Shelter From the Storm

A growing body of literature suggests that asset limits in public assistance are associated with low savings rates among low-income families. Several states have begun eliminating or significantly increasing asset limits in an attempt to address potential disincentives. The primary concern for other states, however, appears to be the possibility that caseloads would increase to unsustainable levels, especially in times of economic recession. Five states that eliminated or increased asset limits during the Great Recession were analyzed for changes in caseload size after the rule change. Results suggest that there is no significant relationship between asset limits and caseload size.

Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 3.0 License (http://www.creativecommons.org/licenses/by/3.0/) which permits any use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access page (http://www.uk.sagepub.com/aboutus/openaccess.htm).

Article
Many American domestic policies are aimed at encouraging middle-class families to save and build financial assets, including tax-sheltered retirement contributions, 529 plans, and mortgage interest deductions. These assets play an important role in a family's economic and psychosocial wellbeing. Household savings can be especially critical in times of recession and high unemployment. Programs aimed at the very poor, however, often discourage savings and financial independence.
Beginning in the early 1980s, social welfare policy moved toward a "stick" approach in public assistance eligibility, eliminating the "carrot" of wealth accumulation that plays a prominent role in the middle-class tax policies mentioned above (McDonald, Orszag, & Russell, 2005). Meanwhile, private sector policy shifted in a similar direction with the decline of employer provided pension funds. These changes in the public and private sectors have coincided with a growing gap in wealth accumulation between the highest and lowest income households (McKernan, Ratcliffe, Steuerle, & Zhang, 2013). Of the myriad explanations for this disparity, restrictive eligibility rules may play an important role. For example, low resource limits (maximum allowable financial assets for eligibility) in the Temporary Assistance for Needy Families (TANF) program are associated with lower savings rates among all low-income families, not just those receiving TANF (Nam, 2008).
For this reason, several states have eliminated or significantly increased asset limits for TANF eligibility. While some worry that allowing applicants to maintain greater assets would cause an unsustainable swelling of caseload size, two early adopters of elimination from Ohio (eliminated in 1997) and Virginia (eliminated in 2004), found that caseloads actually decreased after this change (Parrish, 2005). However, five other states that eliminated or significantly increased their asset test (Alabama, Colorado, Delaware, Los Angeles, and Maryland) did so between 2007 and 2011, in the midst of the Great Recession, so it is important to determine whether this policy continues to be workable during periods of increased social need. The aim of this study is to ascertain whether elimination or significant increase of the asset test is associated with changes in caseload size since the beginning of the Great Recession.

Literature Review
Asset building is an important source of security for families, especially in times of economic crisis. In essence, "poverty is a trap of low assets" (Schreiner & Sherraden, 2006, p. 385). Low-income households that tend to have limited savings run the risk of financial ruin due to a sudden medical emergency, immediate household needs, and so on. Ownership of assets also shifts people's concepts about the future and by casting the future in a positive light, actually improves wellbeing in the present. A study by Yadama and Sherraden (1996) based on data collected from the Panel Study of Income Dynamics (PSID), an ongoing longitudinal survey of 5,000 families, supported the idea that assets positively influence expectations and confidence about the future and encourage people to make specific plans with regard to work and family. Furthermore, a positive association exists between a mother's assets (home ownership and savings) and a child's educational achievements (Zhan & Sherraden, 2003).
Asset limits in social programs, such as TANF, have traditionally been put in place to allocate assistance to people without the means to support themselves. Those who fall above the set asset limits are deemed capable of self-sufficiency and are therefore excluded from assistance programs. However, some have argued that asset tests actually discourage savings (Chen & Lerman, 2005;Rand, 2007). A study by Hubbard, Skinner, and Zeldes (1995) found that asset-based social programs are associated with low savings accumulation among low-income households. The authors suggest that this is due to a lack of incentive to save, because saving will disqualify families from social programs instead of guaranteeing self-sufficiency. With many limits as low as US$2,000 (Urban Institute, 2010), saving even US$2,001 would make a family ineligible for assistance but practically unable to weather any but the smallest financial setback.
States were allowed greater flexibility in setting asset limits for Aid to Families with Dependent Children until the Omnibus Budget Reconciliation Act of 1981, which created much lower allowances (US$1,200 for a vehicle and US$1,000 in other liquid assets) than many states had previously set (McDonald et al., 2005;Nam, 2008). Powers (1998) studied changes in savings behaviors among lowincome, female-headed households before and after 1981 and discovered that savings decreased by an average of 25 cents for every dollar decrease in asset limit. States were once again allowed to set asset limits after the passage of the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 (PRWORA); however, many states have yet to change limits from levels established in the 1980s (McDonald et al., 2005).
Increased resource limits are associated with the accumulation of multiple types of assets, including savings accounts, vehicle ownership, and savings for retirement. Hurst and Ziliak (2006) reviewed changes in savings among lowincome families after passage of the PRWORA. In states with expanded TANF asset limits, single mothers were 13% more likely to own a car for every US$1,000 increase in asset allowance. Similar studies have suggested that lessening asset restrictions on vehicle ownership can lead to eligible individuals increasing vehicle assets (Bansak, Mattson, & Rice, 2010;Owens & Baum, 2012;Sullivan, 2006). Owens and Baum (2012) found that eliminating vehicles from asset tests is associated with increased vehicle ownership by anywhere from 8% to 20%. Vehicle ownership in turn makes it easier for people to access jobs and resources, further increasing a family's economic security.
Since the 1980s, many companies have made a vast shift from employer-managed pension plans toward retirement saving plans managed mainly by employees, such as Individual Retirement Accounts (IRAs) and 401Ks. Meanwhile, most asset-limit restrictions have not been reevaluated. As is, low asset-limit rules are in fact a disincentive for many families to save for retirement (McDonald et al., 2005). Significantly, the longer a state has allowed increased assets in TANF eligibility, the more likely lowincome families are to accumulate savings and have bank accounts (Nam, 2008).
In summary, wealth accumulation for low-income households is an essential element of both financial and psychosocial well-being. Eligibility rules for TANF have the potential ability to either reward or penalize this behavior. Ohio was the first state to eliminate TANF asset limits in 1997 (Rand, 2007). Although a slight increase in caseload was expected, caseloads actually remained at a record low and Ohio received high-performance bonuses for labor force attachment. Virginia has experienced a similar lack of caseload increase since an administrative rule change in 2004 that eliminated TANF asset limits. Virginia's Department of Social Services has also predicted that eliminating the asset test could "increase the assistance provided by US$127,200 for 40 families and provide US$323,050 savings in administrative staff time annually" (Rand, 2007, p. 629), in essence saving the state money. These anecdotal findings ought to be more thoroughly investigated in order to determine the practical viability of elimination. The aim of this study, therefore, is to investigate the effect of significantly raising or canceling the asset limit on TANF caseloads by state.

