Econometrics
There is More Behind Government Spending Than You Think by: Chen Yeh The aftermath of the recent credit crisis is starting to show its signs: after a bunch of bailouts and fiscal investments, many governments over the world are forced to cut their spending. However some governments seem to cut down their budgets more drastically than others. While the magnitude of these economic measures differ from country to country, economists wonder which countries’ spending patterns are similar and in particular why they are similar. In traditional macroeconomics one can find models that partially address these questions, however Persson and Tabellini (American Economic Review, 2004) adopt a different approach by combining views from political science and economics. Their empirical paper is based on sophisticated arguments and addresses determinants of government expenditure that are not straightforward.
Introduction Traditional models of government expenditure often involve economic variables, however the field of political economy also considers institutional factors. Persson and Tabellini (2004, henceforth P&T) consider the effect of constitutional rules on fiscal policy outcomes. These constitutional rules consist of electoral rules and forms of government. In most democracies, the electoral rule is either proportional or majoritarian. When a country adopts a proportional voting rule, the distribution of votes closely matches the distribution of seats in the representative democracy, whereas the majoritarian rule is equivalent to “the winner takes all” system.1 For forms of government, P&T consider presidential and parliamentary ones. In their definition of presidential countries, the chief executive or cabinet (regarding issues of economic policy) is not accountable to the legislature through a vote of confidence, i.e. its power is independent of any political institution. As a result, a country is called parliamentary where this vote of confidence is actually required.
In their paper, P&T are trying to address a simple question, namely how do electoral rules and forms of government influence fiscal policy? The authors empirically test a series of claims, hypothesized by both political scientists and economists. Their basic setup is the simple, yet trustworthy, technique of ordinary least squares (OLS). They also relax the necessary OLS assumptions and perform some robustness tests to check whether their empirical results are reliable. It seems that standard OLS regressions confirm earlier theoretical work, however their other results are not that strong, but do give interesting insights regarding constitutional rules and fiscal policy outcomes.
The interaction between constitutional rules and fiscal policy outcomes: some theory Although P&T’s contribution is mainly empirical, the foundation of their econometric framework is firmly motivated by theoretical models. A stream of economic models predict that proportional electoral systems and
The proportional voting rule is for example used in the Netherlands for parliamentary elections, whereas the majoritarian voting rule is adopted in the United States during their elections to vote for a president. 1
In this issue of AENORM, we continue to present a series of articles. These series contain summaries of articles which have been of great importance in economics or have caused considerable attention, be it in a positive sense or a controversial way. Reading papers from scientific journals can be quite a demanding task for the beginning economist or econometrician. By summarizing the selected articles in an understanding way, the AENORM sets its goal to reach these students in particular and introduce them into the world of economic academics. For questions or criticism, feel free to contact the AENORM editorial board at
[email protected]
14
AENORM
vol. 18 (70)
December 2010
Econometrics
parliamentary regimes should be associated with more public goods, a bigger and broader program of welfare spending and a larger overall size of government.2 While these models are all somewhat different in nature, they do reach a similar conclusion: under proportional elections, the composition of public spending is more biased towards large groups in the population. To satisfy the needs of these large groups, a higher level of government spending is induced when compared to the majoritarian voting rule. The main arguments are twofold. The first reason is district magnitude (how large a share of the legislature is elected in a typical district): political parties have strong incentives to seek support of broad and large parts of the population under proportional elections as legislatures are elected in mainly large districts. The other argument is based on the electoral formula (how vote shares are converted to seat shares in the legislature). In a majoritarian system, a political candidate usually only needs 50 percent of the district vote. Moreover the politician would actually suffice with 50 percent of those particular votes, which implies that theoretically this candidate could win the entire election with just 25 percent of the total amount of votes. Under a proportional system, 50 percent of the national vote is needed to win the elections. Thus when facing a proportional voting rule, politicians are once again induced to target larger segments of the population. Even though the effect of forms of government on fiscal policy has not received much attention, there are a few formal studies. As mentioned earlier, the key difference between presidential and parliamentary regimes is the required confidence vote. Diermeier and Feddersen (1998) argue that confidence requirements induce more legislative cohesion: when the chief executive receives the support of a stable majority of legislators, they vote together on legislation, pursuing the joint interest of its represented voters. This means that government spending is providing benefits to a majority of voters. Moreover, these voters become the residual claimant on additional revenue and therefore prefer high taxes and spending.3 By contrast, in presidential regimes there are often powerful minorities (typically the constituency of powerful officeholders) and none of them are residual claimants, which implies that they do pay (relatively) high costs but do not reap the benefits. Thus they resist high government spending.
