ESTIMATION PROCEDURES

The creation of a pension database has required gathering social insurance legislation for a significant sample of countries and a long time span. As for country cases, I have tried to cover classical categorizations of welfare traditions which may translate into significant differences in pension outcomes. Hence, selected countries include France, Germany, Italy, Portugal, Spain, Sweden, the United Kingdom, and the United States.Footnote ⁹

Pension calculation rules have been examined in these countries from the time when state pensions for dependent workers were first introduced until the present. Therefore, our database spans from 1889, when Bismarckian Germany introduced its pioneering system of old age pensions, until 2000. Pension legislation has been examined every ten years. For each of the periods and countries, information has been collected on the coverage of the system, top and bottom income thresholds, qualifying periods, calculation rules, maximum and minimum pensions, and other factors. Importantly, I have concentrated exclusively on public and general regime pensions. Although I will focus on pension outcomes and replacement rates in particular, other variables, particularly rules on coverage, help determine the generosity of pensions.Footnote ¹⁰

Pension calculation rules have been applied to workers with a given level of earnings and work history with the aim of providing a measure of the differential treatment of earnings by pension systems. Estimates have been developed for nine types of workers earning, respectively, 25, 50, 75, 100, 150, 200, 250, 300, and 400 percent of the average wage earnings before taxes.

Several assumptions have been made. First, earnings are flat during all working years, in a way that, for instance, a worker retiring in 1950 with earnings equaling the average level in the economy in that year is assumed to have had that nominal salary during his whole past work history.Footnote ¹¹ Second, I assume that workers provided labor without interruptions from the age of 20 to the normal retirement age scheduled in each pension act. Third, in order to stick to the analysis of benefits strictly resulting from the provision of work and not increasing due to any personal circumstance, pensions are calculated for the case of single men without dependents. Fourth, and most importantly, I assumed that a worker retiring in year i had been subject during his entire working life to pension calculation rules applicable in year i. This assumption is analogous to demographers’ use of “period rates,” as opposed to cohort rates, which have been used in calculating life expectancy or divorce rates. Therefore, the analysis will capture only a short-run snapshot of each year's pension policy impact.

My data add to two other sources on pension generosity of widespread use. One is the Walter Korpi and Joakim Palme's Social Citizenship Indicator Program (SCIP) database, which covers 18 countries and provides information on different programs for the years 1930, 1933, 1939, 1947, 1950, and thereafter every fifth year up to 2000. The SCIP measures the replacement rates corresponding to a single worker with the gross average industrial wage. The second source is Lyle Scruggs’ data set of Welfare State Entitlements. In this case, female employees’ are explicitly included in the calculation of the average wage. The data is offered yearly for 18 countries between 1971 and 2002. Table 1 compares these two sources against my own results.

Table 1

REPLACEMENT RATES OF PENSIONS WITH RESPECT TO EARNINGS, DIFFERENT SOURCES (workers with average level of earnings)

* Corresponds to 1939.

* Corresponds to 1961.

*** Corresponds to 1971.

Sources:

Scruggs, Welfare State Entitlements; Korpi and Palme, Social Citizenship; and author's own elaboration.

LONG-TERM TRENDS IN PENSION OUTCOMES

The history of old age insurance has translated into significant transformations of pension policies. This evolution is not only related to changing economic cycles in a century-long time span, but also to different welfare traditions. Esping-Andersen's classification of the worlds of welfare was an influential attempt to identify some of the most relevant differences and set welfare categories accordingly. His concept of the “world of welfare” was static, in essence, because it assumed that structural factors like culture or political tradition not only had determined the present shape of welfare regimes, but were to guide their future adaptations too.

