Bailout for sale? The vote to save Wall Street

The original political agency models were novel in that they treated the electorate as the principal instead of the government. Rather than a benevolent government trying to maximize the welfare of an adversarial polity, as had been the tradition in public economics, this research supposed instead that it was the government who was out to game the electorate. Elections are a means of “controlling” politicians (Barro 1973). Besley (2006) reviews the literature on political agency.

This is similar in spirit to Snyder and Ting (2008), for example, who differentiate politicians according to the intensity of their preferences for holding office. In my model, politicians are differentiated by the characteristics of the districts they represent.

Bowman and Rugg (2010) have collected polling data from around the time of the bill’s vote. For example, in October 2008, a CBS/NYT poll asked “Do you approve or disapprove of the economic bailout plan that was passed by Congress and signed into law by the president?” Over October 10–13, the approval rating was 28%. Over October 19–22, the approval rating was 32%. Public support for the bill was never greater than 50% in the polls and declined steadily in the following months. Approval rates were higher for questions that did not include the word “bailout”.

The budgeted cost of the TARP bill was $700 billion and required a 25% increase in the federal budget for fiscal year 2008 and a 7% increase in the total debt position of the United States (Congleton 2009). The initial TARP proposal by then Treasury secretary Paulson was a remarkably vague two-page document that provided virtually no legislative oversight for the secretary’s decisions as to how he would distribute the $700 billion requested from (and granted by) Congress. Li (2011) demonstrates that TARP recipients were politically connected and estimates that only one third of the TARP capital was used to initiate new loans, which was one of the bill’s marketed intents. At the time, the public perception was that the taxpaying public financed a redistribution “up the income distribution” to the executives of the major Wall Street banks to whom TARP funds would be allocated. Indeed, the banking bailout led to deeper consolidation of market power within the commercial banking sector and some of TARP’s beneficiaries counted record profits (and record executive bonuses) in the following fiscal year (Johnson and Kwak 2010). Additionally, many commentators have noted that the further consolidation of market power may have exacerbated incentive problems in the largest commercial banks, sowing the seeds of a larger “boom-bust-bailout cycle” in the years to come (Boone and Johnson 2010; Johnson and Kwak 2010; Dorsch 2010, for example). If these banks were considered too-big-to-fail before the bailout, they are even bigger after the consolidation of market power achieved with the help of Washington. The moral hazard problem of banking executives, none of whom lost their jobs, has thus been amplified and encourages riskier lending than before the crisis. The lasting unpopularity of the TARP will not be found in the fact that the financial system was recapitalized, but the manner in which it was accomplished. On point (iv), one is reminded of Allan Meltzer’s famous lament about IMF subvention of East Asian governments in the 1990s: “Capitalism without failure is like religion without sin. It doesn’t work.”

Empirically as well, the bailout bill seems well-suited to investigate the role of campaign contributions in legislative voting. Voting patterns in Congress did not follow party lines and the presidential race between Senators Obama and McCain was tight at the time of the vote. Additionally, the initial proposal did not pass in the House, being defeated 228–205.

Campaign contributions help incumbents maintain political office, which gives politicians utility. In this sense, contributions can approximate cash rents. For reasons of data availability, most work has focused on registered campaign contributions, which are legal monetary payments to politicians that are not explicitly quid-pro-quo. The difference between campaign contributions and bribery may largely be legalistic. As a theoretical economic transaction, they are equivalent (Welch 1974).

A model where second period contributions depend on the first period vote is especially relevant for legislative votes that are existential for the special interest. Getting support from a cadre of congressional legislators by contributing to their political campaigns is akin to an existential hedge. For more on the “services” aspect of what campaign contributions buy, see Baron (1989) and Snyder (1990), for example.

The only assumption I make is that special interest legislation is unpopular; whether the policies are beneficial or harmful for the economy is not considered here. Political accountability is modeled in a very reduced-form way. Implicitly, voters use retrospective strategies to keep incumbents accountable. See Schmidt et al. (1996) for evidence from US senatorial elections that voters use retrospective strategies when voting incumbents out.

A simple empirical analysis presented in the next section supports this conjecture. Voting to support the bailout increases the probability that a representative lost her seat in 2008 or the 2010 midterms, but the effect is weaker for representatives of districts with a greater proportion of the employed working in the financial services sector. Additionally, note that, in reality, θ _ij would be a vector of constituency characteristics rather than a scalar.

Note that elections affect legislative voting decisions in a very reduced-form way in my model. I do not, for example, allow campaign contributions to affect the probability of getting re-elected as in Austen-Smith (1987), Congleton (1989) and Prat (2002). As an anecdote, cable news political commentator (and former Congressman) Joe Scarborough repeatedly notes that incumbent politicians rarely get voted out of office on the basis of voting against controversial legislation.

In Besley (2006) and other models that focus on political selection, the assumption is that some politicians are cognitively “dissonant” by nature and, by implication, always for sale. In my set-up, incumbent legislators cast “rationally dissonant” votes to support the industrial special interests on an issue-by-issue basis.

