutional setup is that all individuals face the same choice between different retirement options. For individuals on the brink of retirement, moreover, the crisis had no direct effect on their accumulated second pillar retirement wealth. If the retirement decisions of the different subgroups diverge after the crisis, most likely this is due to an indirect effect of the crisis - for example via an increase in risk aversion. A before-after and difference-in-differences analysis reveals that the average increase in the annuity rate after the economic downturn is predominantly driven by individuals working in the export-oriented sectors. For the latter, the average increase in the annuity rate was over 4% points larger than in the non-export oriented sectors. As in most other countries, Switzerland s export-oriented sectors suffered much more from the crisis than other sectors. Hence, individuals working in these sectors were more exposed to the consequences of the crisis than individuals in other sectors. The differential impact of the crisis on individual annuitization decisions by sector shows that macroeconomic events can have an effect on risk aversion or risk perception. This is even true for individuals who are not directly affected by the crisis, provided that exposure to the crisis is intense enough. This papers adds to the understanding of demand for annuities and its relationship to macroeonomic events, both of which is relevant for policy makers and the design of employer-provided pensions. 21

25 Appendix.1 Tables and Figures Table 8: Descriptive statistics for the full sample, including individuals who receive full disability insurance Variables N mean s.d. min max Covariates: Income 14,620 65,681 59, ,085,000 Sex 14, Disability 14, Wealth 14, , , ,824,000 Married 14, Age at retirement 14, PV 14, Outcome variable: Annuity rate 14, Notes: Sex=1 if female, 0 if male. Disability= receipt of Swiss disabiliy insurance, it ranges from 0 (no disability) to 100 (full disability). Last income denotes last annual income before retirement. 22

26 Table 9: OLS regression of annuity rate on individual characteristics 23 (I) (II) (III) (IV) Sex *** *** (0.0105) (0.0106) (0.0104) (0.0236) Married *** *** *** ** ( ) ( ) ( ) (0.0188) Log wealth 0.127*** 0.125*** 0.131*** ( ) ( ) (0.0101) Log income *** *** *** ( ) ( ) (0.0227) Age at retirement ** * ** ( ) ( ) ( ) ( ) PV *** *** *** *** ( ) ( ) ( ) ( ) Wealth 1.92e-07*** Income (1.79e-08) -4.81e-07*** (1.02e-07) Constant *** *** *** (0.201) (0.204) (0.190) (0.469) Observations 13,725 13,725 13,725 3,589 R-squared Notes: (I) excludes individuals with full disability insurance, (II)-(IV) exclude individuals with full and part disability insurance. (III) controls for level income and wealth. (IV) for individuals from the export-oriented sector only. Std. errors in parenthesis. Level of significance denoted by *( 10%), **( 5%), ***( 1%). Age=age at retirement, P V =present value of annuity calculated from time of retirement at age 65 until expected end of life (using survival probabilites). Sex=1 if female, 0 if male. Married=1 if married, zero otherwise.

27 Table 10: Regression of Annuity Rate on CH bonds with maturities of 1, 2, 3, 5, 10, 20, 30 years Outcome variable: annuity rate Years to maturity Interest rate *** *** *** *** *** *** *** ( ) ( ) ( ) ( ) ( ) ( ) ( ) Constant *** *** *** *** *** *** *** (0.217) (0.217) (0.217) (0.217) (0.217) (0.217) (0.217) Observations 9,399 9,399 9,399 9,399 9,399 9,399 9,399 R-squared Covariates Yes Yes Yes Yes Yes Yes Yes Notes: Years 2007 to Standard errors in parenthesis. Level of significance is denoted by *( 10%), **( 5%), ***( 1%). 24

28 Figure 6: Average wealth, income and age at retirement by year,