Selection of judges and law enforcement - Stanford University
Transcript of Selection of judges and law enforcement - Stanford University
Selection of judges and law enforcement
Julia Shvets�
LSE and Cambridge University
August 2005
Abstract
This paper analyzes the impact of judicial selection procedures on lawenforcement by courts. The model contrasts selection of judges by thelegislator with that by the executive, two common ways of appointingjudges around the world. It is based on the assumption that, while bothpolitical o¢ ces are interested in rents, the decisions of the executive aresubject to judicial review, but the decisions of the legislature are not. Themodel predicts that relative to the legislator, the executive has incentivesto select judges that are less favourable towards small �rms.
I contrast two regimes empirically, using a rare natural experiment inappointment procedures in Russian commercial courts. I put together apanel dataset of decisions of individual judges. The method of appoint-ment di¤ers both over time and across courts. Consistent with the model,the empirical analysis shows that judges selected by the executive tendto favour of small �rms less than those selected by the legislative. Themodel constraints the data in a number of ways, and the richness of thedata allows to test for main alternative explanations. The results are alsorobust to allowing for selection into litigation, and correcting for potentialendogeneity of appointment procedures.
�I would like to thank Tim Besley for his guidance, Mark Schankerman, Heski Bar-Isaac,Jordi Blanes i Vidal, Charles Goodhart, Shira Klien, Ellen Meade, Inger Munk, Martin Old-�eld, Elena Panova, Kwok Ton Soo, Chenggang Xu and LSE and Cambridge seminar partic-ipants for helpful comments and suggestions. I am also grateful to the Legal Reform Projectof the Stockholm Institute of Transition Economics for providing �nancial support for thisresearch.
1
1 Introduction
A growing stock of empirical evidence suggests that with better law enforcementenjoy better economic performance1 . Given the signi�cance of law enforcementfor economic development, it is important to understand why some court sys-tems are more e¤ective than others. In particular, does the organization ofjudiciary a¤ect performance of judges, and, through it, economic growth?This paper studies the impact of selection procedures for judges on law and
policy enforcement with implications for �rms. Di¤erent countries entrustjudicial appointments to di¤erent political o¢ ces, typically either the executiveor the legislative branches of the government, or the combination of the two2 .During the last �fty years, several Western countries have introduced substantialchanges in the process of selection and appointment of judges3 . Yet, whetherthese di¤erences in organization of judicial selection a¤ect the process of lawenforcement remains an open question.This paper o¤ers two main contributions. First, I empirically contrast law
enforcement by judges appointed by the executive on the one hand and thelegislature on the other. To do so, I take advantage of a natural experiment inRussian commercial courts. There, due to an exogenous constitutional change,the way judges had been selected varies both across courts and over time.Second, I develop a model which contrasts incentives of the legislator to
those of the executive when each is in charge of selecting judges. The model isbased on a premise that politicians are (at least partly) interested in rents. Itclasses all �rms into those that are able to pay rents (�large �rms�) and thosethat are not (�small �rms�). Judges di¤er in their preferences in favour of small�rms. Through their decisions judges can a¤ect rents politicians collect fromlarge �rms. Therefore, politicians select judges strategically.The di¤erences in incentives faced by the legislator and the executive when
selecting judges arise because the executive�s decisions are typically subject toa judicial review, while the legislator�s are not. The executive who collectsrents from large �rms will not select judges who protect interests of small �rms.In contrast, the legislator writes laws for the judges he selects. When lawsare incomplete, large �rms may view litigation a substitute for legislation, andthe legislator may �nd himself competing against the judiciary for rents4 . Todampen this competition, the legislator has an incentive to increase the costs oflitigation for large �rms by selecting judges who favour small guys.
1For example, Mauro (1995) and Keefer and Knack (1995) report signi�cant links betweenquality of law enforcement and growth. Later papers, such as La Porta et al (1997), Pistoret al (2000), and Demirguc-Kunt and Maksimovic (1998), show a relationship between lawenforcement and �nancial market development.
2See Skordaki (1991) and Thomas (1997) for cross-country surveys of judicial appointmentprocedures.
3 In 1990s, proposals for reforms of judicial appointment procedures were also discussed inthe UK and Australia (see, for instance, Kendall (1997)).
4Shapiro (1995) writes interest groups in the US: "...lobbying of courts by interest groupsbent on winning in court what they cannot win in the political process ... is the stu¤ ofjudicialization."
