Rating Politics? Partisan Discrimination in Credit Ratings in Developed Economies

Abstract

How does government partisanship influence sovereign credit ratings of developed countries? Given the convergence of fiscal and monetary outcomes between left and right governments in the past decades, credit rating agencies (CRAs) should in principle not discriminate according to ideology. However, we hypothesize that CRAs might lower ratings for left governments as a strategy to limit negative policy and market surprises as they strive to keep ratings stable over the medium term. A panel analysis of Standard & Poor’s, Moody’s, and Fitch’s rating actions for 23 Organisation for Economic Co-Operation and Development (OECD) countries from 1995 to 2014 shows that left executives and the electoral victory of nonincumbent left executives are associated with significantly higher probabilities of negative rating changes. We find no evidence of similar systematic partisan bias in spreads on government bonds, but spreads do adjust to partisan-biased downgrades. This suggests that CRAs may introduce partisan discrimination into sovereign credit markets.

Keywords

political economy sovereign credit ratings business and politics

Credit rating agencies (CRAs) have emerged as important gatekeepers to sovereign debt markets over the past three decades. The power of the largest CRAs—Moody’s, Standard & Poor’s (S&P), and Fitch¹—over the fiscal fortunes of prosperous developed countries was thrown into sharp relief when they initiated a series of downgrades that triggered and perpetuated a succession of sovereign debt crises in Europe between 2008 and 2012. Although CRAs’ decisions are mostly motivated by the emergence of news about a country’s fiscal and economic performance, anecdotal evidence (such as the 2011 downgrading of the United States over the debt-ceiling debate or the 2012 collective downgrade of nine Eurozone member states over perceived governance problems within Europe’s common currency area) indicates that CRAs incorporate political considerations into their assessment of highly rated (and politically stable) sovereigns’ debt.

Although sovereign ratings received ample and growing scholarly attention in the past two decades, it remains unclear what sort of political factors influence the sovereign credit rating scores of prosperous developed countries. Most studies have concentrated on identifying key economic and institutional factors that have the greatest effect on rating scores (e.g., Afonso, 2003; Afonso, Gomes, & Rother, 2007; Cantor & Packer, 1996). Scholars who examined the influence of political factors on CRA decisions did so strictly in the context of developing economies, whose access to global financial markets is especially sensitive to the pronouncements of rating agencies. Most focused on large-scale systemic factors, such as regime type or political stability, which show greater variation in (and therefore, have particular relevance for) developing countries (e.g., Archer, Biglaiser, & DeRouen, 2007; Beaulieu, Cox, & Saiegh, 2012; Biglaiser & Staats, 2012; Block & Vaaler, 2004; Haque, Mathieson, & Mark, 1998; Vaaler, Schrage, & Block, 2005).

We explore whether government partisanship has an effect on the sovereign credit ratings of prosperous developed countries. Thereby, we seek to make three contributions. First, we focus on sovereign credit rating activity in an important, but so far scantily explored, group of countries (developed economies) whose vulnerability to changes in credit ratings has been highlighted by the unprecedented frequency of downgrades since 2008. Although emerging market economies are more likely to encounter debt crises and defaults due to their underdeveloped financial markets (Reinhart & Rogoff, 2008), recent events suggest that sovereigns with fully developed financial markets are not as immune to sovereign risk as previously thought. Because developed countries characteristically have more stable economies and better access to financial markets than developing economies, domestic politics is likely to play a more significant role in the variation of perceived sovereign creditworthiness.

Second, we shift the attention from systemic political factors (which are at the focus of most existing scholarship) to the role of day-to-day politics (elections and government partisanship) in sovereign credit ratings. Two studies suggested that CRAs discriminate against the center-left, particularly after elections in developing economies (Block & Vaaler, 2004; Vaaler et al., 2006). We seek to establish the presence or absence of such partisan discrimination in developed countries, which (by virtue of being stable democracies) provide particularly fertile ground for exploring how everyday politics enters assessments of creditworthiness, independently of concerns about political instability.

Third, by exploring whether and why CRAs treat executives of different partisan color differently in developed countries, we contribute to the debate over how markets influence and constrain politics in advanced economies. Evidence of partisan discrimination would lend support to claims that economic globalization places governments into a “golden straitjacket,” curtailing the scope for democratic choices in return for providing abundant access to financing (Rodrik, 2000). The absence of partisan discrimination, however, would suggest that developed countries (unlike their developing counterparts) have substantial “room to move” in their political and policy choices even in a globalized financial world (Mosley, 2000). Furthermore, exploring whether and how CRAs (highly visible market actors specializing in judging creditworthiness) discriminate between governments of different ideologies sheds light on the mechanisms through which politics may influence market outcomes.

We argue that although CRAs should in principle not differentiate on the basis of government partisanship once relevant policy differences are controlled for, they might espouse the expectation that left governments pursue less credit-friendly policies as a mechanism to insure against negative policy and market surprises over the medium term as they cannot frequently adjust their ratings to incorporate new information as it arises. Keeping ratings stable and reliably reflective of all potential risks to credit quality is essential if a CRA wants to ensure that its ratings are used as third-party credit risk measures in financial governance documents, which is crucial for retaining their credit rating market share. CRAs themselves emphasize in their official communications the need to combine reliable assessment of risks with relative stability. A conservative strategy to limit the negative surprises arising from the uncertainty involved in making rating decisions over the medium term is to preemptively incorporate all perceived risks into ratings, which implies assigning lower ratings to governments who have less political incentive to pursue policies that CRAs deem conducive to good credit quality. Bond markets, in contrast, do not operate under similar medium-term “stability” constraints because they can make adjustments upon receiving new information about governments’ actual policy choices instantaneously.

In our quantitative analysis, we test for the presence of partisan bias in the ratings of each of the big three agencies as well as the market’s pricing of sovereign risk to gauge the extent to which different market actors employ partisan discrimination. We find that CRAs discriminate against the left, but markets do so only insofar as they incorporate partisan-biased credit ratings into their decisions. We employ a linear probability model (LPM) panel analysis of negative ratings decisions by S&P, Moody’s, and Fitch for 23 Organisation for Economic Co-Operation and Development (OECD) economies from 1995 to 2014,² and then compare these results with an identical (ordinary least squares [OLS]) panel analysis of spreads on long-term government bonds (a proxy of market attitudes toward sovereign creditworthiness) for the same sample. We find that, ceteris paribus, downgrades were more likely from S&P and Moody’s under left-controlled executives (i.e., executives headed by left prime ministers). Fitch does not display similar biases against left prime ministers in their (negative) ratings decisions, but they do exhibit partisan bias in downgrades against executives with strongly left-oriented manifestos and cabinets with high shares of seats occupied by left parties (i.e., single-party left administrations). In addition, the probability of negative rating changes is even higher for S&P and Moody’s upon the election of nonincumbent left governments. Whereas Moody’s rating penalties against left governments is weak to robustness checks, and appears to be driven by the European debt crisis, S&P’s left government and left electoral victory biases precede 2009 and are robust to numerous alternative model specifications. In contrast, spreads on long-term government securities are not directly influenced by the partisanship of ruling executives, or by (nonincumbent) left electoral victories. Yet bond spreads are affected by CRA negative rating decisions, which implies that CRAs’ rating changes may be a crucial transmitter of partisan effects into sovereign bond markets.

The evidence of systematic negative discrimination by CRAs against left governments in developed countries contradicts the notion that developed countries are unencumbered by the political influence of markets. Furthermore, the absence of similar discrimination in spreads suggests that CRAs may be a source of the partisan impetus, rather than markets at large. In an effort to avoid frequent rating changes as well as negative policy and market surprises, credit ratings systematically give markets partisan signals, which are not linked to actual differences in fiscal performance under the left and the right. Thereby, they introduce theoretically questionable factors into assessments of creditworthiness. This reinforces already existing concerns about giving credit ratings a central role in prudential regulations.

Partial Arbiters? Creditworthiness, Partisanship, and Business Strategies in the Credit Rating Market

Sovereign CRAs wield considerable influence over countries’ access to credit through their pronouncements about creditworthiness. Although investors are in principle free to ignore CRAs’ opinion (and sometimes do³), credit ratings systematically influence the price of government debt (Afonso, Arghyrou, & Kontonikas, 2015; Brooks, Faff, Hillier, & Hillier, 2004; also see our results below). This is partly due to the authority and visibility of the three large CRAs, and also because upgrades and downgrades trigger automatic portfolio adjustments by investors who need to comply with portfolio regulations (de Haan & Amtenbrink, 2011; Sinclair, 2008).