Econometric methodology: the average treatment effect and assumptions Constitutional reforms have been very rare, i.e. time
variation is minimal and thus little can be gained by estimating panel data. Hence their conclusions about the effect of constitutions on policy outcomes must be identified from cross-country regressions. P&T therefore use a cross-section of 80 democracies in the 1990’s. The main problem however is that constitution selection is not random: countries with different constitutions also differ in many other aspects. Thus how are the effects of constitutional rules on fiscal policy outcomes isolated from other observable and non-observable policy determinants? P&T’s econometric model consists of two parts. The first deals with the constitution selection of country i Si; thus what electoral rule (Si = 1 when majoritarian and 0 otherwise) or form of government (Si = 1 when presidential and 0 otherwise) a country i uses. To simplify their model, these two aspects of constitutional rules are estimated separately with the following equation:
⎧1 if G ( X i ) + ei > 0 Si = ⎨ o/w ⎩0 where Xi is a vector of observables, including colonial origin and geographic location.4 Countries have changed their constitutional rules very few times in the last 40 years (coined constitutional inertia) and political scientists suggest that this information on constitutional history can be exploited to explain cross-country variation. Thus three time dummies are also included in Xi that indicate the origin of the current constitution.5 The second part of their framework determines the fiscal policy outcome Yi, given by6 = Yi F ( Si , Z i ) + ui Thus P&T’s goal is to estimate the effect on fiscal policy of a hypothetical shift from Si = 0 to Si = 1, the so-called average treatment effect of constitutional reform. When estimating this framework with OLS, two assumptions are imposed. Conditional independence implies that the error terms in both equations (i.e. ei and ui) are uncorrelated and linearity demands that F(.) is linear with constant coefficients. However these assumptions might be restrictive and not realistic. Conditional independence is a strong assumption given the non-random distribution of constitution since historical variables determining constitution could also influence policy outcomes. Naturally we do not have a problem when all common historical determinants of government policy are included in the constitution selection equation. However how do
Persson and Tabellini (1999), Milesi-Ferreti et al. (2002) and Persson et al. (2000) are recent examples of the broad theoretical literature about the link between constitutional rules and fiscal policy. 3 A residual claimant is entitled to the net benefits, i.e. the amount of surplus that is left after deducting costs from income. 4 P&T also rely on other cultural and geographic variables such as: distance from the equator, percentage of population with English or a European language as the mother tongue, ethno-linguistic fractionalization and population size. 5 These dummies include the periods of pre 1920, 1921 – 1950 and 1951 – 1980. The period after 1980 is taken as the default, i.e. equals zero when a country’s constitution originates from this period. 6 Several regressions are estimated, each using a different measure for size of government. These include central government spending (cgexp), central government revenues (cgrev) and government deficit (dft). All these are ratios of GDP. 2
AENORM
vol. 18 (70)
December 2010
15
Blijft het bij talent...
SNS REAAL
SNS REAAL is een innovatieve retailbank-verzekeraar met een balanstotaal van ruim 129 miljard euro en zo’n 7.500 medewerkers (fte). SNS REAAL bestaat uit een aantal sterke merken waar gezonde ambitie en plezier in het werk hand in hand gaan. Denk bijvoorbeeld aan SNS Bank, REAAL, Zwitserleven, ASN Bank en SNS Property Finance. Wij groeien omdat we nanciële zaken op eigen wijze aanpakken. De klant staat hierbij altijd centraal. We willen hem of haar aangenaam verrassen en zoeken daarom professionals die geloven dat zaken anders kunnen en anders moeten.
...of wordt het een glansrijke carrière?