2
The empirical analysis focuses on Russia, where restructuring of the judi-cial system has been named one of the top reform priorities5 . There, frequentinstitutional changes have given rise to a variation in selection procedures incommercial (arbitrazh) courts. Depending on the region and year of appoint-ment, commercial court judges working side-by-side today have been selected bythree di¤erent political bodies: the federal legislature, the executive governmentand regional assemblies.I matched the method of appointment of individual judges to their decisions
made between 1995 and 2002. Sizes of �rms which participated in each courtcase were obtained from a separate database of enterprise statistics. I analyzewhether judges appointed by di¤erent methods display di¤erent policy prefer-ences in their decisions. In doing so, I use an identi�cation method based onreversal rates of judicial decisions developed in an ealier paper (Shvets (2005)).My results show that judges appointed by the federal legislature tend to
favour small �rms in disputes with government o¢ ces more than judges ap-pointed by the executive. This is consistent with predictions of the model.Several alternative hypotheses are tested and rejected. The results are robustto a number of controls and to inclusion of court �xed e¤ects. The identi�cationmethod is also robust to endogenous selection of disputes into litigation.This study contributes to, and has been inspired by, a growing body of
research on design of institutions and their economic impact. In particular,political economy literature shows that speci�c institutional arrangements areimportant for government performance (see, for example, Persson and Tabellini(2000) and Besley and Case (2003)).Judicial institutions have been analyzed to a much lesser extent. The con-
sequences of re-electing judges instead of giving them permanent tenure aremodelled by Maskin and Tirole (2004) who show that popular elections gen-erate incentives to pander to the views of the electorate. Besley and Payne(2003) �nd empirical support for this in the US state courts. They show thatmore workers �le employment discrimination cases with courts where judgesare re-elected than with courts where judges are reappointed. Hanssen (1999,2000) shows that the US states were judges are elected systematically di¤erfrom those where judges are appointed: the former have lower volumes of liti-gation and smaller number of government agency employees. In a 2004 paper,Hanssen demonstrates both theoretically and empirically using the US as anexample that state politicians tend to choose the method of judicial rententionstrategically.The issue of judicial selection received relatively little attention. To my
knowledge, there is no theoretical work modelling the consequences of di¤er-ent selection procedures. Empirical evidence on such procedures is limitedto the US. The early work of Canon (1972) and Glick and Emmert (1987)links appointment procedures to judicial characteristics using association ta-bles. They �nd that judges selected by the legislature tend to come from a
5See, for instance, Black et al (2000) who argue that lack of �decent legal and enforcementinfrastructure�was largely responsible for the many failures of Russian privatisation. See alsoGray and Hendley (1997), Kreuger (2002), and Biletsky et al (2002).
3
di¤erent professional background than elected judges and those appointed bythe executive. There have been very few attempts to link selection procedures ofjudges to their performance, and they do not paint a consistent picture. Besleyand Payne (2003) show that whether judges are initially elected or appointedby a politician does not matter for the number of employment discriminationcharges they attract. Ashenfelter et al (1995), analyzing two US courts, donot �nd a relationship between judges�decisions and the party of the presidentthat appointed them. Revesz (1997), however, does �nd that judges chosen byRepublican presidents tend to side with industry more often in environmentaldisputes.This paper analyzes judicial selection both empirically and theoretically. My
model of judicial selection has been in�uenced by political economy literaturealready mentioned above, as well as by several papers on law and economics.In particular, the observation that judges a¤ect politicians�rents and therefore,politicians have a strategic interest in how judiciary is organized was �rst madeby Landes and Posner (1975). The concept of incomplete law, which hasinspired my view of the roles of judges and the legislature vis-a-vis each other,is developed in Xu and Pistor (2002).This study has also been in�uenced by the literature linking variation in
court performance to the economic and �nancial development of countries.Countries with better judicial institutions have been shown to have better prop-erty rights protection and more developed �nancial markets (La Porta et al(1997), La Porta et al (2004), and Pistor et al (2000)), a smaller uno¢ cial econ-omy (Johnson et al (1998)), and more investment (Johnson et al (2002)). The�ndings of this study contribute to this literature by shedding light on possiblesources for the observed di¤erences in court performance.The paper is organized as follows: section 2 presents a theoretical model
of legislative and executive appointment of judges. Institutional features ofRussian commercial courts and the data are described in section 3. Section 4lays out the empirical model, and section 5 presents empirical results. Con-cluding remarks are in section 6.
