It's a Matter of Confidence: Institutions, Government Stability and Economic Outcomes

The effect of constitutional structures (such as the effect of a presidential vs. a parliamentary system) over policy outcomes has been widely studied in the economic literature. In this paper, we investigate whether stable parliamentary systems and unstable parliamentary systems behave differently in terms of the policy outcomes they implement. We show that accounting for the stability of parliamentary systems generates results that are more robust compared to the previous literature. More precisely, we find that stable parliamentary systems are significantly different both from presidential and from unstable parliamentary ones. Moreover, we show that this result is robust to changes in the set of countries, and to changes in the definition of stability. Finally, we discuss how these results are consistent with the presence of a selection effect in parliamentary systems.


Introduction
It is universally acknowledged that constitutional structures do shape the incentives of the political actors so to transform political ideas into policies.
Starting with the seminal work of Persson and Tabellini (2003) -hereafter PT -most of the existing literature has focused on the differences between parliamentary and presidential systems. Recent literature (e.g. Blume et al., 2009) has highlighted how those findings are not robust to changes such as, for example, the set of countries and the time span. Furthermore, some authors (e.g. Voigt, 2011b) have suggested that the distinction between parliamentary and presidential systems may simply be too coarse and that possible extensions include the use of more fine grained variables to classify constitutional systems.
We remain in the same stream of thoughts and, hence, consider the presence of the confidence requirement the key variable to distinguish parliamentary from presidential systems. 1 Interestingly, such constitutional feature of parliamentary systems is used in different ways, depending on the country. In some cases, the confidence requirement does indeed generate frequent changes of government, thus replacing possibly bad politicians and generating a different government composition (selection effect). In other countries, the confidence requirement acts as a credible threat and may induce either the executive to behave better (disciplining effect) or the parliament to accept more frequently the executive's misbehavior. Hence, parliamentary systems perform differently in presence of different structural characteristics of the politicians such as, for example, the quality of the information available and/or the alignement of interests with the citizens. Given this complexity, we investigate more deeply the characteristics of countries that adopt a parliamentary constitution by considering the stability of the government as a proxy to distinguish different parliamentary systems. We measure stability as inversely related to the frequence of changes in the government composition, which is clearly correlated with the effective use of the confidence requirement. Hence, we contribute to the literature by refining the standard classification (see PT, 2003) introducing stable and unstable parliamentary systems.
We explore whether stable and unstable parliamentary systems behave differently in terms of the policy outcomes they implement. We show that this finer classification of constitutional systems (presidential, stable parliamentary, unstable parliamentary) delivers more robust results. In detail, we find that stable parliamentary systems are significantly different both from presidential and unstable parliamentary ones. Moreover, we find that this result is robust to changes in the set of countries included in the dataset and in the definition of stability. These empirical findings are consistent with the theoretical analysis of Cella et al. (2014), where the authors link stability to a lower quality of legislators which in turn implies that selection is less effective in stable parliamentary systems. As a consequence, these systems generate policy outcomes that are further away from the presidential ones. This paper links two strands of the political economy literature. First of all, it belongs to the field of empirical constitutional economics, as we compare the effects that constitutions have on policy outcomes (for a survey of the recent developments see Voigt, 2011aVoigt, , 2012. The novelty of our work is in the introduction of the finer classification of parliamentary systems, according to our stability measures. This paper belongs also to the theoretical political economy literature that focuses on the differences (and their consequences) between parliamentary and presidential structures. Persson et al. (1997Persson et al. ( , 2000 claim that parliamentary constitutions generate a political system that translates into higher legislative cohesion if compared to presidential ones, that instead guarantee a better separation of powers and therefore accountability. Diermeier and Vlaicu (2011) show that higher cohesion implies higher legislative success and they highlight how the presence of the confidence requirement changes the intertemporal incentives of politicians. The idea that being a member of the ruling coalition lengthens the time horizon for politicians 3 is present also in Feddersen (1998a, 1998b). Voigt (2011a), following Brennan and Kliemt (1994), discusses how parties may respond differently to the confidence requirement due to the longer time horizon they face compared to individual politicians such as presidents in presidential systems. This difference in the discount factors may lead to the implementation of different policies across systems.
The paper is organized as follows: Section 2 discusses in depth the empirical and theoretical background, Section 3 presents the data, Section 4 introduces the results and Section 5 concludes.

