Estimating the Effects of Earnings Uncertainty on Families' Saving and Insurance Decisions.

• Author: Palumbo, Michael G.

Michael G. Palumbo [*]

This paper investigates whether families save partially to self-insure against uncertain future earnings and estimates the extent to which pooled insurance substitutes for saving as a precaution against earnings risk. An econometric model is estimated using unique household survey data from the nineteenth century to examine families' joint saving and insurance decisions. The historical microdata are of interest because they predate widespread social insurance programs, which may stunt self-insurance through private saving among contemporary families. The econometric results imply that two independent measures of idiosyncratic earnings risk matter for families' saving decisions. Additionally, families whose primary wage earners belong to labor unions significantly save less frequently than others, all else being equal, but this tendency is not apparent among members of private benevolent societies.

1. Introduction

   Households face considerable uncertainty regarding their future economic circumstances. Two responses to idiosyncratic economic risks that families can undertake are saving and insurance. The increase in saving (or the reduction in current consumption) a prudent family uses to self-insure against possible future economic emergencies is called precautionary saving. [1] In an insurance arrangement, a family pools its risk with other families to guarantee receiving an insurance benefit, should an unfortunate economic contingency occur.

   Saving and insurance both represent potentially attractive financial instruments for protection against earnings risk, for example, but for different reasons. If this year's draw of annual earnings is normal (i.e., if a low-income contingency concerning the family does not occur), assets that have been set aside in the past as self-insurance are available to finance current consumption or can be used again to self-insure against future economic risks. Insurance premiums are paid prior to the resolution of uncertainty and are consumed regardless of the earnings realization. These arrangements make self-insurance through private saving attractive to families relative to pooled insurance coverage. On the other hand, families who self-insure must set aside one dollar today for every dollar they wish to have tomorrow in case of an emergency. Insurance purchasers, meanwhile, buy a dollar of insurance benefits for less than a dollar premium because they have pooled their idiosyncratic risks with other families. In this sense, a given desired benefit level can be achieved inexpensively through pooled insurance coverage relative to self-insurance.

   In this paper, I estimate an econometric model using historical microdata to investigate three empirical issues related to families' saving and insurance decisions. The first is whether uncertainty regarding the annual earnings of a family's primary wage earner can be identified as an important factor for its saving decision. Many other studies address this same issue, but a consensus of results has not yet arisen (see Browning and Lusardi's survey, 1996). Second, the empirical analysis estimates the extent to which pooled insurance coverage substitutes for self-insurance. Third, the model and data employed allow me to identify which types of pooled insurance arrangements families used, in an era predating government-provided social insurance and in addition to self-insurance, to guard against uncertain future earnings.

   Thus, this paper relates to several recent studies in the literature. First, Cochrane (1991) and Mace (1991) report empirical tests for full consumption insurance using microdata for American families during the 1980s. These researchers reach opposing conclusions: Cochrane rejects the null hypothesis of complete consumption insurance, and Mace's preferred specification reveals no statistically significant effect of a family's income realization on its spending, conditional on aggregate expenditure. [2] More recently, Nelson (1994) revisits Mace's specific empirical application, using the same and some additional data, and presents convincing evidence to overturn Mace's original results.

   Though complete consumption insurance does not seem to adequately describe the U.S. contemporary economy, one might well turn to the task of trying to identify just which types of financial arrangements allow families to smooth consumption across uncertain circumstances. Indeed, Cochrane (1991, p. 974) concludes his paper by suggesting that future research should "try to isolate the (insurance) alternatives rather than simply reject the null (complete consumption insurance)." The database I study contains information about three insurance alternatives that might substitute for self-insurance through private saving as protection against uncertain future earnings, so I undertake his suggestion in this paper. The potential insurance information in the data are memberships of families' primary wage earners in labor unions, memberships in private benefit societies, and indicators of life insurance policies held by primary earners in an era before social insurance in the United States.

   Also related to this study are empirical papers by Engen and Gruber (1995) and Kantor and Fishback (1996), which investigate how families' saving decisions respond to variations in government-provided, social insurance programs. [3] Engen and Gruber study the impact of unemployment insurance benefits on saving among contemporary American families; Kantor and Fishback report on the introduction of workers' compensation policy in the United States around 1918. Both papers take cross-state variation in government insurance programs to be exogenous, thereby justifying an empirical comparison of annual saving (Kantor and Fishback 1996) or wealth (Engen and Gruber 1995) across states. Both papers report statistically significant effects of social insurance benefits on family saving, holding other household-specific factors constant.

   While the two papers just cited consider the implications of social insurance on private saving, the data studied here cover a period in U.S. economic history when families required private means to protect themselves against economic emergencies, such as unemployment spells, illnesses, accidents, strikes, and premature death of primary earners. Thus, this paper actually is most similar to Starr-McCluer's (1996). Both she and I examine how families use their own saving and privately arranged insurance coverage, rather than government-provided mandatory coverage, to protect themselves against risky economic outcomes. Starr-McCluer studies health insurance coverage and saving among contemporary American families; as stated, I investigate labor union and benevolent society memberships, in addition to life insurance coverage, before the turn of the twentieth century. These two papers use similar regression models and, thus, encounter similar econometric issues--endogeneity in the context of limited dependent var iables.

   My results, on the other hand, align more closely to those reported by Engen and Gruber (1995) and Kantor and Fishback (1996) than to Starr-McCluer's (1996). I argue that my use of a more appropriate econometric approach to estimation, and more importantly, access to better instrumental variables, at least for the case of labor union membership, yields consistent estimates of the substitution parameter of interest. My conclusion is based on finding results closest to Starr-McCluer's (1996)--no apparent substitution between self-insurance and private insurance--for the cases of benevolent society memberships and life insurance coverage, for which specification tests indicate an inadequate set of instrumental variables for purging endogeneity from the regressions.

   Thus, the paper's econometric results support the hypothesis that working-class families during the late nineteenth century saved in part as a precaution against uncertain future earnings of their primary wage earners. Families whose primary wage earners have relatively large earnings variances due to tendencies to be out of work save more frequently than others. Additionally, the number of wage earners belonging to a family is inversely related to its past saving propensity. Empirical relationships between saving and insurance coverage are more mixed. I report robust evidence that labor union membership by a family's primary wage earner reduces the probability of it saving, but neither benevolent society membership nor life insurance coverage apparently affects family saving. Without additional information, I cannot determine whether the inconsistent results arise because of differences in benefits provided by labor unions benevolent societies or because of lack of econometric power due to ineffective instr umental variables in the latter case.

   This paper is organized as follows. Section 2 describes the microlevel, historical data on which the paper's results are based. A general econometric model for a family's decision to save and to purchase insurance consistent with the available data is specified in section 3. That section also describes problems involved in estimating the general saving and insurance model because of unobserved heterogeneity among families and limitations in the database (censored endogenous regressors). The empirical results are presented in section 4, and section 5 summarizes the paper's findings.

2. Saving and Insurance Data for Working-Class Families in Maine during 1890

   In 1890, the Bureau of Industrial and Labor Statistics for the State of Maine dispatched "faithful and competent special agents" to gather economic information through personal interviews with working-class families (Ransom and Sutch 1989b). The survey data contain many variables needed to estimate the saving and insurance model, as well as those required to estimate a time series variance of earnings for each primary wage earner in the...