2 The model
The model contrasts incentives of the legislator to those of the executive whenselecting judges. It is based on a premise that both the legislator and theexecutive collect rents. Judges, through their decisions, can a¤ect these rents,and, therefore, politicians select judges strategically.After judicial selection is complete, the executive allocates a resource to
potential bene�ciaries. I divide all policy bene�ciaries into two types: thosethat can pay rents to politicians (�large �rms�) and those that cannot (�small�rms�). The legislator writes a law which governs allocation of resources by theexecutive. The judge scrutinizes the allocation (made by the executive) for itsconsistency with the law (stipulated by the legislator), with social welfare, andwith his own preferences vis-a-vis the bene�ciaries of the two types.
4
Since the executive is interested in rents, he allocates the resource to thosewho pay, and receives less rent when the chances of the court striking down hispolicy are higher. The executive shares rent with the legislator to secure a lawfavourable to the executive and to those who pay rent.The decisions of the judge (selected by either the legislator or the executive)
may be appealed to a higher court. To keep in line with Russia�s institutions,which are the focus of the empirical investigation in section 5, I assume thatthe appellate court judge is appointed by the executive. I draw empiricalimplications from the model by linking the selection of judges to reversal ratesof their decisions.
2.1 Set up
Let the economy consist of two �rms, a small and a large one, and three govern-ment branches: the executive, the legislator, and the judiciary. The latter hastwo-tiers: the lower court judge and the appellate court judge. The executiveselects the appellate court judge. Under what I call regime 1 he also selects thelower court judge. Under regime 2 the legislator selects the lower court judgeinstead.The large �rm pays rents to the executive in return for a favourable policy.
I assume that the small �rm cannot pay rents, but can take the executive tocourt if his policy is unfavourable to it.The executive decides on a policy, which entails allocating a resource of
value V between the two bene�ciaries. Let Y be the net welfare bene�t fromallocating the resource to the large �rm as opposed to the small one. Ex ante,let Y have a bell-shaped probability distribution g(Y ): For simplicity, I assumethat the executive cares only about rents. Therefore, he transfers all of V tothe large �rm. If the executive is taken to court by the small �rm, he investse¤ort e to persuade the lower court judge that the allocation of the resource tothe large �rm is justi�ed on the grounds of social welfare.The legislator writes the law P; which speci�es rules for how the executive
can allocate the resource. This law is characterized by how much it favoursallocation to the large �rm vis-a-vis that to the small one. Assume that thelegislator can choose between two values of P = f0; vg; where v > 0 is morefavourable to the large enterprise. To secure P = v; the executive pays thelegislator some of the rent that he collects from the large �rm. I assume thatthe legislator, just like the executive, only cares about rents, and, thereforealways chooses P = v in equilibrium.The two judges are characterized by their preferences vis-a-vis the small �rm.
Let z capture the extent to which the lower court judge favours the small �rm.I assume that z is drawn from an interval with the lower bound z0:The lower court judge observes Y +" where " is noise with distribution f(");
which is independent of Y and has mean zero. Let d denote a decision standardof the lower court judge, which measures the minimum net welfare bene�t fromallocating the resource to the large �rm that the judge must observe to upholdthe executive�s allocation. Hence, the lower court judge rules in favour of the
5
executive if Y +" � d and in favour of the small �rm if Y +" < d. The decisionstandard of the judge is a¤ected by three factors: the law on the books P , thee¤ort e that the executive invests into persuading the judge that it is optimal toallocate V to the large �rm, and the extent of the judge�s preferences in favourof the small �rm z: For simplicity, let d = z � P � e:Let Z capture the preferences of the appellate court judge in favour of the
small �rm, drawn from the same range as z. Let the appellate judge observeY directly. I assume that the executive cannot successfully invest e¤ort intolitigation at appeal. Let D be the appellate judge�s decisions standard, so thatthe executive wins the appeal if Y � D, while the small �rm wins it if Y < D:The appellate court judge reviews the lawfulness of the lower court decision,
but does not reexamine the facts of the case. Instead, he takes these facts asgiven by the lower court judge. I capture this by letting D = (1��)Z +�(z�e) � v, where � 2 [0; 1] measures the degree to which preferences of the lowercourt judge in�uence the appellate judge�s decisions through the fact �ndingstage:The dispute between the executive and the small �rm does not necessarily go