Method
Data on the asset limits for TANF eligibility by state are available via the Urban Institute's (2010) Welfare Rules Databook. Data on the caseloads by state and total caseloads for the United States as a whole are available via the Administration for Children and Families (2013). The assetlimit data available by state was first screened to identify states that had substantially raised, or abolished, their nonvehicular asset limits between 2007 and 2011. Substantially, here, implies a raise of the limit to ≥US$10,000. This amount was chosen, as it is consistent with recommendations by advocacy organizations such as the Corporation for Enterprise Development and proposed in President Obama's 2011 budget (Greer & Levin, 2014). The five states fitting these criteria are listed in Table 1.
Alterations in the non-cash asset limits (vehicles, IRAs, 529s, Individual Development Accounts (IDAs), 401Ks, etc.) by state will not be considered for the purposes of this analysis for two reasons. First, these criteria vary significantly in type and detail between states, which makes a cross-state comparison very difficult and of little explanatory value. Second, in the instance of vehicles, the non-cash asset limit is likely to have little effect on the saving habits of lower-income individuals and families, because one cannot, by definition, save a non-cash asset.
Of the states that fit our criteria for asset-limit changes in the time frame, the total number of individual TANF recipients per month was analyzed. Monthly caseload data are given in 15-month periods, running from the start of the Federal fiscal year to the end of the following calendar year. This means that the data for October to December for each year overlap with the following year's data from the Administration for Children and Families website and occasionally are subject to revision. This was not the case for any of the states in this study, so no adjustments to the figures were necessary.
The TANF caseload analysis will answer two questions: (a) How has the caseload changed since the rule change for each state? and (b) How has the rate of increase (or decrease) of new caseloads changed since the asset-limit change for each state?
We have transformed the monthly caseload data from the ordinary calendar style so that all months are now reckoned relative to the month of the rule change in that state, which we denote as Month 0. To test for significant changes in caseloads over time, we use a segmented regression model (Wagner, Soumerai, Zhang, & Ross-Degnan, 2002). The Durbin-Watson test was used to evaluate the level of autocorrelation among errors (Durbin & Watson, 1950), and when significant autocorrelation was present, we used an Autoregressive Integrated Moving Average (ARIMA) model to reduce bias in the estimates of intercept and slope changes for each state.
A second method to address the question of how abrupt policy changes may be affecting caseloads over time is to look for unique "change points" in the time series trends. To find the most likely change points in the time series for each state, a bootstrapping search algorithm was applied to the entire time series (Muggeo, 2003(Muggeo, , 2008. Starting values of 24 months and 32 months were used for the search patterns to allow for any lag in the implementation of the policies. Even with these starting points, the algorithm can efficiently search all possible segments dividing points for the most likely change point values. All analyses were performed using the R statistical programming language (R Development Team, 2013).