The practical implications of a potential partisan effect on sovereign credit ratings are clear. If left governments consistently face higher probability of downgrades than right ones, all else being equal, they face more difficult market conditions and political challenges. First, left governments could experience higher borrowing costs and, consequently, tighter constraints on their fiscal policies. Second, as perceived sovereign creditworthiness can have knock-on effects on the private sector’s access to foreign capital, lower ratings can generate more adverse financial conditions for businesses under left governments independently of the government’s policy performance. Third, given that downgrades can trigger capital flight at times of economic and financial turmoil, countries may be at greater risk of plunging into a debt crisis under left governments. Finally, beyond these financial disadvantages, partisan discrimination in ratings can also create political costs for left governments if proneness to negative rating actions is interpreted as a sign of inferior ability to manage the economy.

The theoretical relevance of potential partisan discrimination in sovereign credit ratings lies in shedding light on the markets–politics nexus and of the distinctive role of CRAs in shaping that nexus. As governments tapped into international credit markets to gain access to more abundant and cheaper funding, scholars have debated to what extent their political and policy choices would be constrained by markets. Some argued that increased trade and capital mobility has limited the autonomy of governments in macroeconomic policy, and highlighted the convergence in macroeconomic indicators, especially deficits and inflation, as a sign of these constraints (Boix, 2000; Garrett & Lange, 1991). Others went further to argue that markets place governments in a “golden straitjacket”: they severely limit the role of democratic politics in policy choice in general (Rodrik, 2000). Mosley qualified this hypothesis by claiming that markets treat developed and developing countries differently. She argued that “financial markets’ influences on developed democracies are somewhat strong, but somewhat narrow”: While they pay attention to macroeconomic indicators, markets are less concerned about supply-side policies such as welfare provisions or the size of the public sector (Mosley, 2000, p. 766; 2004). Prosperous developed countries are given greater “room to move,” because their ability to repay their debt is not fundamentally questioned (Mosley, 2005). However, concerns about the power of financial markets over democratically elected (but underfunded) governments strongly resurfaced in the wake of the global financial and economic crisis (Streeck, 2014).

Market constraints manifest themselves as partisan discrimination if governments of different partisan color pursue systematically different policies. The rational partisanship theory posits that as representatives of labor, left parties are more likely to pursue expansionary policies to keep unemployment low, at the expense of higher inflation and deficits (Hibbs, 1977). By the same token, they are assumed to have stronger commitment to generous welfare policies than to fiscal stability. The expectation of higher deficits from the left received some empirical support in the past (Boix, 2000; de Haan & Strum, 1994), but has lost empirical validation in recent decades. Newer studies report little systemic relationship between left incumbents and deficits or the willingness to implement fiscal cuts to consolidate public finances (Hübscher, 2016; Mierau, Jong-A-Pin, & de Haan, 2007). Other political factors (i.e., power-sharing in multiparty or minority governments) have been found to be more important in explaining macroeconomic and government spending outcomes than partisanship (Bawn & Rosenbluth, 2006; Breen & McMenamin, 2013). Indeed, our own difference-in-means tests demonstrate that left executives in our OECD sample presided over similar levels of average debt and deficits levels as right executives since 1995. Although there may have been a relationship between partisanship and policy choices in the past, this relationship has faded over time in developed countries (Cusack, 1997).

If there is no systematic connection between partisanship and fiscal and monetary outcomes, markets should in principle not discriminate between left and right; but evidence suggests they do. Markets have been shown to react to the prospect of a left government in office with drops in stock prices (Bechtel, 2009; Sattler, 2013), increased inflationary expectations (Fowler, 2006), and greater exchange rate volatility (Bernhard & Leblang, 2002) even in developed countries. However, sovereign bond yields (indicators of sovereigns’ creditworthiness) do not seem to be systematically affected by government partisanship over time according to recent political economy literature. The impact of ideology on yields is modulated by the extent to which institutional constraints allow governments of different partisan color to pursue significantly differing policies (Breen & McMenamin, 2013). The conundrum between immediate market reactions to changes in government partisanship and the lack of systematic effect on bond yields in the long run can be resolved if we consider that markets flexibly adjust prices as information about actual policy choices becomes available, diminishing the initial effects of partisan expectations over the longer term.⁴

In contrast to markets, CRAs potentially have motivation to systematically employ partisan expectations in their rating decisions due to a combination of three factors: an express preference for social and economic policies (above and beyond maintaining fiscal and monetary stability) that right-wing governments tend to embrace, the inability to adjust ratings upon new information as flexibly as markets do, and potentially the urge to insure against downside policy and market risks when dealing with the uncertainty entailed in making rating decisions over the medium run. We elaborate on all three of these factors in turn.

Although none of the CRAs discuss the effect of government ideology on sovereign creditworthiness, S&P and Fitch express preferences for economic and social policies commonly associated with the right in their sovereign credit rating methodologies.⁵ All iterations of the big three’s sovereign rating methodologies⁶ steer clear of commenting on the effect of government partisanship on sovereign creditworthiness, not even mentioning the words “ideology,” “party,” “left,” or “right.”⁷ At the same time, all CRAs’ methodologies express a preference for conservative macroeconomic policies, such as low and stable inflation, and low deficits and public debt. Moreover, earlier editions of S&P and Fitch’s methodology (Fitch, 2002; S&P, 2006, 2008) also clearly advocated trade, labor market, and financial liberalization; praised small governments and low taxation; cautioned against large state-owned enterprises and entitlements; and urged welfare reforms to increase fiscal sustainability amid population aging (Moody’s methodologies and later iterations of S&P’s and Fitch’s methodologies do not explicitly comment on these economic and social policy issues). Such preferences make it plausible that CRAs prefer political forces that embrace such policies although in principle they should only make a distinction between right and left insofar as the relevant policy variables actually differ significantly under the tenure of the two sides.

CRAs might have incentives to use ideological labels as indicators of future policy choices—instead of waiting to see the actual policy choices—because they cannot flexibly adjust ratings every time there is a new policy development. CRAs need to keep their ratings stable (and reliably reflective of all possible downside risks) because rating stability and reliability is key to the use of their ratings as third-party credit risk measures in financial governance documents like portfolio mandates, contractual stipulations, or even public regulations. Widespread use of a CRA’s ratings in such documents is crucial to retaining market share. CRAs offer their services to borrowers who want to advantageously place their bonds on the market. Although this creates short-term incentives for CRAs to inflate ratings to attract bond issuers (Becker & Milbourn, 2010), CRAs can only retain market share in the long term by maintaining investors’ willingness to use their ratings to orient investment decisions because even the most favorable rating is useless for a bond issuer unless it induces investors to hold the bond. The dominant channel through which a given CRA’s ratings motivate large-scale buying and selling decisions is their use in financial governance rules. Most mutual funds, pension funds, insurance companies, private endowments, and foundations use credit ratings to set minimum credit standards (IMF, 2010, p. 92), and the portfolios of such institutional investors automatically have to be adjusted when ratings change. (Ratings are unlikely to achieve the same effect purely through providing risk analysis to investors, as most investors rely on in-house research to orient their decisions besides ratings). Making a CRA’s ratings suitable for use in financial governance documents requires not only that they reliably reflect all reasonable risks to credit quality, but also that they cannot change too frequently because the transaction costs associated with automatic portfolio adjustments would make the given agency’s ratings prohibitively costly to use (Cantor & Mann, 2007).

A conservative approach to dealing with the trade-off between rating stability and reliability is to incorporate all perceived possible future threats to credit quality into ratings ahead of time. Of the three rating agencies, S&P explicitly embraces such an approach. In its Credit Stability Criteria of 2016 (S&P 2016), S&P explains that it assigns lower ratings (to any type of issuer) than would arise from the analysis of current credit quality, if the issuer’s creditworthiness could slightly/significantly deteriorate in 1 or 3 years when encountering hypothetical scenarios representing “moderate stress” on credit quality. The document does not spell out the range of hypothetical scenarios (for any class of issuers), so it remains unspecified what sources of stress are taken into consideration when testing the susceptibility of sovereign credit quality. Presumably, however, they include significant adverse changes in the variables and policies specified in S&P’s sovereign rating methodology. This would not only imply policy choices disapproved by CRAs, but also potential adverse knock-on effects on economic and financial variables as a result of market responses to such policy choices.

The other two CRAs’ approach to the stability-reliability trade-off is more difficult to gauge. In a Special Comment (Cantor & Mann, 2007), Moody’s justifies its commitment to ensuring rating stability, provides estimates of the losses of accuracy that result from greater stability, and concludes that the trade-off seems to suit investors’ needs. However, Moody’s does not disclose whether this trade-off is achieved by factoring contingent future risks into ratings (similar to S&P’s practice) or by foregoing adjustments upon receiving new information as long as it does not drastically affect credit quality. We have not been able to locate similar official communication about rating stability and reliability from Fitch. Therefore, while we expect S&P to systematically discriminate between left and right governments in its rating decisions, it is not clear whether the other two would translate their similar policy preferences into preemptive rating changes linked to partisanship.