to court. After the small �rm threatens the executive with litigation, the lattermakes an out-of-court settlement o¤er to the former. I assume that the small�rm has the bargaining power, and gets the maximum the executive is willingto pay6 . The executive and the small �rm estimate their respective chancesof winning litigation based on signals � and � of Y that each of them receives,respectively. De�ne and � as noise being drawn from identical distributionsh( ) and h(�) independent of Y and with mean zero, and let � � Y + and� � Y + �. Assume that each party bases its beliefs about Y only on its ownsignal.The dispute goes to the lower court only if the small �rm rejects the ex-
ecutive�s o¤er. After the decision of the lower court, the small �rm and theexecutive negotiate on whether to take the case to appeal. Assume that theydo not update their information after the lower court decision. Let c be thetotal costs of litigating in the lower and appellate courts, borne by each party.The timing of the model is the following:t = 0 : The executive (under regime 1) or legislator (under regime 2) selects
the lower court judge out of a pool of candidates with di¤erent z: The executivealso selects the appellate court judge from the same pool.t = 1 : The executive allocates V to the large �rm in exchange for a payment
determined by Nash bargaining between the executive and the large �rm. Thelegislator selects P = v in exchange for a payment �L from the executive. Thesize of �L is determined by Nash bargaining between the executive and thelegislator.t = 2 : Y is realized. The small �rm threatens the executive with litigation.
They bargain to decide whether to litigate or to settle out of court. If theysettle, the game ends.t = 3 : If the case goes to the lower court, the executive invests e into
6This assumption is without loss of generality, but it allows to simplify the exposition.
6
litigation. The lower court judge makes a decision in favour of either theexecutive or the small �rm.t = 4 : The loser threatens the winner with appeal. The case is either settled
or goes to the appellate court. If it is settled, the game ends.t = 5 : If the case goes to the court of appeals, the appellate judge makes a
decision which either upholds or reverses that of the lower court:
2.2 Analysis
In regime 1 the executive government is in charge of judicial selection. Theexecutive chooses the lower court judge and the appellate court judge to maxi-mize the payment he receives from the large �rm net of expected litigation costs.This amounts to optimizing z and Z:In order to write down the executive�s expected payo¤, I �rst analyze the
decision of the small �rm and the executive to litigate at t = 2 and t = 4:At t = 2; after being threatened with litigation, the executive makes a settle-
ment o¤er to the small �rm which is equal to his expected loss from litigation.The minimum that the small �rm is willing to accept is its expected gain fromlitigation. The two parties estimate their expected loss/gain based on theirsubjective probabilities of winning the law suit, calculated using their respec-tive signals � and �. The small �rm rejects the settlement o¤er if its expectedgain from litigation is higher than the expected loss of the executive. Thisoccurs when � and � lead to predictions which substantially di¤er from eachother7 . No updating of signals occurs after the lower court decision. Thus, ifthe decision is litigated in the lower court, it will also be litigated in the courtof appeals. Therefore, only the probability of winning in the court of appealsmatters for the parties�ex-ante payo¤s.Given a signal �; denote the executive�s estimated probability of winning at
appeal as p(Y > Dj�): Thus, at t = 2 the executive�s out-of-court settlemento¤er to the small �rm is
[1� p(Y > Dj�)]V + c+ e; (1)
which the small �rm rejects if
p(Y > Dj�)� p(Y > Dj�) > 2c+ e
V: (2)
Since the small �rm has the bargaining power in settlement, the executive�sexpected payo¤ is the same whether the dispute is litigated or not. Therefore,at t = 0; before Y; �; and � are realized, the executive�s payo¤ is
�X =1
2V
Z 1
�1
Z 1
�1p(Y > D j �)h( )g(Y )d dY � c� e: (3)
As one would expect, it is increasing in the probability that the court systemwill support his allocation of the resource to the large �rm.
7This is based on divergent expectations model by Priest and Klein (1984).
7
Proposition 1 The executive�s expected payo¤ is monotonically decreasing inthe preferences of judges in favour of small �rms:
The proof of proposition 1 is in appendix A. It follows from the fact thatthe expected payo¤ of the executive falls with the probability that the judiciaryoverrules his allocation of the resource to the large �rm. The more the judgesfavour the small �rm, the more likely they are to invalidate the executive�spolicy. This is true for both the appellate and the lower court judges. Thepreferences of appellate court judge a¤ect the �nal decision directly, while thepreferences of the lower court judge a¤ect the decision at appeal through thefact �nding stage.Proposition 1 implies that when the executive is in charge of appointing
judges he will choose both the lower and the appellate court judges which favourthe small �rms the least, i.e. those with z = z0 and Z = z0.