However, this conceptual framework has been challenged in several ways. From the mid-1970s and up to the latest pension reforms resulting from the current financial crisis, most welfare states have experienced notable changes. Their extent is a matter of debate between studies suggesting that welfare states have entered a period of “permanent austerity” and those that interpret new policies as adaptations applied to well-rooted historical bases.Footnote ¹²

Independently, change has not only taken place in the public realm. In the last decades a new balance between public and private has emerged in welfare provision.Footnote ¹³ In this way, welfare protection has been partially brought back to the market, even though the impact of this process varies dramatically among particular programs. Some show a greater penetration of private arrangements, such as family spending, and programs like health spending are more inelastic to changes in the public-private mix.Footnote ¹⁴

As general as these trends might be, their effect on welfare states is still a function of the welfare tradition to which they belong. This applies especially in the case of globalization. While it is generally accepted that globalization has redrawn the borders of welfare typologies, the weight and sign of the changes caused by it remain subject to debate.

From a traditional point of view, focusing on the competitive pressure introduced by greater economic integration, public social spending may have shrunk as a result of a loss of independence of states to tax and finance public protection. Authors like Martin Rhodes, and Herman Schwartz argue that globalization has led countries to a retrenchment in social spending policies in order to maintain competitive advantages in the world economy.Footnote ¹⁵ A likely consequence of this is the greater prominence of private arrangements, thus narrowing the gap between welfare regimes with different degrees of state participation. On the other hand, a hypothesis based on the demand side for social spending argues that, in compensation for the undesired effects of greater openness, an expansion of social spending may follow.Footnote ¹⁶ Therefore, convergence in social provision would follow economic integration.

These two contradictory hypotheses may be reconciled after income levels are taken into account. Countries with minimal levels of protection would expand their welfare states to cope with higher economic volatility, while more developed countries with well-developed welfare state structures are bound to retrench.Footnote ¹⁷

In spite of these trends, the transmission of globalization risks is not uniform and only some industries and workers are affected by it. As the last wave of globalization has been skill-biased, earnings structures in developed countries have become more polarized.Footnote ¹⁸ As a result, the increasing threat posed by globalization on the lowest levels of earnings might have underpinned the desire to reinforce pension provision at the lowest levels, assuming that lower earnings translate into lower pensions as a consequence of less insurance contributions paid. Finally, the relationship between globalization and the welfare state has also been denied by some researchers, arguing that domestic and not external pressures are ultimately responsible for welfare state outcomes.Footnote ¹⁹ As a matter of fact, power resources, social coalitions, and the quality of political systems have been found to be critical to determining the sign of the effect of greater internationalization on social policy.Footnote ²⁰ The largest differences arise between democracies and autocracies, but the ability to mobilize power resources is a key element to counterbalance some of the worst effects of openness, as the case of developing countries dramatically shows.Footnote ²¹

Nevertheless, the demand for policy intervention from citizens seems to differ according to each country's previous record of trade openness. When faced with economic openness for the first time, citizens lack a priori knowledge of which firms/sectors will fail. Therefore, economies may tend to endogenize risk and provide for adequate social policy at a general level. After this, further increases in openness would have marginal effects on the demand for social protection.Footnote ²²

In my analysis, which is concentrated on a relatively homogeneous sample of developed countries, one may argue that the redefinition of the welfare state in the last years has been moderate because of the similarity between political regimes or already high levels of integration due to EU membership. Still, the intuition is that the boundaries of welfare state classifications have changed all the same. Therefore, the dynamic nature of welfare states must be stressed once again, as the following analysis will show.

DATA

Figures 1A and 1B, 2A and 2B, and 3A and 3B display the replacement rate estimates corresponding to our sample of pension regimes. Each figure corresponds to a different type of worker defined by her earnings level.Footnote ²³

According to the graphs, pension systems have evolved substantially and replacement rates of own earnings vary across earnings levels. At earnings of 50 percent of average earnings, replacement rates in Figures 1A and 1B for the United Kingdom, Sweden, and Spain were maximized as a result of flat pensions in most pension systems before World War II. After the war, it was mostly minimum pensions that raised replacement rates and generated redistribution.