Een schitterende diamant is het resultaat van zorgvuldig bewerken en bijslijpen. Zo is het ook met professionals die schitteren binnen organisaties. In het begin zijn hun talenten nog verborgen of in het beste geval glinsteren ze aan de oppervlakte. Maar met een inke dosis initiatief en de juiste ondersteuning ligt er een glansrijke carrière in het verschiet. Een Kwantitatief Financieel Traineeship bij SNS REAAL is een uitstekend begin van zo’n carrière.
Kwantitatief Financieel Traineeship Dit traineeship geeft je de kans om samen te werken met onze vakspecialisten op het gebied van Risico-, Asset- en Treasurymanagement. Tijdens je traineeship doorloop je een boeiend opleidingsprogramma en werk je zelfstandig aan diverse opdrachten steeds voor een periode van 3 tot 6 maanden. Deze opdrachten vervul je bijvoorbeeld bij de afdelingen van Group Risk Management, Financial Markets en het Actuariaat. Eén opdracht kies je buiten deze afdelingen. Kortom: uitdagende opdrachten en je draait mee met de dagelijkse gang van zaken op de afdeling. Binnen het traineeship heb je alle mogelijkheden om niet alleen je kennis maar zeker ook je persoonlijke vaardigheden te ontwikkelen. Aan het einde van je traineeship kun je je specialiseren. Voor wie? (Bijna) afgestudeerde academici met een kwantitatieve studierichting kunnen solliciteren op dit traineeship. Bij het bepalen wie in aanmerking komt vinden wij de volgende punten ook belangrijk: • Je bent (bijna) WO-afgestudeerd met maximaal 2 jaar werkervaring. • Een kwantitatieve achtergrond, bijvoorbeeld econometrie, actuariaat, (toegepaste) wiskunde, technische bedrijfskunde of vergelijkbaar. • Je bent in staat om kwantitatieve modellen te begrijpen en te ontwikkelen. • Je beheerst de Nederlandse taal uitstekend in woord en geschrift. • Je gaat een goed adviesgesprek niet uit de weg. • Je geeft de voorkeur aan een omgeving, waarin je dicht bij de besluitvorming staat. • Je bent leergierig en enthousiast. • Je hebt een inke dosis talent: inhoudelijk ben je sterk en je combineert dit met zeer goede sociale vaardigheden.
• Je bent proactief en onderneemt initiatieven ter verbetering van je werkzaamheden. • Je hebt competenties als leervermogen, resultaatgerichtheid, klantgerichtheid, communicatief en samenwerken. Wij bieden In eerste instantie natuurlijk de kans om bij een innovatieve bank/ verzekeraar aan de slag te gaan. SNS REAAL wil zich steeds blijven ontwikkelen. Zo blijft SNS REAAL exibel en in staat om snel op de meest actuele ontwikkelingen te reageren. We verwachten dat jij met ons mee groeit in je ontwikkeling en we stimuleren dit dan ook van harte. We bieden je direct een contract aan binnen SNS REAAL met de mogelijkheid om van hier uit jezelf verder te proleren. Onze arbeidsvoorwaarden: • Volop kansen om jezelf te ontwikkelen en te ontplooien. • Een breed, boeiend en leerzaam opleidingsprogramma. • Intensieve begeleiding door een mentor. • Een tweejarig contract met een uitstekend startsalaris: € 2.500,- bruto per maand op basis van 40 uur. • Een 13de maand. • Een beoordelingstoeslag van 0 tot 15% van je jaarsalaris. • Direct een uitdagende functie, waarna je vervolgens door kunt groeien naar een specialistische rol. Het traineeship start per 1 februari 2011. Meer weten? Voor meer informatie over het traineeship kun je bellen met Arno van Eekelen, Senior Medewerker Group Risk Management, telefoonnummer 030-2915562, mobiel 06-24303733. Voor meer informatie over de procedure kun je contact opnemen met Lonneke Bierens, Recruiter, telefoonnummer 06-57 55 62 21 of solliciteer direct via www.werkenbijsnsreaal.nl/carriere.