Empirical and theoretical background
We present in details the empirical and theoretical background of our findings, in order to highlight the novelty of our work.
On the empirical side, we closely relate to Persson and Tabellini (2003), where the authors compare constitutional systems -presidentialism vs. parliamentarism -and electoral rules -majoritarian vs. proportional -in order to identify the differences, if any, in a number of relevant social and economic indicators. The main result concerns the size of the central government: presidential regimes spend systematically 5% less than their counterpart, and this difference increases by an additional 5% if the country has adopted a majoritarian electoral system instead of a proportional one. 2 The authors also consider the effects on government's revenues, social welfare expenditure as percentage of the GDP, total factor productivity and the corruption level.
However, a number of papers question the accuracy of the authors' claim of causality. The main criticisms can be classified into four categories: i) the overall robustness of the empirical analysis (Blume et al., 2009); ii) the construction of the explanatory variable (Voigt, 2011b); iii) the exogeneity of the explanatory variable (Hayo and Voigt, 2013;Acemoglu, 2005); iv) the transmission channels (Voigt, 2011b;Robinson and Torvik, 2008  This theoretical analysis suggests that fully stable parliamentary systems behave exactly like presidential ones, while those systems that do replace their political establishment should be more different from presidential ones the greater their stability index. The predictions of the model for fully stable parliamentary systems are ambiguous, as we may not be able to empirically identify whether a parliamentary system is stable because it is in the equilibrium in which the executive is disciplined, or because it has a legislative body with a low performance. Hence, we suggest a possible transmission channel between stability and performance, thus partially addressing also the fourth critique.

Data
In order to facilitate the comparison with the previous literature we mantain the essential setting of PT (2003) and we start from their original dataset.
The dataset is composed by 85 countries, somewhat balanced along the continents. It contains a wide set of data ranging from economic to social, cultural and political information.
We then test the robustness of our analysis by making use of the extended dataset as in Blume et al. (2009). The latter dataset enlarges the PT one 7 up to 116 countries and updates some of the variables. 3 We further update certain variables that we use to perform robustness checks. 4 We also use a set

Explanatory variables
The confidence requirement is rarely the explicit reason for a government collapse (approximately 40 cases upon over 800 elections in our dataset).
However, the mere existence of the confidence requirement works -to a certain extent -as an incentive for a questioned government to resign before being dismissed by the use of a motion of no-confidence, thus favouring a cabinet reshuffle. According to this view, we collected data on government stability in terms of the capacity of governments to last till the end of their political term. We believe government stability represents a well-suited proxy for determining the effectiveness of the confidence requirement as an instrument to dismiss a questioned government.
Following PT (2003), we build the dummy variable P RES that identifies countries according to the constitutional design in use, that can be either the presidential system or the parliamentary one, according to the legal existence of the motion of confidence. This is a restrictive definition of presidential systems, as presidential countries where the government is In order to incorporate stability within the group of parliamentary countries, we mainly rely on the information available in the DPI dataset. The objective is to fix specific thresholds to classify parliamentary countries in stable and unstable ones. We construct two additional dummies P ARL ST AB and P ARL U N ST AB to identify the two categories.
The most challenging aspect has been the choice of the index of stability.
Throughout the paper we adopt the partition resulting from the index GOV LIF E, defined as follows: where D i represents the real number of years a government has been in office between two elections, E i is a dummy which is equal to one when an election occurs, and X i , is the legal length of any electoral term according to country-specific constitutional rules. 6 Thus, the numerator stands for the average length of any electoral cycle computed for each country i, and the index weights such length by the legal one. The index ranges from zero to one, with higher values that correspond to higher stability.
Given the index GOV LIF E, we create the dummy variable P ARL ST AB that takes value one for countries with an index values above the median and zero otherwise, and the dummy P ARL U N ST AB that takes value one for countries with an index values under the median and zero otherwise. 7 Therefore, in order to provide robustness checks we have considered three additional stability indexes. 8 6 The index GOV LIF E is built using the indicator yrcurnt from the DPI dataset which is coded zero in an election year, and Xi − 1 in the year after the election. 7 More precisely, we first drop three countries around the median in order to avoid a random assignment of countries due to measurement errors. 8 The robustness checks are reported in Table 7 in the Appendix.