In sum, although there is no robust empirical evidence that currently links left governments to inferior fiscal outcomes, we expect that (some) CRAs might still systematically discriminate against left governments in an effort to manage the uncertainty of having to keep ratings stable over the medium term, given their preference for a range of economic and social policies usually associated with right governments. To better explore the validity of this reasoning, we test a series of hypotheses that examine the effect of (new and incumbent) left electoral victories and the left’s tenure in power on rating decisions. We differentiate between change in government and tenure in government to better understand the effect of political uncertainty around elections and separate it from a more generalized mistrust toward left governments. Our analysis incorporates the possibility that left governments that have been in power for a while benefit from having an established policy record, whereas newly incoming left governments are more subject to negative rating movements.

At the same time, we also test two auxiliary hypotheses about investor reactions to partisanship and about the relationship between investors’ decisions and ratings in an effort to better locate any potential source of partisan discrimination within sovereign credit markets. First, we investigate whether partisanship affects the pricing of sovereign debt to test the results of previous studies that show no systematic differences in perceived creditworthiness driven by government partisanship, once other factors are controlled for. Second, we probe our assumption that ratings have an impact on prices and are, therefore, materially consequential.

Hypothesis 1: Ceteris paribus, negative rating actions⁸ (downgrades and worsened outlooks) are more likely

(a) after the election of a new left executive,

(b) upon the reelection of an incumbent left executive, and

Hypothesis 2: (Auxiliary hypothesis): Ceteris paribus, spreads are higher

(a) after the election of a new left executive,

(b) after the reelection of an incumbent left executive, and

Hypothesis 3: (Auxiliary hypothesis): Spreads are higher in the wake of negative rating actions.

Partisan Discrimination in Sovereign Ratings: Empirical Evidence From 23 OECD Economies

We test the above hypotheses employing a series of panel analyses to determine the extent to which government partisanship affects ratings, and to infer how CRAs’ and markets’ views of government partisanship are related. Therefore, we have two dependent variables. Our primary dependent variable is the likelihood of negative rating actions. Our secondary dependent variable is spreads. Spreads are defined as the interest rate on the long-term treasury bill of the given country minus the interest rate for the 10-year U.S. Treasury bond.⁹ Rating actions are CRA communications about a sovereign’s creditworthiness. They can be not only downgrades, confirmations, or upgrades, but also revisions in the outlook assigned to the rating. Outlooks can be positive, negative, and stable, signaling the likely changes in the country’s rating score over the course of the next 2 years. Outlooks are not necessarily followed by a rating change and do not trigger automatic portfolio changes like downgrades or upgrades. However, they transmit important information to markets about the changing opinions of the CRA about a country. In fact, the IMF (2010, p. 105) finds that CRAs influence market prices to a greater extent through outlooks and credit watches than through actual rating changes.

We focus on rating actions as our dependent variable rather than actual rating scores for two reasons. First, the ratings of developed sovereigns are characterized by severe clustering at the top end of the rating scale. AAA ratings make up roughly 55%, 65%, and 55% of country-years in our panels for S&P, Moody’s, and Fitch, respectively. (Moreover, only 1.5%, 2.1%, and 2.2% of country-years for our S&P, Moody’s, and Fitch panels, respectively, are speculative grade.) This produces extreme “ceiling effects.” Extreme ceiling effects lead to biased parameters, particularly for binary independent variables, which our political variables are coded as (see Wang, Zhang, McArdle, & Salthouse, 2009). In this respect, our developed country sample generates methodological challenges that comparable studies working with broader samples of sovereign ratings do not face.¹⁰ Second, focusing on rating actions allows us to take into consideration those signals of changing CRA opinion (specifically worsened outlooks) that do not necessarily result in actual downgrades, but are nevertheless consequential for prices (as the above-referenced IMF study suggests). Despite these methodological considerations, we also test our hypotheses about partisanship and elections using rating scores instead of ratings changes as the dependent variable and report our results in Online Appendix A.¹¹

Our analysis examines the partisan determinants of rating actions for each CRA individually, rather than collectively, to better understand each agency’s strategy toward government partisanship. The source of information on the date and nature of rating actions is S&P’s Sovereign Rating and Country T&C Assessment Histories (September 4, 2013) and various rating reports, Moody’s Rating List,¹² and Fitch’s Sovereign Ratings History.¹³ Between 1995 and 2014, Fitch, Moody’s, and S&P initiated 39, 39, and 52 downgrades and 26, 25, and 50 worsened outlooks, respectively (see Figure 1).

Figure 1.

Rating events by credit rating agency (23 OECD countries, 1995-2014).

We use two codings for our “rating action” dependent variable. One measures whether a country experienced a downgrade in Year t (1 if yes, 0 if otherwise), while the other measures whether a country experienced a negative rating event more broadly (a downgrade OR a worsened outlook decision) in Year t (1 if yes, 0 if otherwise). Table 1 provides descriptive statistics of these dependent variables for our annual and quarterly data, as well as our independent variables outlined below.

Table 1.

Descriptive Statistics.

	1995-2014 (annual data)		1995-2014 (quarterly data)
	M	SD	M	SD
Dependent variables
Downgrade, S&P’s (1 = yes)	0.078	0.269	0.026	0.159
Downgrade, Moody’s (1 = yes)	0.059	0.235	0.020	0.140
Downgrade, Fitch (1 = yes)	0.060	0.238	0.020	0.141
Downgrade or worsened outlook, S&P’s (1 = yes)	0.135	0.342	0.048	0.213
Downgrade or worsened outlook, Moody’s (1 = yes)	0.085	0.279	0.034	0.180
Downgrade or worsened outlook, Fitch (1 = yes)	0.098	0.298	0.034	0.180
Spread on government bonds	0.374	2.067	0.174	1.977
Independent variables
Left-controlled cabinet (1 = yes)	0.399	0.490	0.391	0.484
Cabinet manifesto score	−0.335	16.489	NA
Government debt (% of GDP)	67.327	39.005	NA
Δ in debt	1.128	5.871	NA
Net government lending (% of GDP)	−1.646	4.929	NA
Δ in government lending	0.136	2.603	NA
Trade balance (% of GDP)	3.241	7.910	NA
Unemployment	7.120	3.821	NA
Inflation	2.073	1.545	2.074	1.642
Election year (1 = year)	0.272	0.445	0.068	0.252
Left incumbent electoral victory (1 = yes)	0.054	0.227	0.014	0.116
Left nonincumbent electoral victory (1 = yes)	0.046	0.209	0.011	0.106
Minority government (1 = yes)	0.233	0.423	0.234	0.421
Coalition majority government (1 = yes)	0.563	0.497	0.564	0.493
Single-party majority government (1 = yes)	0.193	0.395	0.191	0.392
Government expenditure (% of GDP)	44.758	7.072	NA

Source. Standard & Poor’s Sovereign Rating and Country T&C Assessment Histories, European Union’s (EU) AMECO database, OECD, Armingeon, Knöpfel, Weisstanner, and Engler (2014).

S&P = Standard & Poor’s; AMECO = Annual Macroeconomic; OECD = Organisation for Economic Co-Operation and Development.

Our data cover 23 OECD countries¹⁴ from 1995 to 2014. Although the earliest sovereign credit ratings were issued in the mid-1970s, we conduct our analysis from the mid-1990s onward for two reasons. First, one of the “big three,” Fitch, entered the ratings market relatively late, issuing its first ratings in 1994. Moody’s and S&P also extended their sovereign rating portfolio gradually. Although they covered most of the developed world by the late 1980s, some outliers (e.g., Luxembourg) were only rated as late at 1994. As a result, 1994 is the first year for which all sovereigns are rated by the “big three” and, therefore, 1995 is the first year for which ratings changes can be observed. Our second, and more significant reason for starting our investigations in the mid-1990s is that it allows us to observe the “mature phase” of the sovereign ratings market. From the 1970s to the mid-1990s, the nature of the sovereign rating market transformed considerably as the weight of institutional investors increased, more than doubling both as a share of the economy and as a share of claims held (Davis & Steil, 2004, pp. 5-7). This shift coincided with increased economic volatility in developed countries, exemplified by the 1992 European Monetary System (EMS) crisis, arguably increasing the appeal of more conservative rating strategies. Starting our investigations in the mid-1990s allows us not only to work with a relatively constant set of countries, but also to observe CRAs’ behavior under uniform financial market characteristics and economic conditions.

We conduct our panel analysis for this time period using both annual data (results in Tables 2 and 3) and quarterly data (results in Table 4). We chose to complement our annual panel with one using quarterly data, to better capture the effect of within-year events, like elections and changes in government composition.¹⁵ Our results based on annual and quarterly data are largely consonant. In our discussion below, however, we prioritize results from the annual data for two reasons. First, quarterly economic data are less available than annual data. Crucially, fiscal indicators such as gross debt and net lending, which are arguably the single most important determinants of sovereign creditworthiness, are only available on an annual basis in all major macroeconomic databases (OECD, Eurostat, the IMF, etc.). Excluding fiscal data from the analysis poses a significant omitted variable bias problem (especially if fiscal balances are affected by partisan decisions). While the main economic indicators that we use in our annual panel analysis are available on a quarterly basis, some (notably unemployment) demonstrate a higher frequency of missing observations for quarterly data than annual data. Second, our annual data demonstrates more variation both in downgrades and negative ratings decisions, as well as electoral events than quarterly data (see Table 1).