Econometrics
Table 1. Results from the OLS regressions. *, ** and *** indicate significancy at the 10, 5 and 1 percent level respectively (Source: Persson and Tabellini, 2004). Dependent variable
cgexp
pres
-5.18 (1.93)*** -6.32 (2.11)***
maj propres majpar pajpres pres_newdem
cgexp
-6.56 (3.01)*** -6.96 (3.72)*** -10.37 (3.03)***
cgrev
dft
cgexp
cgexp
cgexp
-5.00 (2.47)** -3.68 (2.15)*
0.16 (1.15) -3.15 (0.87)***
-2.65 (2.70) -1.45 (2.32)
-7.75 (2.70)*** -7.94 (3.74)**
-6.46 (2.98)** -6.33 (2.48)**
3.50 (2.72) 3.58 (4.03) -4.08 (2.23)*
maj_newdem newdem pres_baddem maj_baddem baddem F-test (pres) F-test (maj) Sample Observations R²
0.43 1990's 80 0.71
1990's 80 0.70
1990's 76 0.68
we know that we have dealt with this omitted variables bias? P&T justify their methodology by using a proper set of control variables, however there are always other factors that can be overlooked. Therefore the authors use Heckman correction which very loosely can be described as estimating the original two-part model with no adjustments at all and the corresponding bias.7 Finally the earlier estimates are then corrected with the estimated bias and should lead to unbiased results. Another option involves the classical Instrumental Variables (IV). Lastly, the linearity assumption can be relaxed. Linearity is taken as a convenient local approximation of a more general model. However the constitutional effect on policy outcomes may be stronger in older or better democracies. As these features differ systematically across constitutional groups, the local approximation may no longer be tenable and the linear estimates are biased. Their solution consists of using matching methods. More weight is given to the comparisons of similar countries to reduce the effect of any nonlinearities. The basic idea is that we should compare the performance of similar countries, because their selection into different constitutions is largely random. Thus for each country with a particular constitutional rule, we try to find its “twin” or a “set of close relatives” to it with the alternative constitutional rule.
7
1990's 72 0.50
4.01** 3.18* 1960-1973 1990's 42 80 0.79 0.72
2.42 (4.16) 2.06 (5.97) -5.73 (3.46) 1.40 0.66 1990's 80 0.70
Econometrics at work: results of the regressions In their basic OLS setup, the results are in line with the earlier mentioned theoretical frameworks. A switch from proportional to majoritarian elections in a country chosen at random reduces total government spending by about 6 percent of GDP, which can be seen in the first column of table 1. Similarly the estimates for form of government indicate that presidentialism reduces the overall size of government by roughly 5 percent of GDP. The effects of the two constitutional features also appear to be additive. A F-test does not reject the null hypothesis that the estimated coefficient of majpres in column 2 of table 1 equals the sum of the estimated coefficients of propres and majpar. Thus these estimates seem to indicate that introducing a presidential form of government and majoritarian electoral rules in a proportionalparliamentary country would reduce central government spending by a significant amount of 10 percent of GDP. When other measures of size of government are used, most of the results remain intact. While the effects are smaller, the signs are preserved in the government revenues setup (column 3 of table 1): a hypothetical switch from presidential to parliamentary government would increase both spending and revenues by the same
This is feasible due to some identification assumptions which we will not discuss in this article.