In regime 2, the legislator is in charge of selecting the lower court judge. Thelegislator chooses z to maximize the payment he receives from the executive inreturn for P = v; the law that favours the allocation of the executive�s resourceto the large �rm. The maximum that the executive would be willing to pay thelegislator is the di¤erence between the executive�s payo¤s under the two typesof laws, P = v and P = 0: Assuming Nash bargaining, the legislator�s payo¤ is
�L =1
2[�X(P = v)� �X(P = 0)]: (4)
When the legislator adopts the law in favour of the executive, he increasesthe probability that the executive�s allocation to the large �rm will be upheldby the court system. Both the legislator�s policy P = v and the e¤ort theexecutive invests into litigation e shift the decision standard of the lower court(and subsequently the appellate court) in favour of the executive�s chosen policy.Therefore, the two are substitutes in the eyes of the executive.
Assumption 1 There exists z� > z0 such that when P = v;
argmaxe�X
�= 0 8 z 2 [z0; z�)> 0 8 z � z�
This assumption states that when the judge does not favour small �rms toomuch (z < z�), the legislator�s policy v increases the probability of a judicialdecision in favour of the executive by a substantial enough amount so that theexecutive does not invest any more e¤ort into litigation. Thus, when z < z�;the legislator�s policy v allows the executive to save e; the amount he wouldhave otherwise had to spend to in�uence judicial decision: The more the judgefavours the small �rm, the more the executive needs to spend on persuading thejudge to uphold his allocation, and the more the executive pays to the legislatorto avoid having to do this.Therefore, the legislator�s payo¤ increases with judicial preferences in favour
of small �rm, as long as z < z�: This gives rise to the following proposition.
8
Proposition 2 When assumption 1 holds, the legislator maximizes his payo¤by choosing judges with z � z�.
Proof of Proposition 2 is in appendix B.The driving force behind proposition 2 is the substitutability of the roles of
the legislator and the judiciary in the eyes of the executive when the latter wantsto ensure enforcement of his policy. The legislator perceives this competitionfor rents from the judge, and strategically appoints a judge who is relativelyfavourable to the small �rm to reduce the outside option of the executive.
2.3 Testable implications
I now map predictions of this model onto observables. Recall that under bothregimes 1 and 2, the appellate court judge is appointed by the executive. Thus,the model predicts that he favours small �rms the least, so Z = z0: Under regime1, the lower court judge is also appointed by the executive, and, therefore, hasthe same preference as the appellate court judge, so z = Z = z0:Under regime 2, the lower court judge is appointed by the legislator and,
therefore, has z � z� > z0: Thus, under regime 2, the lower court judge favourssmall �rms relative to the appellate court judge.Although judicial preferences z are not observed directly, di¤erences in policy
preferences of the lower and appellate court judges can be identi�ed from thecourt of appeals reversals of the lower court decisions. A model developedin Shvets (2005) shows this. The main result of that model relevant here issummarized in the following proposition:
Proposition 3 If the lower court judge favours the small �rm more than theappellate court judge, the probability that the decision of the lower court judgeis reversed at appeal is higher if this decision were in favour of the small �rmthan if it were in favour of the executive, as long as the variance of and � isnot too large.
The proof of this proposition is in appendix C.Thus, there are three testable implications. First, when a judge is appointed
by the legislator, the probability that his decision is reversed on appeal is higherif it is in favour of a small �rms. Second, when a judge is appointed by theexecutive, the probability that his decision is reversed at appeal does not dependon who the decision is in favour of. Third, as long as judges selected by di¤erentmethods have the same variance of signal noise "; decisions in favour of small�rms by the legislator�s judge will be reversed more often than decisions in favourof small �rms by the executive�s judge. The converse is true for decisions againstsmall �rms.