Table 2.

Political and Economic Determinants of Negative Credit Ratings and Bond Spreads (Annual Data).

	DV is downgrade			DV is downgrade and worsened outlook			DV is spreads
	S&P	Moody’s	Fitch	S&P	Moody’s	Fitch	On govt bonds
	1	2	3	4	5	6	7
Partisan controls
Left-controlled executive (1 = yes)	0.071***	0.034*	0.037	0.003	0.019	0.006	−0.148
	(0.009)	(0.077)	(0.200)	(0.927)	(0.353)	(0.853)	(0.218)
Election year (1 = yes)	0.002	0.01	0.017	−0.034	0.029	0.051*	0.127*
	(0.949)	(0.538)	(0.427)	(0.277)	(0.146)	(0.079)	(0.088)
Election win for incumbent left executive (1 = yes)	0.001	−0.028	−0.029	0.011	−0.025	−0.049	−0.015
Election win for incumbent left executive (1 = yes)	(0.977)	(0.353)	(0.432)	(0.824)	(0.409)	(0.313)	(0.926)
Election win for nonincumbent left executive (1 = yes)	0.113**	0.084**	0.046	0.106*	0.058	0.004	−0.213
Election win for nonincumbent left executive (1 = yes)	(0.027)	(0.016)	(0.350)	(0.096)	(0.164)	(0.948)	(0.248)
Economic and fiscal controls
Public debt (% of GDP)	−0.001	0.001	−0.002**	0.001	0.001	0.001	0.012**
	(0.635)	(0.513)	(0.012)	(0.439)	(0.381)	(0.246)	(0.021)
Trade balance (% of GDP)	−0.018***	−0.002	0.003	−0.009*	0.003	−0.009	−0.045*
	(0.000)	(0.559)	(0.589)	(0.069)	(0.438)	(0.242)	(0.084)
Unemployment	0.025*	0.010	0.001	0.010	0.013**	0.000	0.338***
	(0.077)	(0.242)	(0.973)	(0.417)	(0.039)	(0.993)	(0.000)
Δ public debt (% of GDP)	0.010***	0.008***	0.004	0.008**	0.007***	0.006*	−0.027*
	(0.003)	(0.001)	(0.220)	(0.024)	(0.005)	(0.085)	(0.095)
Inflation	0.007	0.031***	−0.003	0.006	0.045***	0.009	0.238***
	(0.663)	(0.007)	(0.844)	(0.705)	(0.000)	(0.670)	(0.000)
Government size (% of GDP)	0.013**	0.019***	0.025***	0.015**	0.016***	0.031***	0.009
	(0.038)	(0.000)	(0.000)	(0.011)	(0.001)	(0.000)	(0.730)
Government type
Minority government (1 = yes)	−0.105**	0.027	0.124**	−0.158***	0.016	0.026	0.516**
	(0.030)	(0.552)	(0.029)	(0.004)	(0.761)	(0.740)	(0.021)
Coalition majority (1 = yes)	0.003	0.085**	0.108***	−0.112**	0.05	0.011	0.606**
	(0.944)	(0.017)	(0.009)	(0.043)	(0.228)	(0.830)	(0.019)
Credit rating controls (for spreads only)
Downgrade by big three (1 = yes)	—	—	—	—	—	—	0.494**
	—	—	—	—	—	—	(0.019)
Worsened outlook by big three (1 = yes)	—	—	—	—	—	—	−0.076
	—	—	—	—	—	—	(0.619)
Improved outlook by big three (1 = yes)	—	—	—	—	—	—	−0.277*
	—	—	—	—	—	—	(0.060)
Upgrade by big three (1 = yes)	—	—	—	—	—	—	0.056
	—	—	—	—	—	—	(0.629)
N	424	424	417	424	424	417	399
R ²	.405	.483	.409	.509	.525	.442	.648
Chi-square (p value)	0	0	0	0	0	0	0

Estimators used were linear probability models (for ratings changes), and OLS (for spreads), for 23 OECD countries from 1995 to 2014. All models include a panel-specific AR1 disturbance. N − 1 country, time dummies, path dependency dummies (for credit rating actions regressions), and constant term included but not shown. The p values are in parentheses (panel corrected standard errors used). DV = dependent variable: S&P = Standard & Poor’s; OLS = ordinary least squares; OECD = Organisation for Economic Co-Operation and Development; AR1 = first-order autoregressive.

*, **, and *** indicate significance on a 90%, 95%, and 99% confidence interval.

Table 3.

Robustness Checks for Partisanship Beta Coefficients.

	Original model	Random effects	No Greece	No United States	Economic variables lagged	1995-2008	No PSAR1	Common AR1
Standard & Poor’s
DV = Downgrade
Left govt (1 = yes)	0.071***	0.046*	0.072***	0.064**	0.079**	0.041*	0.084***	0.085***
New left victory	0.113**	0.106**	0.116**	0.126**	0.155***	0.144***	0.112*	0.100*
DV = Downgrade and worsened outlook
Left govt (1 = yes)	0.003	−0.021	−0.004	0.004	0.021	−0.007	0.013	0.012
New left victory	0.106*	0.115*	0.111*	0.115*	0.125*	0.153***	0.139*	0.122*
Moody’s
DV = Downgrade
Left govt (1 = yes)	0.034*	0.023	0.027	0.037*	0.026	−0.032***	0.033*	0.038*
New left victory	0.084**	0.045	0.065*	0.087**	0.085**	0.015	0.057	0.049
DV = Downgrade and worsened outlook
Left govt (1 = yes)	0.019	0.006	0.015	0.026	0.013	−0.035***	0.031	0.031
New left victory	0.058	0.035	0.029	0.070*	0.069	−0.018	0.001	−0.0005
Fitch
DV = Downgrade
Left govt (1 = yes)	0.037	0.023	0.036	0.044	0.045	−0.011	0.039	0.033
New left victory	0.046	0.036	0.045	0.043	0.043	0.075***	0.027	0.016
DV = Downgrade and worsened outlook
Left govt (1 = yes)	0.006	0.002	0.027	−0.009	0.012	−0.012	0.032	0.023
New left victory	0.004	−0.002	0.019	0.024	0.012	0.081	0.018	0.005
Spreads
Left govt (1 = yes)	−0.148	−0.144	−0.087	NA	−0.228*	0.044	0.032	−0.146
New left victory	−0.213	−0.204	−0.019	—	−0.42*	0.237**	−0.283	−0.171

Model used for the downgrade dependent variables are Models 1 to 3 in Table 2. Model used for the downgrade and worsened outlook dependent variables are Models 4 to 6 in Table 2. Only the beta coefficient is shown. PSAR1 = panel-speciﬁc ﬁrst-order autoregressive; AR1 = first-order autoregressive; DV = dependent variable.

*, ** and *** denote significance at a 90%, 95%, and 99% confidence interval.

Table 4.

Political and Economic Determinants of Negative Credit Ratings and Bond Spreads (Quarterly Data).