However, between 1920 and 1940 a different pattern had already emerged in Italy, Germany, and Portugal, where rates approached full replacement by 1940. This had to do with the social policy of totalitarian regimes, where very high benefits were granted legally, even if these may not have translated in reality after waiting periods and subsequent legislation changes are taken into account. However, Italian and German differentials persisted after World War II, until around 1970. Sweden joined this group in 1960 after the country introduced the earnings related allmän tilläggspension (ATP) tier.

In general, the postwar era gave rise to “modern” pension systems, with a new treatment of pensions that was more sensitive to an adequate substitution of lifetime earnings levels at old age. Contributions as a function of earnings were therefore privileged. In fact, after 1950 all earnings levels were covered whereas, prior to that year, income limits had excluded higher earnings from benefits.Footnote ²⁴ The prioritization of contributions to determine benefit levels produced the rise of replacement rates between 1950 and 1980. Italy and Germany experienced the earliest and most striking rises, but the increase became general between 1950 and 1980. Increases took place at all earnings levels but rose most for higher earnings. For instance, replacement rates rose on average for all the countries in the sample by 42.8 percent between 1950 and 1980 for workers with the average level of earnings. The rise was slightly lower (39.9 percent) in the case of pensioners who had earned 50 percent of average earnings, but for workers with 300 percent of average earnings, replacement rates rose by 120.5 percent during the same period. Therefore, the golden age of welfare states achieved the inclusion of high earners in exchange for a more generous growth of replacement rates than for other earnings categories. This may be interpreted as a way to avoid opting out from those workers, whose contributions were key to the financial sustainability of pension systems if even moderate redistribution within systems was desired. In fact, minimum pensions developed in parallel to this trend. See, for instance, the case of France, which in 1950 granted a minimum pension that was binding for earnings categories up to 100 percent of average earnings before falling to 75 percent of average earnings by 1970. In Germany, minimum pensions for low-income groups rose significantly, as the lifetime earnings of low-income workers with at least 25 years of coverage were raised for pension computation purposes to 75 percent of the national average wage.Footnote ²⁵ However, this “agreement” between pensioners at various earnings levels was only feasible under continued economic growth and demographic equilibrium among age groups. That, of course, would not be the case in the following decades.

Figure 1A

REPLACEMENT RATE OF PENSION BENEFITS AS PERCENTAGE OF OWN EARNINGS FOR INDIVIDUALS WITH 50 PERCENT OF AVERAGE EARNINGS

Figure 1B

REPLACEMENT RATE OF PENSION BENEFITS AS PERCENTAGE OF OWN EARNINGS FOR INDIVIDUALS WITH 50 PERCENT OF AVERAGE EARNINGS

Sources: Based on calculations described and documented in Appendix 1.

Figure 2A

REPLACEMENT RATE OF PENSION BENEFITS AS PERCENTAGE OF OWN EARNINGS FOR INDIVIDUALS WITH 100 PERCENT OF AVERAGE EARNINGS

Figure 2B

REPLACEMENT RATE OF PENSION BENEFITS AS PERCENTAGE OF OWN EARNINGS FOR INDIVIDUALS WITH 100 PERCENT OF AVERAGE EARNINGS

Sources: Based on calculations described and documented in Appendix 1.

Figure 3A

REPLACEMENT RATE OF PENSION BENEFITS AS PERCENTAGE OF OWN EARNINGS FOR INDIVIDUALS WITH 300 PERCENT OF AVERAGE EARNINGS

Figure 3B

REPLACEMENT RATE OF PENSION BENEFITS AS PERCENTAGE OF OWN EARNINGS FOR INDIVIDUALS WITH 300 PERCENT OF AVERAGE EARNINGS

Sources: Based on calculations described and documented in Appendix 1.

In the latest period, 1990–2000, it is interesting to look at the different effects of the wave of retrenchment on welfare states. Pensioners with low and average earnings experienced almost negligible changes in replacement rates. The exceptions were France, where replacement rates in the two lowest categories increased quite notably due to a significant rise in the minimum pension; and the United States and Spain, where pensioners with 100 percent of average earnings saw replacement rates rise mildly after the removal of relatively low pension ceilings. On the other hand, retrenchment occurred in the higher earnings categories (250 to 400 percent of the average), despite the fact that replacement rates grew in Spain, Sweden, the United States, and Italy after the removal of benefit capping.