9
The first one -GOV EN D -is the fraction of governments that are successful in reaching the legal term of the mandate. Again, higher values of the index correspond to higher stability.
The second index -Y EAR EXEC -is the average tenure of the head of the executive weighted by the legal length of any electoral term. 9 The higher the value of the index, the higher the stability. Note that this index

Dependent variables
We We also provide results for other dependent variables which include the expenditure on education, the social protection expenditure, the country openness, the level of corruption, the productivity, and the total government expenditure. 11 This last variable is of particular importance given that the accuracy of central government expenditure as a measure of the government size has been questioned in the literature (see Voigt, 2011b).
We also account for the executive ideological position under the assumption 9 This index is built using the indicator yearof f from the DPI dataset, which collects information about the number of years the head of the executive has been in office.
that leftist executives should exhibit a larger public expenditure (Becher, 2013). 12

Control variables
We

Results
The empirical specification follows PT (2003). We modify the original setting to introduce a finer partition for parliamentary countries. We apply the dummy coding technique to account for heterogeneity in parliamentary systems. This choice generates three categories: P RES, P ARL ST AB and

P ARL U N ST AB.
We estimate the model by making use of the multiple regression procedure, with two predictive dummy regressors that enter the regression equa-12 Source: DPI, 2012. Results are reported in the Appendix, Table 8. 13 The variables con20, con2150, con5180, respectively dating the constitution's origin before 1920, between 1921-1950, and between 1951-1980. 11 tion: where P RES is the baseline category that represents the control group in our setting. We are interested in testing whether presidential systems differ from stable parliamentary ones (β 1 = 0), whether presidential systems differ from unstable ones (β 2 = 0) and whether stable and unstable parliamentary systems differ from each other (β 1 = β 2 ). On the other side, we cannot reject the null-hypothesis that unstable parliamentary systems behave as presidential ones. Once again this result holds also when we adopt the larger dataset. 14 This evidence corroborates the idea that the criterion we follow to account for the heterogeneity in parliamentary countries is meaningful and statistically robust.
We then replicate the same exercise using the central government revenue -tax revenue -as dependent variable (Table 3). We show that the difference between presidential and stable parliamentary systems is the only significant one. Then -following the same PT (2003) structure -we check whether presidential governments have a smaller share of social welfare expenditure compared to parliamentary ones (Table 4). Theoretically, parliamentary systems stimulate collective action between the executive and the 14 The Wald test for β 1 = β 2 reports a p-value of 0.0132 and 0.0179 for regressions in Table 2 [columns 2 and 4, respectively].  We test the robustness of the latter claim by changing the dependent variable. In Table 5,   Notes: SSW is the central government expenditure in social services and welfare; PRES is the dummy that identifies presidential systems; PARL STAB is the dummy that identifies stable parliamentary systems according to the index GOV LIFE; PARL UNSTAB is the dummy that identifies unstable parliamentary systems according to the index GOV LIFE. White heteroskedasticity-consistent standard errors in parentheses. PT-modified and BCIMmodified refer to PT (2003) specification of the model where the authors include all the standard controls -age, lyp, prop65, gastil, federal, oecd, africa, asiae, laam, col uka, col espa, col otha -except that lpop, prop1564 and trade are missing. Then, we re-estimate the model using the same specification as in previous tables.  (2)] is the central government social protection expenditure as defined by the IMF/GFS; we report the average from 1990 to 2012. EDSPEND [Column (3)-(4)] is the expenditure on education as percentage of the GDP, as defined by the IMF/GFS; we report the average from 1995 to 2012. PRES is the dummy that identifies presidential systems; PARL STAB is the dummy that identifies stable parliamentary systems according to the index GOV LIFE; PARL UNSTAB is the dummy that identifies unstable parliamentary systems according to the index GOV LIFE. White heteroskedasticityconsistent standard errors in parentheses. Controls include: age, lyp, prop65, gastil, federal, lpop, prop1564, trade, oecd, africa, asiae, laam, col uka, col espa, col otha. * p < 0.1, * * p < 0.05, * * * p < 0.01 is effectively used are increasingly similar the more unstable they are. As a consequence, as already mentioned in Section 2, we predict: (i) a significant difference between parliamentary systems with an intermediate level of stability and presidential ones, (ii) no difference between unstable parliamentary systems and presidential ones. Moreover, the theoretical prediction on the difference between the most stable parliamentary systems and the presidential ones is ambiguous.