	DV is downgrade			DV is downgrade and worsened outlook			DV is spreads
	S&P	Moody’s	Fitch	S&P	Moody’s	Fitch	On govt bonds
	1	2	3	4	5	6	7
Partisan controls
Left-controlled executive (1 = yes)	0.027**	0.003	0.004	−0.004	0.002	−0.009	−0.137*
	(0.031)	(0.777)	(0.776)	(0.830)	(0.854)	(0.597)	(0.075)
Election quarter (1 = yes)	0.000	0.021	0.014	−0.018	−0.007	0.007	0.151**
	(0.995)	(0.153)	(0.379)	(0.414)	(0.700)	(0.705)	(0.012)
Election quarter (t − 1)	0.018	0.030*	−0.004	0.018	−0.002	0.003	0.189**
	(0.360)	(0.071)	(0.820)	(0.457)	(0.931)	(0.904)	(0.011)
Election quarter (t − 2)	0.02	0.015	0.000	0.029	−0.017	0.008	0.134*
	(0.309)	(0.377)	(0.986)	(0.225)	(0.369)	(0.723)	(0.084)
Election quarter (t − 3)	0.019	0.001	−0.016	0.019	−0.012	−0.006	0.032
	(0.313)	(0.970)	(0.377)	(0.421)	(0.542)	(0.790)	(0.662)
Election quarter (t − 4)	0.009	0.001	−0.025	−0.007	−0.008	−0.011	0.015
	(0.584)	(0.958)	(0.118)	(0.756)	(0.663)	(0.586)	(0.807)
Left incumbent election win (1 = yes)	−0.016	−0.017	−0.02	0.023	−0.003	0.013	−0.132
	(0.589)	(0.534)	(0.530)	(0.552)	(0.934)	(0.744)	(0.270)
Left incumbent election win (t − 1)	−0.035	−0.029	−0.002	0.018	0.029	−0.013	−0.133
	(0.284)	(0.324)	(0.959)	(0.678)	(0.396)	(0.758)	(0.346)
Left incumbent election win (t − 2)	−0.028	−0.024	0.022	−0.01	−0.003	0.007	−0.153
	(0.399)	(0.420)	(0.540)	(0.810)	(0.935)	(0.877)	(0.330)
Left incumbent election win (t − 3)	−0.001	−0.001	0.000	0.015	0.026	−0.004	−0.086
	(0.976)	(0.979)	(0.991)	(0.727)	(0.448)	(0.927)	(0.559)
Left incumbent election win (t − 4)	0.017	0.025	0.006	0.038	0.019	0.021	−0.116
	(0.570)	(0.333)	(0.860)	(0.337)	(0.576)	(0.583)	(0.382)
Left nonincumbent election win (1 = yes)	0.013	−0.017	0.053	0.017	0.007	0.088*	−0.187
Left nonincumbent election win (1 = yes)	(0.732)	(0.598)	(0.195)	(0.743)	(0.870)	(0.081)	(0.159)
Left nonincumbent election win (t − 1)	0.004	−0.044	−0.015	0.018	−0.037	0.017	−0.157
	(0.923)	(0.208)	(0.742)	(0.735)	(0.406)	(0.755)	(0.334)
Left nonincumbent election win (t − 2)	0.071	0.086**	0.057	0.114**	0.098**	0.033	−0.195
Left nonincumbent election win (t − 2)	(0.101)	(0.018)	(0.219)	(0.041)	(0.033)	(0.556)	(0.253)
Left nonincumbent election win (t − 3)	−0.052	−0.028	0.023	−0.038	−0.044	0.016	0.048
	(0.226)	(0.438)	(0.625)	(0.490)	(0.340)	(0.776)	(0.765)
Left nonincumbent election win (t − 4)	−0.063	−0.049	−0.018	−0.015	−0.007	0.046	0.023
	(0.111)	(0.144)	(0.667)	(0.769)	(0.864)	(0.371)	(0.861)
Economic controls
Inflation	−0.002	0.009**	−0.001	0.007	0.006	0.000	0.052**
	(0.772)	(0.039)	(0.855)	(0.321)	(0.258)	(0.966)	(0.038)
Real GDP growth	−0.011***	−0.010***	−0.008***	−0.008**	−0.013***	−0.009***	−0.037***
	(0.000)	(0.000)	(0.003)	(0.028)	(0.000)	(0.008)	(0.003)
Government type
Minority government	−0.025	−0.006	0.028	−0.045	−0.009	0.003	−0.254**
	(0.294)	(0.780)	(0.377)	(0.124)	(0.735)	(0.938)	(0.035)
Coalition majority	−0.01	−0.012	0.008	−0.009	−0.002	−0.004	−0.029
	(0.689)	(0.533)	(0.707)	(0.741)	(0.917)	(0.881)	(0.805)
Credit rating actions (for spreads)
Downgrade by big three							0.203**
							(0.042)
Worsened outlook by big three							0.152**
							(0.033)
Improved outlook by big three							0.006
							(0.932)
Upgrade by big three							0.005
							(0.962)
N	1,732	1,732	1,700	1,732	1,732	1,700	1,650
R ²	.201	.27	.161	.197	.229	.168	.401
Chi-square (p value)	0	0	0	0	0	0	0

Estimators used were linear probability models (for ratings changes), and OLS (for spreads), for quarterly data for 23 OECD countries from 1995 to 2014. All models include a panel-specific AR1 disturbance. N − 1 country, time dummies, path dependency dummies (for credit rating action regressions only) and constant term included but not shown. p values are in parentheses (panel corrected standard errors used). t − 1, t − 2, t − 3, and t − 4 indicate a one, two, three, and four quarterly lag, respectively. DV = dependent variable: S&P = Standard & Poor’s; OLS = ordinary least squares; OECD = Organisation for Economic Co-Operation and Development; AR1 = first-order autoregressive.

*, **, and *** indicate significance on a 90%, 95% and 99% confidence interval.

Model Specification and Estimator

We employ a fixed effects, OLS estimator for both rating actions and spreads to keep the results on the two different dependent variables comparable. For rating actions, an OLS model is the equivalent to an LPM as the variable is binary. While logistic regression overcomes several problems with using OLS to estimate binary outcomes, we opt for an LPM in our credit rating decision analysis for two reasons. First, when using a logistic estimator, we suffered quasi-separation problems (results either did not converge or some of our independent variables were dropped) in some of our models because several country and time dummies perfectly predicted zero outcomes (see Carter & Signorino, 2010). Second, related to separation problems, employing a fixed effects model in logistic regression would result in the dropping of panels where there are no downgrades (affecting 10 of the 23 countries in our OECD sample) or negative rating events (affecting two of the 23 countries within our sample). Consequently, if we employed fixed effects for both the spreads’ dependent variable using OLS, and the credit rating decisions using logistic regression, we would effectively be comparing two different samples.

We model credit rating decisions as follows:

\Pr (y_{i, t} = 1) = β_{0} + β_{1} (L e f t_{i, t}) + β_{2} \sum^{​} K_{i, t} + β_{3} \sum^{​} L_{i, t} + β_{4} \sum^{​} M_{i, t} + β_{5} \sum^{​} N_{t} + β_{6} \sum^{​} P_{i} + ε_{i . t},

y_i,t is the binary manifestation of the two rating actions that we describe above in country i at time t. Left_i,t is a binary variable that indicates whether a left party controls the executive in country i at time t: it equals 1 if the prime minister is from a left party, 0 if otherwise. We also use the proportion of cabinet seats occupied by left parties and the manifesto scores of incumbent governments as alternative measurements of partisanship (we present these results in Online Appendix C).¹⁶ Executive partisanship data stem from Armingeon, Knöpfel, Weisstanner, and Engler (2014), while manifesto scores are taken from the Wissenschaftszentrum Berlin für Sozialforschung’s (2016) Manifesto Project Database. For election years, Armingeon and his coauthors weight their partisanship data by the number of days an executive is in office. If a rating change happens during an election year (or election quarter), however, we code executive partisanship solely as the executive in power when the rating change happens.

∑K_i,t is a vector of electoral controls and their interaction with the left executive dummy, which enables us to test our two electoral hypotheses; whether CRAs are more likely to impose negative ratings when a new left government comes into office (Hypothesis 1a) and when a left incumbent is reelected (Hypothesis 1b). These controls include an election year/quarter dummy (1 if the year/quarter is an election year, 0 if not), an interaction term between an election year/quarter and whether a left incumbent won control of the executive (1 if yes, 0 if no), and an interaction term between an election year/quarter and whether a left nonincumbent won control of the executive (1 if yes, 0 if no). Nonelection years/quarters serve as the baseline category of all of these dummies. The left incumbent and nonincumbent interaction terms highlight the additional (or conditional, if the electoral dummy is nonsignificant) negative rating penalty associated with an election year/quarter. Hence, the beta coefficient on the election dummy is indicative of a right executive electoral victory; the betas on the left incumbent/nonincumbent interaction terms indicate the added electoral effect for left government victories. Election data also stem from Armingeon et al. (2014).

∑L_i,t is a vector of economic and fiscal controls that CRAs claim impacts their ratings within their methodologies for assessing countries’ solvency. These include the debt to GDP ratio, net government lending as a percentage of GDP (positive/negative values indicate an annual fiscal surplus/deficit), the trade balance (positive/negative values indicate an annual trade surplus/deficit) as a percentage of GDP, unemployment, and inflation. We also include year-on-year differences in debt and net lending as CRAs are concerned not only about debt and deficit levels, but also sudden negative changes in them. To test Mosley’s “room to move” hypothesis, we control for the size of the government (government spending as a percentage of GDP).¹⁷ Because we were unable to obtain fiscal and government spending indicators or trade balance data from the OECD on a quarterly basis, these variables are absent from our quarterly analysis.

Multicollinearity problems within our economic controls prompted us to exclude real GDP growth as a control from our annual panel data set (even though all agencies cite it as an important variable) as it shares significant correlations with our fiscal variables. However, we incorporate real GDP growth into our quarterly panel data set because fiscal controls were not available (moreover, we use real GDP growth in the place of unemployment in the quarterly panel because quarterly real GDP data are complete for all our countries, whereas quarterly unemployment succumbs to missing data problems¹⁸). We stress that including GDP growth in our models for the annual data does not alter the results of our political variables.¹⁹ To avoid multicollinearity problems between the government debt and net public lending variables, we included these controls in separate regressions (Tables 2 and 3 incorporate public debt, while Online Appendix B incorporates net government lending instead of debt for the same models). Debt,²⁰ unemployment, and trade balance data were taken from the European Union’s (EU) Annual Macroeconomic (AMECO; European Commission’s Directorate General for Economic and Financial Affairs, 2015) database, while inflation, net government lending,²¹ and government expenditure²² data were taken from the OECD.