2.4 Discussion
In this section I discuss the main assumptions of the model. I begin withthose concerning interactions between politicians and courts. The di¤erences
9
in incentives of the executive and the legislator arise because I assume that thedecisions of the executive are subject to judicial review, while the decisions ofthe legislator are not. This is indeed the case in most countries. The exceptionare constitutional courts, which typically review legislation for its consistencywith the country�s founding principles. Therefore, my theory is not relevantfor such courts.In the model, the welfare consequences of the executive�s allocation Y are
realized after the legislator passes his law, and laws are (implicitly) assumed tobe incomplete. In other words, the legislator cannot tell the judge what to dofor every Y; giving rise to judicial discretion. Therefore, the judge �lls in thegaps in laws with his decisions. This notion has been implicitly employed inmuch of the literature on economic impact of legal systems, and is con�rmed byseveral empirical studies.A critical assumption of the model, not made explicit thus far, is that those
who pay rents cannot capture judicial selection in the same way as policy de-cisions. This assumption re�ects the substantially larger free-rider problemsthat face �rms or lobbies who want to capture judicial appointments: judgestypically di¤er along some broad dimensions (for example, pro-industry vs. pro-government), while policy can often be tailored to a particular lobbyist. Fur-ther, capital constraints may make the �purchase�of judicial appointments dif-�cult since the legislator would demand now the discounted value of rents heexpects to receive in the future.The assumption that the executive rather than the large �rm transfers rents
to the legislator is for simplicity. Relaxing it does not change the results.I make two assumptions on the organization of the appeals process. First,
the appellate court does not re-examine the facts of the litigation, but takesthem as given by the lower court. This is consistent with the mandate ofappellate courts in most countries. If instead I assumed that decisions of thelower and appellate courts were unconnected, the preferences of the lower courtjudges, if perfectly observed, would not have an e¤ect on outcomes, since only�nal judicial decisions matter. Second, I assume that the executive cannotsuccessfully invest e¤ort into litigation at appeal. This assumption capturesthat appellate courts typically have higher professional standards than lowercourts. A weaker assumption that investment at appeal is su¢ ciently lessproductive than at the lower court is enough for the results to hold.I also make several simplifying assumptions regarding the preferences of the
legislator, executive and judges. Although above I suppose that both politicalo¢ ces only care about rents, the results continue to hold if I allow social welfareto enter their utility. For simplicity, the judges in the model are concerned onlywith social welfare (Y ) and their own policy preferences. The results remainrobust if judges are allowed to be partial rent-seekers. This can be easily shownby reinterpreting the executive�s investment into litigation as rents for the judge.Several assumptions are used to derive testable implications. In the data,
the ability of �rms to pay rents is not observable. Empirically, I take size tobe an indicator of this ability and, in line with it, in the model I assume thatonly large �rms can pay rents (see also section 3.2). Second, I assume that the
10
executive branch selects the appellate court judge. This is what happens inRussia, and assuming it in the model helps me draw testable results. The modelcan be easily generalized to allow the appellate court judge to be appointed bythe legislator or more than one method.
3 Background and data
To identify the di¤erences in behaviour of judges selected by the legislator onthe one hand and the executive on the other, I take advantage of a naturalexperiment in judicial selection which occurred in Russian commercial courts in1990s. This section provides the background on procedures for selecting judgesin these courts, and then describes the data.
3.1 Institutional background8
Ine¤ective court system is widely considered a major deterrent to investmentand economic growth in Russia9 . In this section I �rst describe the functions,hierarchy and main institutional features of Russian commercial court system,and then turn to focus on procedures for selecting judges.
3.1.1 Russian commercial (arbitrazh) courts
Economic disputes between �rms, individual entrepreneurs and the state fallunder the jurisdiction of Russian arbitrazh courts. These are professional courtsorganized in a three-tier structure: courts of �rst instance (regional courts),appellate courts (okrug courts), and the Supreme Arbitration Court.Regional courts were created in 1991 in 81 of Russia�s 89 administrative
regions. They replaced the Soviet system of arbitration tribunals which haddealt with con�icts between state-owned enterprises under central planning.Although the old name was kept, the new system was set up quite di¤erentlyfrom the Soviet one10 .The jurisdiction of each regional arbitrazh court coincides with the admin-
istrative borders of the region11 . The plainti¤ is required to �le his suit inthe arbitrazh court of the region where the defendant is o¢ cially registered,preventing �venue shopping�. Each case is heard by a single judge, with theexception of bankruptcy suits and suits to invalidate a government sanction. Bylaw, these are heard by at least three judges together.
8This section describes the organization of Russian commercial courts between 1992 and2002. In 2003, several changes were introduced into the system. These are beyond the scopeof the current study.