12
In order to investigate this intuition, we propose an even finer partition of parliamentary systems. We further divide parliamentary countries in three categories: stable, unstable and partially stable countries (PARL PARSTAB ). 15 We expect partially stable parliamentary systems to be further away from presidential ones than unstable systems. We show that the partially stable parliamentary systems are significantly different both from the presidential and the unstable parliamentary ones which is consistent with our theoretical prediction. As for the fully stable parliamentary systems results are ambiguous, as the performance of countries in this group is not significantly different from any other category. 16 The results are therefore compatible with the theoretical predictions in Cella et al. (2014).
Hence, we conclude that the selection effect that operates less effectively in parliamentary systems with a low quality legislators may be identified as a possible transmission channel of the difference in performance between the constitutional systems.

Conclusion
This paper analyses the effect of constitutional structures on policy outcomes with a specific attention to the role of the confidence vote. In particular, the novelty of the paper rests with the understanding of the link between government stability and economic outcomes, particularly for parliamentary systems. Hence, the empirical analysis we perform introduces finer partitions 15 We exploit the same stability index as before -GOV LIFE -but classifying parliamentary systems into the three categories. 16 We can never reject the null-hypothesis when performing the Wald test.    Notes: GGEXP is the total government expenditure; OPEN is the country openness (imports plus exports over the GDP); PRES is the dummy that identifies presidential systems; PARL STAB is the dummy that identifies stable parliamentary systems according to the index GOV LIFE; PARL UNSTAB is the dummy that identifies unstable parliamentary systems according to the index GOV LIFE. White heteroskedasticity-consistent standard errors in parentheses. The regressions include the following controls: age, lyp, trade, prop1564, prop65, gastil, federal, oecd, lpop, africa, asiae, laam, col uka, col espa, col otha. The specification section refers to the difference between the PT explanatory variable PRES and our division between stable and unstable parliamentary countries.   (3) and (4)]; PRES is the dummy that identifies presidential systems; PARL STAB is the dummy that identifies stable parliamentary systems according to the index GOV LIFE; PARL UNSTAB is the dummy that identifies unstable parliamentary systems according to the index GOV LIFE. White heteroskedasticity-consistent standard errors in parentheses. Following PT (2003), all the regressions include the following controls: age, lyp, trade, gastil, federal, oecd, lpop, africa, asiae, laam, col uka, col espa, col otha, avelf, prot80, catho80, confu.
The specification section refers to the difference between the PT explanatory variable PRES and our division between stable and unstable parliamentary countries. * p < 0.1, * * p < 0.05, * * * p < 0.01