∑M_i,t is a vector of government type dummies (minority, coalition majority, or single-party majority, the latter serves as the baseline category). Government type data also stem from Armingeon et al. (2014). In addition, we incorporated a number of political institution controls that may influence debt ratings: capital account liberalization, a country’s central bank independence index, and the World Bank political stability and absence of violence index (higher values indicate greater political stability). These controls could only be included in our annual results as they are not available on a quarterly basis. We also included a control variable for whether a country is an EU member in time t, given previous scholarly findings that membership regularizes market expectations about a country’s future policy choices (Gray, 2009, 2013), as well as whether a country was an Economic and Monetary Union (EMU) member in time t as the creation of the Euro eliminated currency risk for countries with the common currency. To preserve space, we do not present these results below, but they are provided in Online Appendix D, where we offer a fuller interpretation of their impacts on negative ratings and spreads. We discuss how the inclusion of these variables affects our results below.

∑N_t is a vector of (n − 1) time dummies (to control for omitted time shocks that may affect ratings), as well as four-path dependency dummies for CRAs: whether a CRA bestowed a downgrade upon country i in the past 3 years (1 = yes, 0 if no), a worsened outlook in the past 3 years (1 = yes), an improved outlook in the past 3 years (1 = yes), and an upgrade in the past 3 years (1 = yes). ∑P_i is a vector of country-specific fixed effects, which account for omitted variables that are constant over time but differ across countries (we test the sensitivity of our results to random effects in Table 3). Finally, Wooldridge statistics indicated that first-order serial correlation was present in the majority of our models. Consequently, we incorporated a panel-specific first-order autoregressive (AR[1]) disturbance, in addition to panel corrected standard errors to correct for heteroskedasticity within panels (see Beck & Katz, 1995; Table 3 also provides results if a panel-specific AR(1) disturbance is excluded, as well as if a common AR(1) disturbance is used).

In testing our auxiliary hypotheses about spreads, we use the following model specification:

i_{i, t} - i_{U S, t} = β_{0} + β_{1} (L e f t_{i, t}) + β_{2} \sum^{​} C R A_{i, t} + β_{3} \sum^{​} K_{i, t} + β_{4} \sum^{​} L_{i, t} + β_{5} \sum^{​} M_{i, t} + β_{6} \sum^{​} N_{t} + β_{7} \sum^{​} P_{i} + ε_{i . t},

i_i,t – i_US,t is the spread of country i’s nominal interest rate on long-term government bonds vis-à-vis the United States’s nominal interest rate on 10-year T-bills at time t (higher values indicate a country’s long-term government bond is more risky relative to the United States’s). Though we use the spread level for our regressions below, we stress that our results are similar if we use changes in spreads as the dependent variable (partisanship does not affect spread changes, but negative ratings do). Interest rate data for long-term government bonds stem from the OECD.

To test the effects of rating changes on spreads (Hypothesis 3), we include CRAs’ decisions as an independent variable in modeling spreads. ∑CRA_i,t is vector of rating changes bestowed on country i in year t from at least one of the big three (our results are identical if we examine ratings changes of each CRA separately). This vector includes four rating decisions (all coded as binary if one or more of the big three bestow them in a given year): downgrades, worsened outlooks, improved outlooks, and upgrades. All other partisan, electoral, economic, fiscal, and institutional explanatory variables in our spreads models are the same as in the model for rating actions, as are the time and country fixed-effect dummies and error structures (however, the CRA path dependency dummies in vector ∑N_t are not included in the spreads models because these actions are incorporated into vector ∑CRA_i,t).

Results

Table 2 presents our results for the annual panel data for the big three’s rating decisions and for bond spreads when debt is used as the primary fiscal control. Online Appendix B provides our results when annual public lending is used as the primary fiscal control. The dependent variable in Models 1 to 3 in Table 2 and Online Appendix B are downgrades for S&P, Moody’s, and Fitch, while the dependent variable in Models 4 to 6 are negative rating actions (downgrades or worsened outlooks) for S&P, Moody’s, and Fitch. The dependent variable in Model 7 in Table 2 and Online Appendix B is government bond spreads. Online Appendix C provides our results when the proportion of cabinet seats occupied by left parties (Models 1 to 4) and manifesto scores (Models 5 to 8) are used to measure partisanship. Tables D1, D2, and D3 in Online Appendix D provide our results for the rating decisions of S&P, Moody’s, and Fitch, respectively, when alternative political institutional controls are included (capital account openness, central bank independence, political stability, EU, and EMU membership) while Table D4 provides these results for bond spreads. Results for credit rating decisions can be interpreted in terms of changes in probabilities; that is, if the beta coefficient is 0.03, this translates to an increased probability of a downgrade (or worsened outlook) by 3 percentage points.

Government Partisanship

The results for our left executive dummy in Models 1 to 3 in Table 2 and Online Appendix A indicate that two of the big three CRAs (S&P and Moody’s) were more likely to downgrade left-controlled executives. Left governments were 7.1% and 3.4% more likely to witness downgrades from S&P’s and Moody’s, respectively than their right counterparts. The magnitude of these downgrade penalties is slightly larger for S&P in Online Appendix B when net government lending is used as a control, but becomes insignificant for Moody’s. Executive partisanship’s impact on downgrades also holds when using alternative measures for left partisanship, examined in Online Appendix C. For all three CRAs, the higher the proportion of cabinet seats occupied by left parties, the more likely a downgrade (cabinets that are solely occupied by left parties exhibit downgrade likelihoods of 10% from all three CRAs). For Moody’s and Fitch, higher manifesto scores (indicative of more right-wing executives) are associated with lower probabilities of downgrades. A one standard deviation increase in an incumbent’s manifesto score (16.489) is associated with a 4.9% and 3.3% lower likelihood of a downgrade from Moody’s and Fitch, respectively.

The effect of government partisanship is very robust to different model specifications for S&P. Table 3 provides the beta coefficients, and their associated significance levels, for the left-controlled executive dummy (and the nonincumbent left election victory dummy, which we discuss below) for the following alternative specifications of our models in Table 2: (a) if random effects are used, (b) if Greece is excluded (downgrade and worsened outlooks were relatively numerous under its left executives from 2009 onward), (c) if the United States is excluded (we do this because our spreads regressions omit the United States, given that its bond yield is used as the baseline for spreads), (d) if our economic variables are lagged, (e) if our time period spans 1999 to 2008 (excluding the European debt crisis), (f) if the panel-specific AR(1) disturbance is excluded, and (g) if a common AR(1) disturbance is used. S&P’s (significant) left executive downgrade penalty is robust to all respecifications in Table 3. Moreover, S&P’s results for downgrade premiums associated with left executives is robust when central bank independence, capital account liberalization, political stability, EU, and EMU membership are controlled for (see Online Appendix Table D1) and when we used quarterly data (see Table 4 below).

The effect of government partisanship is not as robust for Moody’s. In Table 3, partisanship holds its significance and sign for Moody’s if the United States is excluded and if a panel-speciﬁc ﬁrst-order autoregressive (PSAR[1]) or common AR(1) error term is used. However, Moody’s left executive downgrade penalties lack significance for the other specifications in Table 3 and if other political controls are included in our models (see Models 1 to 5 in Online Appendix Table D2). Intriguingly, left executives are less likely to witness downgrades from Moody’s between 1995 and 2008, indicating that this CRA’s left executive bias in downgrade decisions is moved solely by the inclusion of the years of the debt crisis within our models. We provide an explanation for this finding in our conclusion below.

In contrast to ratings, spreads on government bonds are not affected by the partisanship of the executive (the left executive dummy is nonsignificant for spreads in Table 2, Online Appendices B and C, and all model specifications in Online Appendix Table D4). We only find partisanship effects under two specifications (when net government lending is used as a fiscal control, see Online Appendix B, or when all economic variables are lagged, see fifth column in Table 3). However, in these cases, left executives are associated with lower bond spreads than right executives.

Elections and Partisanship

Elections, on their own, do not affect the likelihood of downgrades or negative ratings relative to nonelection years, as indicated by the insignificant election year dummy for all CRAs. Election results, on the contrary, had significant effects. The reelection of incumbent left governments have no effect on the likelihood of negative rating action by any of the big three, as indicated by the insignificance of the left incumbent election win dummy. However, S&P and Moody’s do mind left nonincumbent electoral victories in their downgrade and negative rating decision, while Fitch only demonstrates a significantly greater likelihood of bestowing downgrades after nonincumbent left electoral victories in the precrisis 2000s (see results in Table 3 for the 1999-2008 panel). We explain this finding in our conclusion. In other words, the election of right executives (incorporated into the election dummy) and left incumbents exhibit no influence on adverse rating changes. CRAs only appear critical of the election of new left executives.