Findings
To compare the different state caseload trends visually on the same graphic, caseload numbers were transformed to represent a percentage increase or decrease compared with the zero time point. Also, the time variable was centered such that the zero point for each state represents the date of the policy change. A cursory visual inspection of the time series for the five states under consideration reveals no common trend in behavior in the 2 years following the rule change ( Figure 1). Colorado (asset-limit change in October 2006) 1 shows a drop in caseload of around 30% relative to the Month 0 caseload. Virginia (September 2004) shows a slight drop in caseload of less than 10%. Alabama, Delaware, and Maryland show minor or moderate increases in caseload (<20%) relative to Month 0. Louisiana shows a periodic rise and then fall of caseloads, which may indicate a seasonal employment relationship of some kind.
Estimates for any change in caseload levels and slopes are shown in Table 2. As is evident from Figure 1, every state had first-or second-order trends in the time series. Durbin-Watson tests for each timeline indicated that error terms were in fact correlated, so the point estimates in Table 2 are from ARIMA regression models.
When the existing trends in the data are taken into account, Louisiana is the only state with significant changes in either level or slope of caseloads. Compared with the 24 months prior to the policy change in Louisiana, there was an average decrease in the TANF caseload of approximately 1,000 cases and concomitant rise in the number of cases added per month of approximately 230 cases.
The change point analysis reveals a similarly inconsistent pattern of changes in caseload trends both before and after the policy changes. Estimated change points and fitted regression lines for the resulting segments are shown in Figure 2. The two estimated change points for Louisiana, for example, were approximately 8 and 12 months before any policy changes took effect, so it is unlikely that any increases in caseloads over the 4-year period examined could be attributed to any change in asset limits for TANF. The states of Alabama, Delaware, and Maryland had estimated change points after the change in asset limits (14.5, 14.6, and 19.7 months, respectively). In all three of these cases, the change points were associated with sharp downward trends in TANF caseloads. In the case of Colorado, there does seem to be a leveling off in what had been an already decreasing trend in TANF caseloads that occurs at approximately 10.3 months after the change in asset limits. This could be a lagged effect of the policy changes, but it would be difficult to assign a cause without further analysis.
In summary, a large increase in asset limits for TANF eligibility, or the abolishment of those limits, does not appear to have any causal affect to increase either the number of caseloads or the rate of caseload growth in any of the states examined.

Discussion
We are aware that the data set is small, and conclusions cannot be drawn from it without caution. However, the fact that the states in this data set enacted their rule changes during the height of the worst economic recession in living memory and still found no associated increase in caseload or rate of caseload growth is, we submit, powerfully suggestive of the idea that the asset limit does not affect TANF caseloads. Because each state appears to be experiencing its own individual caseload trajectory independent of the asset-limit rule change, it is possible that local-level factors are influencing the change, including economic conditions or additional policy changes that either restrict or liberalize eligibility. It would be difficult to design a deeper test of this idea than to substantially raise the asset limits for TANF eligibility in three states whose unemployment rates were, at the time of the rule change, at least three percentage points higher than the average rate in the decade prior to the recession (Bureau of Labor Statistics, 2013). We regard the lack of impact of asset-limit changes on caseload in these cases as powerful evidence suggesting that these two factors are not correlated.
The ability to build assets is critical to the financial stability of low-income families, especially in times of economic recession. However, previously discussed research suggests that low asset limits in public assistance create a disincentive for the poor to save and sends a message that savings is discouraged. The greatest barrier to remediation appears to be a political one (Sprague & Black, 2012). Legislators and state human service administrators in many states fear that this rule change would "open the floodgates" to persons with large financial assets and eventual Note. Change point indicates month at which caseload change is estimated in the 24 months before and after the rule change, ranging from −24 to 24. Therefore, a Point-1 value of −6.3 would indicate that the first change is estimated to have occurred 6.3 months before the rule change (Month 0), and a Point-2 value of 14.5 is estimated to have occurred 14.5 months after the rule change. TANF = Temporary Assistance for Needy Families. misuse of public funds. These fears have been fueled by media reports of a lottery winner continuing to receive assistance after elimination of the asset test (Susman, 2012). This anecdotal information must be supplemented by empirical research regarding the true effect on caseload size. This analysis of five states that have eliminated their asset test in the midst of the greatest economic decline since the Great Depression is especially relevant as a measure of the safety net's stability in times of greatest citizen need. Future research ought to explore whether the elimination or significant increase of the asset limit has indeed boosted savings rates among low-income families, as predicted by research conducted prior to these changes (Hurst & Ziliak, 2006;Nam, 2008).