Stratmann (1991) finds evidence to support this logic. In his study of congressional voting on farm subsidies, he finds that representatives of “wheat states” took fewer campaign contributions from the wheat interest group.

I thank an anonymous referee for recommending that I concentrate on the initial bailout vote. Congleton (2009) and Johnson (2009) provide accounts of the bill’s ultimate passage.

Data are freely available at the Center’s website, www.​opensecrets.​org. It is worth noting that the website collects data from government records about reported campaign contributions, so is likely to underestimate the amount of money that actually found its way into legislative politics (Welch 1974). Moreover, there are in-kind compensations for legislators, which are more difficult to quantify, such as seats on corporate boards, jobs for girlfriends and nephews, invitations to play in exclusive foursomes at Augusta National, and so on.

Note that N=433 for vote 674. Representative Jerry Weller did not cast a vote on that measure.

The DW-Nominate score is an ideology rating ranging between −1 and +1.339 (in the 110th Congress), based on historical voting records on government intervention in the economy. Higher DW-nominate scores indicate legislators with a less interventionist voting history. A priori, more interventionist legislators should be more likely to support the bailout, due to higher ideological costs (Δ) of not doing so. A negative coefficient estimate therefore is expected on the dw variable.

Note that the probit marginal effects are non-linear. To find discrete change effects, first use the estimated probit coefficients to predict the probability of supporting the bill for the average congressman, that is, \(\widehat {\mbox{Pr}(\mathit{vote}\,674=1)}=\varPhi(\widehat{\alpha}+\widehat{\beta}\times \overline{\mathit{contrib}}+\widehat{\gamma'}\overline{\mathbf {X}})=0.4880\), where the overline indicates sample mean values. Next, predict the probability of supporting the bill for an average congressman who receives one standard deviation above the mean contribution from finance, that is, \(\widehat{\mbox{Pr}(\mathit{vote}\,674=1)}=\varPhi(\widehat{\alpha}+\widehat{\beta }\times(\overline{\mathit{contrib}.}+\mathit{std}.\mathit{dev})+\widehat{\gamma'}\overline {\mathbf{X}})=0.7389\). The change in predicted probability reported in the text, 25.49, is the difference between these two probability estimates. The calculations of the effect of a standard deviation increase are performed in the same manner for the commercial banking specification presented in Table 2 as well.

Chappell (1982) was the first to point out the problem of endogeneity in estimating the effect of campaign contributions on voting behavior. Ansolabehere et al. (2003) reviews the literature and emphasizes that papers which control for endogeneity using instrumental variables find less support for the notion that votes are responsive to campaign contributions. Stratmann (2005) reviews the literature and rejects the notion that campaign contributions do not influence congressional voting.

The pairwise correlations of the instruments with contributions from finance (p-values) are 0.293 (0.00) for Financial Services Committee membership and 0.176 (0.00) for Ways and Means Committee membership. Contributions from the commercial banking sub-sector are also strongly correlated with memberships on these committees. The pairwise correlations of the instruments with vote 674 are 0.029 (0.55) for Financial Services Committee membership and 0.066 (0.17) for Ways and Means Committee membership. Details of the relevance and strength of the instruments are presented in the discussion of the two-stage least squares estimations below.

It is possible, however to fail to reject the null hypothesis of exogeneity if the instruments are weak. I thank an anonymous referee for making this point. The strengths of the instruments are investigated in more detail below.

The results are very similar if party affiliation is used as the ideology control rather than the DW-Nominate ideology measure.

The Shea Partial R ² is large in both cases, and the irrelevance of the instruments is rejected at the 1% significance level. As for the strength of the instrument, the table reports the Kleibergen-Paap Wald rk χ ². In all specifications, the null hypothesis that the model is under-identified is rejected at the 1% significance level. Finally, Table 3 reports the Kleibergen-Paap Wald rk F-statistic for testing the null hypothesis that the instrument is weak. For background on these tests, see Shea (1997), Stock and Yogo (2005), Baum et al. (2007), and Kleibergen and Paap (2007). When interaction terms are not considered, the null of weak identification is easily rejected. In sum, the two committee memberships are relevant and non-weak instruments for use in 2SLS estimation.

Since 2SLS is a linear estimator, the impact of a one standard deviation increase in contributions from finance is simply \(\widehat{\beta}\times \mathit{std}.\mathit{dev}\).

The weak instrument F-statistics are lower when considering the second endogenous variable, though the weak instrument null is rejected using Stock-Yogo critical values (Stock and Yogo 2005). In the total finance specification, the null hypothesis of weak identification is rejected at a reasonable significance level, but the evidence against weak identification is more modest for the total commercial banking specification. Note, however, that the results of the 2SLS regressions with interaction terms are robust to weak instruments, according to the Anderson-Rubin F statistic. The null of the Anderson-Rubin test, that the estimates are robust to weak instruments, cannot be rejected.

I thank an anonymous referee for suggesting these calculations.