S&P is 11.3% more likely to issue a downgrade, and 10.6% more likely to initiate negative action (downgrade or worsened outlook) if a new left government wins an election: this penalty is in addition to the 7.1% downgrade penalty applied to left executives in general (see Models 1 and 4, Table 2). These S&P ratings penalties for nonincumbent left electoral victories are highly robust. With the exception of Model 4 in Online Appendix Table D1, the ratings penalty (both for downgrades on their own, and negative ratings more broadly) that S&P associates with nonincumbent left electoral victories remains significant for all alternative model specifications in Table 3, as well as all model specifications in Online Appendices B, C, and D. The ratings penalty that Moody’s bestows upon nonincumbent left electoral victories is not as extreme or robust as S&P’s reactions. Moody’s is 8.4% more likely to issue a downgrade upon the election of nonincumbent left executives but these electoral victories do not significantly affect negative ratings more broadly (see Models 2 and 5 in Table 2). Moody’s downgrade electoral penalty remains significant if Greece and the United States are excluded from the sample or if lagged economic variables are used as controls (see Table 3), if net lending rather than public debt is used as the primary fiscal control (see Online Appendix B), using alternative measures of left partisanship (Online Appendix C), and if capital account openness, central bank independence, and political stability²³ are included in our models. Outside of these specifications, however, the downgrade penalty Moody’s bestows on nonincumbent left electoral victories is nonsignificant.

S&P’s and Moody’s downgrade and negative ratings penalties are also present when our regressions are run with quarterly data (results presented in Table 4). As mentioned above, we were limited in terms of the types of controls we could include in our quarterly models as all of our fiscal and government expenditure data were unavailable on a quarterly basis (similarly, we also omitted unemployment as a control in our quarterly models, due to widespread missing observations for some of our countries, but we included real GDP growth, for which every country had complete data). We structure the models for the quarterly data in exactly the same fashion as our annual data, with one exception. We provide a series of four quarterly lags for our election variables because elections may not have immediate results on ratings changes (due to the fact that some countries may have transitional or coalition-formation periods between an election and when its victor formally enters office, and/or that CRAs may take some time to observe a newly elected left executive once it has entered office). Lagged nonincumbent left party electoral victories had a significant impact on downgrades for Moody’s (Model 2 in Table 4; significance was just under the 90% confidence level for S&P, p value = 0.101) and negative ratings more broadly (Models 4 to 6) for both S&P and Moody’s. More specifically, electoral victories of new left governments produce a higher probability of a downgrade (8.6% for Moody’s), and a higher probability of a negative ratings change (11.4% more likely for S&P and 9.8% for Moody’s), two quarters after they occur (we highlight this result in bold in Table 4). In other words, ratings penalties associated with new left government victories do not emerge immediately (as indicated by the insignificant present value of this variable), but rather emerge in the 6 months afterwards. Likewise, Fitch demonstrates an immediate worsened outlook ratings penalty for a nonincumbent left executive electoral victory in the quarterly data. Fitch is 8.8% more likely to bestow a negative ratings change in the same quarter that new left executives win an election (see Model 6, Table 4).

Just as in the case of government partisanship, we find that elections affect bond spreads differently from ratings. Elections do matter for bond spreads, but not their outcomes. Results in Table 2, Online Appendices B, C, and D and Table 4 demonstrate that election years are associated with increased bond spreads (Table 4, Model 6 indicates that these effects are spread over three quarters). However, the electoral victories of left incumbent governments or new left governments do not have a significant impact on bond yields (when they do, as in Tables 3 and Online Appendix Table D4, left electoral victories are associated with smaller, rather than larger, Only in the years prior to the debt crisis (1995-2008), are non-incumbent left electoral victories associated with higher spreads). The lack of robust effects for electoral outcomes suggests that markets display greater partisan neutrality than CRAs during elections: Although markets appear to price elections into bond yield spreads for our countries, they do not price left electoral victories (either incumbents or nonincumbents) into them.

In regard to other economic and institutional controls, downgrades and worsened outlooks were more likely by the big three during periods of high sudden increases in public debt (Δ debt), high levels of unemployment (for S&P and Moody’s), higher trade deficits (for S&P), higher sudden increases in government deficits (for Fitch), higher inflation (for Moody’s), and higher levels of government expenditure as a percentage of GDP (for all of the big three). Compared with single-party majority governments, minority governments were less likely to encounter downgrades or negative ratings from S&P, but were more likely to encounter these decisions from Fitch, while coalition majority governments were more likely to encounter downgrades from Moody’s and Fitch (but were less likely to encounter negative ratings from S&P). Results for bond yields align with some of these results. Bond spreads are significantly larger for higher levels of public debt, unemployment, inflation, and trade deficits. Spreads also increase under minority and coalition majority governments (compared with single-party majority administrations). Importantly, spreads increase when downgrades are initiated by the big three. Moreover, the positive impact of downgrades on spreads remains significant for all of our specifications in Tables 2 to 4, and Online Appendices B, C, and Table D4.

In sum, whereas CRAs and the markets are aligned in their responses to variation in economic, fiscal, and institutional variables, their approach to politics deviates strongly. S&P and Moody’s are more likely to initiate negative rating actions against left governments and even more so when a new left executive comes into office. Markets, however, are moved by the uncertainty created by elections, but do not directly respond to variation in government partisanship. This latter result confirms Breen and McMenamin’s (2013) findings that partisanship does not systematically affect bond yields over time. More importantly, because spreads respond to downgrades, the tendency of CRAs to downgrade left governments more often than right ones indirectly feeds partisanship into market outcomes.

Discussion and Conclusion

Our results point to an intriguing dynamic in the market–politics nexus. CRAs have systematically discriminated against left governments in the past two decades. S&P, the agency with the most influential and largest number of sovereign ratings (IMF, 2010, pp. 87, 115), has been consistently more likely to downgrade left executives than their right-wing counterparts, and the probability of negative rating actions increased even further when nonincumbent left executives first enter office. Moody’s and Fitch have also penalized left governments, but not as consistently as S&P. While the likelihood of downgrades was correlated with the ideological composition of the executive (and the partisanship of the prime minister) for Moody’s, the partisan stance of the prime minister lacks importance for Fitch. Both Moody’s and Fitch applied downgrade penalties against newly elected nonincumbent left governments, but Fitch only did so in the years preceding the crisis (see Table 3). We found that partisan credit rating cues were incorporated into the pricing of sovereign debt, as markets systematically acted upon downgrades, even though bond prices otherwise did not betray partisan bias. This suggests that CRAs instigate systematic partisan discrimination in sovereign credit markets.

As CRAs do not acknowledge or justify the influence of partisan considerations on rating decisions in their communications, we hypothesize that discrimination against left governments is a strategy that conservative CRAs can use to minimize downside policy and market risk while keeping ratings stable, in an effort to make them attractive as third-party risk indicators. The pattern of variation in partisan discrimination across CRAs and across time lends some support to this hypothesis. The finding that S&P discriminates most systematically against the left is consistent with S&P’s track record as the most cautious in its risk assessment of the big three—leading the other two 74% of the time in downgrading developed economies between 2005 and 2010 (IMF, 2010, p. 115)—as well as with its professed strategy of preemptively incorporating all potential future risks into ratings. Moody’s less stringent partisan discrimination before the crisis is consistent with a less conservative attitude, reflected in its role as a follower to the other two CRAs in downgrades.²⁴ The appearance of the left electoral penalty in Moody’s downgrades to new left governments only with the inclusion of the years of the crisis likely betrays the adoption of a more conservative approach as a reaction to rating failures during the crisis. In a similar vein, but with opposite effect, Fitch’s abandonment of discrimination against electoral victories of new left governments after 2008 can also be attributed to its effort to regain credibility after the crisis. In 2011, Fitch transformed its sovereign rating methodology, putting it onto a predominantly quantitative footing and making it fully transparent by publishing the models that form the basis of its rating reports,²⁵ which left less room for the incorporation of intangible factors like partisanship.

Our findings about partisan discrimination in credit ratings have several important implications for the role of politics in sovereign debt markets. First, the presence of systematically more adverse market conditions for the left suggest that the “golden straightjacket” exists: markets selectively penalize certain developments in domestic politics. Second, and more surprisingly, the “golden straightjacket” effect does not seem to originate from investors, but from CRAs that provide partisan-discriminated signals to investors, which are then translated into changes in spreads on government debt. Third, the use of credit ratings as third-party risk indicators by institutional investors and public regulators proves to be problematic from more than one perspective. The anomaly of placing unappointed, unelected, and unsupervised private actors at the heart of prudential frameworks has already been pointed out (e.g., Sinclair, 2008). However, our findings suggest that CRAs efforts to maintain this position motivates them to produce systematically distorted assessments of creditworthiness that generate adverse conditions for governments on the left of the ideological spectrum.