Note, however, that the wave of retrenchment was earlier in some countries. For instance, the United Kingdom not only favored a gradual move towards privatization in 1980, 1986, and then 1995 that our analysis is not capturing, but also cut back pension generosity right after the significant progress that the introduction of an earnings-related tier (SERPS) in 1978 had meant. Moreover, these changes added to cuts in the basic, first-tier pension through stricter indexation.Footnote ²⁶

Germany also retrenched earlier, as replacement rates in Figure 3A began to fall from 1960 in the highest categories, and from 1980 in the average ones in Figure 2A. A slight rise followed in the early 1990s, especially in the higher categories but then fell again in 2000 because pension points, which are awarded as a function of earnings and enter the pension formula as a multiplier, were capped at a notably lower level.Footnote ²⁷

The information drawn from our database allows us to test for the cross-country convergence of pension benefits measured as replacement rates. Figure 4 shows the coefficient of variation of replacement rates as a measure of dispersion for each earnings level and year.

The dispersion increased from 1910 to 2000 in every category but the levels of dispersion in 1920 are almost the same as in 2000. There were large changes in between with peaks in 1950 for many higher earnings pensioners. Meanwhile, the dispersion for pensioners at middle and low earnings levels had flatter profiles and peaked around 1930 for categories up to 100 percent of average earnings and in 1940 for pensioners with 150 percent of average earnings. Generally, the dispersion around 1990 and 200 was substantially lower than in the peak years.

In general, the move toward economic integration that followed World War II has been associated with convergence of pension generosity for most of the period with a slight rise in dispersion between 1990 and 2000. However, the evolutions and peaks differ depending on whether high or low earnings levels are being examined. These differences show the important effect for cross-country dispersion of differences in the timing of the introduction of minimum pensions and the tightening of the link between contributions and pensions, by removing ceilings on pensionable earnings and benefits.

Figure 4

DISPERSION OF REPLACEMENT RATES BETWEEN COUNTRIES, FOR EACH LEVEL OF EARNINGS

Note: Each series represents a percentage of average earnings levels. Dispersion is not reported for 1890 and 1900, because Germany is the only observation for those years.Source: Calculations are from the database in Appendix 2.

THE DETERMINANTS OF PENSION POLICY

After examining the evolution of pensions throughout the twentieth century, the factors explaining these outcomes need to be identified. Therefore, I will examine the relationship between the log value of pensions and a number of correlates using my panel database. I have obtained coefficients for each of the nine earnings categories contained in the panel of data. Also, both country and year fixed effects are used to control for the time-invariant heterogeneity across countries and worldwide shocks.

Most of the variables tested are the “usual suspects” identified in the literature to explain welfare states and pension development.Footnote ²⁸ I expect a positive relationship with real GDP per capita and with the share of the population older than 65—for both sexes together and for men only—due to “grey power.” The relationship with democracy is likely to be positive, although it is not straightforward.Footnote ²⁹ I also check whether the share of male agricultural workers represented a burden for pension growth.

The effects of globalization are also measured. The degree of commercial openness may not capture adequately the external dimension of pension policy. Trade openness as a whole may not show how trade and economic interdependences affect the design of pension policy, so it is especially important to look at the country or group of countries with which economic exchange is more intense. Therefore, I examine two different channels of foreign influence on domestic policy, both dealing with bilateral trade. First, I look at the “dynamic adaptation” that may take place when a preferential trade partner has a generous pension regime.Footnote ³⁰ The sign of this effect is ambiguous. On the one hand, a strong trade partnership could induce the replication of pension policy features, for instance, through ideological or policy transfers or, perhaps, as a way to control labor outflows to more generous schemes. On the other hand, the sign of the relationship could be negative if countries gave priority to the competitive advantage opened when trading with a generous welfare state. In this case, there is an incentive to keep pension standards low and compete through lower labor costs.