AENORM
vol. 18 (70)
December 2010
17
Econometrics
Table 2. Results from the Heckman specifications, IV estimations and matching methods. *, ** and *** indicate significancy at the 10, 5 and 1 percent level respectively (Source: Persson and Tabellini, 2004). Dependent variable
cgexp
cgexp
cgexp
cgexp
cgexp
cgexp
cgexp
pres
-5.29 (2.18)**
-11.52 (4.54)**
-6.51 (3.71)*
-4.22 (3.99)
-5.86 (4.53)
-2.54 (2.26)
-7.30 (2.36)***
-4.18 (3.17) col_uka, laam 1990's pres maj 2SLS
-4.86 (3.57)
-6.59 (3.40)*
-5.76 (2.59)**
1990's pres maj Stratification
1990's pres maj Nearest neighbor
1990's pres maj Kernel
65(pres) 67(maj)
65(pres) 67(maj)
65(pres) 67(maj)
maj
-6.21 (2.82)** Conts & Cols Yes Sample 1990's Endogenous selection maj Estimations Rho Chi-2 Adjusted R² Observations
-6.77 (1.98)*** Yes 1990's pres
-4.83 (3.19) col_uka 1990's pres maj Heckman Heckman 2SLS ML ML 0.05 0.62 (0.29) (0.33) 3.29 0.59 75 75 75
amount of 5 percent. However the effect of majoritarian elections is cut in half, resulting in a reduction of about 3 percent of GDP. This basic setup is robust to the specification of the control vector Zi. When less influential controls are dropped or other controls such as income inequality, ethnic and linguistic fractionalization and whether the country was a former social country are added, the results do not change much as the estimated coefficient of majoritarian elections is always negative and (almost) always significant. This also holds true when P&T focus on the age of the constitution and the quality of democracy: constitutional effects appear stronger in old democracies and the effect of presidential regimes is stronger in higher quality democracies, whereas the effect of majoritarian elections remains stable. In table 2 the results are shown when the standard OLS assumptions of conditional independence and linearity are relaxed. The first two columns show the Heckman adjustments and it seems that the results are strengthened: the signs of the coefficients are in line with theoretical results and significant. Furthermore the magnitude of these coefficients are larger in absolute value than under the previous setup which indicates that the OLS regressions were upwards biased. However the other two estimation strategies do not seem to be that successful. The instrumental variables estimations (columns 3 and 4) lead to coefficients that are not significant. Moreover when P&T test for the validity of their instruments it does not lead to promising results. A similar story holds for the matching methods: both stratification and nearest neighbor techniques lead to insignificant coefficients.
18
AENORM
vol. 18 (70)
December 2010
2.23 0.59 75
Conclusion In political science and economics, scientists have hypothesized about the effects of electoral rules and forms of government on fiscal policy outcomes in the form of theoretical models. In their paper, P&T empirically test the predictions of these frameworks in a cross-country data set of 80 democracies. As predicted a hypothetical shift in constitutional rules, coined the average treatment effect, does affect government spending. Majoritarian elections lead to smaller governments when compared to proportional elections: it reduces total government spending by about 6 percent of GDP. The data also seems to support hypothetical changes concerning forms of government. Presidentialism roughly cuts the size of government by about 5 percent of GDP. While these results are robust regarding control variables variation, the basic OLS framework does seem to rely on strong assumptions. However when these assumptions are relaxed, the results still stand, although they do seem to be less firm. P&T’s econometric framework does incorporate another important implicit assumption: constitution affects policy outcomes directly. However political scientists have doubts about this particular mechanism. They argue that constitutional rules have an impact on factors such as party structures, types of government and occurrence of elections or government crises. These in turn may affect fiscal policy outcomes, which implies that constitutional rules only indirectly affect government spending. However this is not reflected in P&T’s empirical model. Acemoglu (2005) argues that
Econometrics
this is crucial and criticizes P&T’s methodology heavily: he considers their results to be cases of robust correlation rather than causal effects. Nonetheless, P&T’s work is a first step in empirically identifying relationships between constitutional rules and fiscal policy and may make you wonder that there is more behind government spending than you think.
References Diermeier, D. and T. Feddersen. “Cohesion in Legislatures and the Vote of Confidence Procedure.” American Political Science Review 92.3 (1998): 611 – 621. Milesi-Ferretti, G.M., R. Perotti and M. Rostagno. “Electoral Systems and Public Spending.” Quarterly Journal of Economics 117.2 (2002): 609 – 657. Persson, T. and G. Tabellini. “The Size and Scope of Government: Comparative Politics with Rational Politicians.” European Economic Review 43.4 - 6 (1999): 699 – 735. Persson, T., G. Roland and G. Tabellini. “Comparative Politics and Public Finance.” Journal of Political Economy 108.6 (2000): 1121 – 1161. Persson, T. and G. Tabellini. “Constitutional Rules and Fiscal Policy Outcomes.” American Economic Review 94.1 (2004): 25 – 45. Acemoglu, D.. “Constitutions, Politics and Economic Growth: A Review Essay on Persson and Tabellini’s ‘The Economic Effects of Constitutions’.” Journal of Economic Literature 43 (2005): 1025 – 1048.
AENORM
vol. 18 (70)
December 2010
19