Footnotes

Acknowledgements

We thank Erik Jones, Waltraud Schelkle, panel participants at the 2015 Council of European Studies conference, three anonymous reviewers, and Comparative Political Studies’s (CPS) editors for their helpful comments.

Authors’ Note

Any errors lie solely with the authors.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biographies

Zsófia Barta (PhD, London School of Economics) is an assistant professor of Comparative Public Policy at the University at Albany. Her main research interest is the politics of public debt. She is the author of In the Red: The Politics of Public Debt Accumulation in Developed Countries (forthcoming, University of Michigan Press).

Alison Johnston (PhD, London School of Economics) is an associate professor of political science at Oregon State University. Her main research interests are the domestic causes of international financial crises, labor market and welfare state politics, and the political economy of European economic and monetary integration. She is the author of From Convergence to Crisis: Labor Markets and the Instability of the Euro (Cornell University Press). Her recent publications include “Global Finance, Labor Politics and the Political Economy of Housing Prices” (forthcoming, Politics & Society), “The Electoral Effects of General Strikes in Western Europe” (2016, Comparative Politics), and “European Monetary Integration and the Incompatibility of National Varieties of Capitalism” (2016, Journal of Common Market Studies).

References

Afonso

(2003). Understanding the determinants of sovereign debt ratings. Journal of Economics and Finance, 27, 56-74.

Afonso

Arghyrou

M. G.

Kontonikas

(2015, April). The determinants of sovereign bond yield spreads in the EMU (ECB Working Paper No. 1781). Frankfurt am Main, Germany: European Central Bank.

Afonso

Gomes

Rother

(2007, January). What “hides” behind sovereign debt ratings? (ECB Working Paper Series No. 711). Frankfurt am Main, Germany: European Central Bank.

Archer

C. C.

Biglaiser

DeRouen

(2007). Sovereign bonds and the “democratic advantage”: Does regime type affect credit rating agency ratings in the developing world? International Organization, 61, 341-365.

Armingeon

Knöpfel

Weisstanner

Engler

(2014). Comparative political data set I 1960-2012. Bern, Switzerland: Institute of Political Science, University of Bern.

Bawn

Rosenbluth

(2006). Short versus long coalitions: Electoral accountability and the size of the public sector. American Journal of Political Science, 50, 251-265.

Beaulieu

Cox

G. W.

Saiegh

(2012). Sovereign debt and regime type: Reconsidering the democratic advantage. International Organization, 66, 709-738.

Bechtel

M. M.

(2009). The political sources of systematic investment risk: Lessons from a consensus democracy. The Journal of Politics, 71, 661-677.

Beck

Katz

(1995). What to do (and not to do) with time-series cross-section data. American Political Science Review, 89, 634-648.

10.

Becker

Milbourn

(2010, September). How did increased competition affect credit ratings? (NBER Working Paper 16404). Cambridge, MA: National Bureau of Economic Research.

11.

Bernhard

Leblang

(2002). Democratic processes, political risk, and foreign exchange markets. American Journal of Political Science, 46, 316-333.

12.

Biglaiser

Staats

J. L.

(2012). Finding the “democratic advantage” in sovereign bond ratings: The importance of strong courts and the rule of law. International Organization, 66, 515-535.

13.

Block

Vaaler

P. M.

(2004). The price of democracy: Sovereign risk ratings, bond spreads and political business cycles in developing countries. Journal of International Money and Finance, 23, 917-946.

14.

Boix

(2000). Partisan governments, the international economy, and macroeconomic policies in advanced nations, 1960–93. World Politics, 53, 38-73.

15.

Breen

McMenamin

(2013). Political institutions, credible commitment, and sovereign debt in advanced economies. International Studies Quarterly, 57, 842-854.

16.

Brooks

Faff

R. W.

Hillier

(2004). The national market impact of sovereign rating changes. Journal of Banking & Finance, 28, 233-250.

17.

Cantor

Mann

(2007). Analyzing the tradeoff between ratings accuracy and stability. New York, NY: Moody’s Investor Service.

18.

Cantor

Packer

(1996). Determinants and impact of sovereign credit rating. Economic Policy Review, 2(2), 37-53.

19.

Carter

Signorino

(2010). Back to the future: Modeling time dependence in binary data. Political Analysis, 18, 271-292.

20.

Chinn

M. D.

Ito

(2006). What matters for financial development? Capital controls, institutions and interactions. Journal of Development Economics, 81, 163-192.

21.

Cusack

T. R.

(1997). Partisan politics and public finance: Changes in public spending in the industrialized democracies, 1955–1989. Public Choice, 91, 375-395.

22.

Davis

P. E.

Steil

(2004). Institutional investors. Cambridge MA: MIT Press.

23.

de Haan

Amtenbrink

. (2011, January). Credit rating agencies (Working Paper No. 278). Amsterdam, The Netherlands: De Nederlandsche Bank.

24.

de Haan

Sturm

J. E

. (1994). Political and institutional determinants of fiscal policy in the European Community. Public Choice, 80, 157-172.

25.

European Commission’s Directorate General for Economic and Financial Affairs. (2015). Annual macro-economic database (AMECO). Retrieved from http://ec.europa.eu/economy_finance/ameco/user/serie/SelectSerie.cfm

26.

Fowler

J. H.

(2006). Elections and markets: The effect of partisanship, policy risk, and electoral margins on the economy. Journal of Politics, 68, 89-103.

27.

Garrett

Lange

(1991). Political responses to interdependence: What’s “left” for the left? International Organization, 45, 539-564.

28.

Gray

(2009). International organization as a seal of approval: European union accession and investor risk. American Journal of Political Science, 53, 931-949.

29.

Gray

(2013). The company states keep: International economic organization and sovereign risk in emerging markets. Cambridge, UK: Cambridge University Press.

30.

Greene

(2003). Econometric analysis (5th ed.). Prentice Hall, NJ: Prentice Hall.

31.

Hall

(2010). Why does the economy fall to pieces after financial crisis? Journal of Economic Perspectives, 24(4), 3-20.

32.

Haque

N. U.

Mathieson

D. J.

Mark

N. C.

(1998, April). The relative importance of political and economic variables in creditworthiness ratings (IMF Working Paper No. WP/98/46.9846). Washington, DC: International Monetary Fund.

33.

Hibbs

D. A.

(1977). Political parties and macroeconomic policy. American Political Science Review, 71, 1467-1487.

34.

Huang

Frangakis

Atkinson

Willke

Leite

Vogel

W. B.

A. W.

(2008). Addressing ceiling effects in health status measures: A comparison of techniques applied to measures for people with HIV disease. Health Services Research, 43, 327-339.

35.

Hübscher

(2016). The politics of fiscal consolidation revisited. Journal of Public Policy, 36, 573-601.

36.

International Monetary Fund. (2010). The uses and abuses of sovereign credit ratings. In Global financial stability report (pp. 85-123). Washington, DC: Author.

37.

Mierau

J. O.

Jong-A-Pin

De Haan

(2007). Do political variables affect fiscal policy adjustment decisions? New empirical evidence. Public Choice, 133, 297-319.

38.

Mosley

(2000). Room to move: International financial markets and national welfare states. International Organization, 54, 737-773.

39.

Mosley

(2004). Government–financial market relations after EMU: New currency, new constraints? European Union Politics, 5, 181-209.

40.

Mosley

(2005). Globalisation and the state: Still room to move? New Political Economy, 10, 355-362.

41.

Reinhart

Rogoff

(2008, March). This time is different: A panoramic view of eight centuries of financial crises (Working Paper No. 13882). Cambridge, MA: National Bureau of Economic Research. Retrieved from http://www.nber.org/papers/w13882

42.

Rodrik

(2000). How far will international economic integration go? Journal of Economic Perspectives, 14, 177-186.

43.

Sattler

(2013). Do markets punish left governments? Journal of Politics, 75, 343-356.

44.

Sinclair

T. J.

(2008). The new masters of capital. Ithaca, NY: Cornell University Press.

45.

Streeck

(2014). Buying time: The delayed crisis of democratic capitalism. London and Brooklyn: Verso Books.

46.

Vaaler

P. M.

Schrage

B. N.

Block

S. A.

(2005). Counting the investor vote: Political business cycle effects on sovereign bond spreads in developing countries. Journal of International Business Studies, 36, 62-88.

47.

Vaaler

P. M.

Schrage

B. N.

Block

S. A.

(2006). Elections, opportunism, partisanship and sovereign ratings in developing countries. Review of Development Economics, 10, 154-170.

48.

Wang

Zhang

McArdle

J. J.

Salthouse

T. A.

(2009). Investigating ceiling effects in longitudinal data analysis. Multivariate Behavioral Research, 43, 476-496.

49.

Wissenschaftszentrum Berlin für Sozialforschung. (2016). Manifesto project database. Available from https://manifestoproject.wzb.eu/

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.98 MB