The second channel I look at is the “persistent” dimension of the effect of trading with an early pension developer.Footnote ³¹ Preferential trade partnership with an early adopter likely has a positive effect on domestic pension policy. My underlying hypothesis here is that trade with an early pension adopter means that some intangible elements, like shared traditions and ideology, in favor of welfare state expansion flow together with economic exchange and result in enhanced pension policy. Note that timing is the key element of analysis, as early adoption is not necessarily equivalent to generosity, either at the onset of the pension system in the leader countries or later.

The effect of migration rates on pension development is measured too. It has been shown that social insurance retains workers in their home countries while “welfare magnets” attract migrants to a small degree.Footnote ³² Inversely, a net share of incoming migrants may result in enhanced social spending for the lowest earnings categories, if policymakers choose to protect them against a more abundant supply of mostly low-skilled labor. The effect of migration is, thus, ambiguous.

Table 2 reports the basic results. The effect of intense trade contact with an early adopter of pension insurance has a positive and significant sign in the high earnings categories of 200 to 400 percent of average earnings. This suggests that contact with a pension leader, irrespective of its later policy path, makes pension policy more generous for higher earners. Note, however, that the relationship is ambiguous and not statistically significant for average and lower earners. Therefore, the effect of trading with an early developer translates into the prioritization of the insurance-related aspects of public pensions. In other words, income replacement through insurance is favored and long, continuous contribution histories rewarded, typically through removal of earnings and contribution maxima. Interestingly, this would constitute a very particular interpretation of what the early welfare state meant, given that insurance principles dominated Bismarckian-based systems but this was clearly not the case in the United Kingdom, which also was an early adopter but more oriented towards universal benefits that guaranteed income security.

On the other hand, if the country is trading actively with a preferential partner with generous benefits, the coefficient is negative and statistically significant for all earnings levels. This means that, when a country trades with a generous welfare state, priority will be given to the competitive opportunity opened by this fact and pensions will consequently decrease.Footnote ³³

There is no statistically significant relationship between migration rates and pension replacement rates until earnings reach 300 or 400 percent of average earnings. This might suggest that, in the light of their very long-term nature, pensions do not act as welfare magnets for migrants. Nor are they used to protect lower earnings against increased competition in labor markets due to new, mostly unskilled entrants. In the cases of net emigration, which dominate in the earliest observations of the sample, the absence of statistically significant results shows that the prospect of foregone pension benefits at the home country is irrelevant for the decision to migrate.

Table 2

DETERMINANTS OF ESTIMATED PENSION OUTCOMES (dependent variable: log value of real pensions)Footnote ³⁴

* Significant at the 10 percent level.

** Significant at the 5 percent level.

*** Significant at the 1 percent level.

Notes:

T-statistics are in parentheses. Democracy corresponds to the POLITY IV index of democracy in CIDCM; Preferential trade with generous country: dummy equal to 1 when a country has preferential trade with a generous pension system; Preferential trade with early pension adopter: dummy equal to 1 when a country has preferential trade with an early adopter of pension; and Migration: migration rate (immigrants + emigrants over total population)*100.

The coefficients on the share of male agricultural workers suggest that countries with larger shares of the agricultural workforce have lower pensions awarded.

Most of the remaining variables perform as expected. GDP per capita is positively related with pensions with an elasticity that ranges from 0.45 at low earnings levels to 0.58 at high earnings levels. The same applies for the share of men aged 65 or more. The positive coefficients show that “gray power” overcomes opposition based on worries about excessive pension burdens in deciding the amount of pensions provided.Footnote ³⁵

The sign of democracy, however, constitutes a surprise, as it shows a negative relationship to pension provision for all earning levels. In light of this result, one may speculate that, if there is a positive connection between democracy and pensions, as hypothesized, it may have to do with the extension of benefit entitlement. This constitutes another side of pension generosity that is different from the benefit levels analyzed here.Footnote ³⁶