9Relative to other transition economies, Russian �rms give very low scores to e¤ectiveness,fairness and reliability of Russian judiciary (Hellman et al (2000)). A recent survey ofbusinesses named legal reform the number two priority in Russia, second only to reduction ofcorruption in public sector (Biletsky et al (2002)).10See, for instance, Hendley (1998).11The exceptions are eight regions which do not have an �own�court.
11
Litigants unhappy with a decision of the regional court can appeal it to acorresponding okrug court of appeals, of which there are ten it total12 ;13 . Thesecourts were established in 1995. The jurisdiction of each appellate court in-cludes from 7 to 11 regional arbitrazh courts14 . In contrast with regional courts,all cases �led with courts of appeal are tried by at least three judges. Thereare no restrictions on the types of cases that can be appealed. Between 1995and 2002, litigants took roughly 5% of all regional court decisions to appellatecourts15 .Litigants can appeal decisions of okrug courts further to the Supreme Ar-
bitration Court of Russia. Yet, it selects and reviews only a small fraction ofsuits �led with it.In an e¤ort to isolate arbitrazh courts from the in�uence of regional and local
authorities, arbitrazh court system is o¢ cially �nanced solely from the federalbudget. Once a judge has been appointed to an arbitrazh court, he has tenureuntil he retires. The salary of a judge cannot be reduced throughout his career.Although initial appointment procedures for arbitrazh judges di¤er across re-gional courts and between regional and okrug courts, most of the time the �nalselection is done by a federal o¢ ce from a shortlist of candidates compiled bya committee of judges. Deciding which judges get promoted is the prerogativeof the presidential o¢ ce, again after a committee of judges had short-listed thecandidates.
3.1.2 Selection of judges
Between 1992 and 2002, Russian regional commercial courts experienced threedi¤erent procedures for the �nal selection and appointment of judges. First, bythe federal legislature, second, by the federal executive (the president of Russia),and third, by regional legislative assemblies16 .Legislature. By constitution, the federal legislature is in charge of writing
commercial law that the arbitrazh courts enforce. When regional arbitrazhcourts were created in 1992, the federal legislature, the Supreme Soviet, wasgiven the powers to select and appoint judges in 56 out of 81 of these courts17 .Regional assemblies. Regional assemblies are elected bodies in charge of pro-
ducing regional laws. They generally do not legislate in the area of commerce,which is the prerogative of the federal legislature. In 1992, regional assemblieswere put in charge of selecting judges in 25 regional commercial courts, thosenot covered by the federal legislature.
12An �okrug�is a large geographical division in Russia which includes several regions.13Before a decision of a regional court comes into force, a litigant who is unsatis�ed with it
can also request a re-consideration by a three judge panel of the same regional court.14The exception is Moscow okrug court which only covers two regional courts.15See The Supreme Arbitration Court (2002).16 Initial short-listing of applicants is done by �qualifying committees�, which consist of other
judges. This is similar to procedures used in many other countries.17That was the �rst democratically elected legislature of Russia�s post-Soviet period. Inside
the Supreme Soviet, three committees were put jointly in charge of selection of judges.
12
The 25 regions where these courts are located have an o¢ cial status of�autonomous.� This re�ects a greater political autonomy from Moscow thathad been granted to them because of their high share of ethnic population.The power to appoint arbitrazh court judges was one small feature in a packageof political privileges these regions enjoyed. Comparison of these regions withthe rest of Russia shows that they have lower levels of economic developmentand more pro-communist politics (see table 6).Executive. Towards the end of 1993, Russia adopted a new constitution
which drastically increased the powers of the president vis-a-vis the legislatureacross a wide range of policy issues. The new powers granted included rightsto veto legislation, dissolve parliament and appoint ministers with minimal par-liamentary involvement18 . One of the powers transferred to the president bythe constitution is that of appointing judges to all regional and appellate com-mercial courts. Thus, from 1994 onwards, the executive government has beenresponsible for judicial selection in commercial courts19 .These di¤erent procedures have led to a situation where judges appointed
by two di¤erent methods work side-by-side in every regional commercial court.In 56 non-autonomous regions, each judge is an appointee of either the federallegislature or the president. In 25 autonomous regions, each judge had beenappointed either by the regional assembly or by the president.
3.2 Data
My data come from three main sources. First, I put together the data onjudicial rulings by coding the texts of commercial court decisions made between1995 and 2002. These texts were obtained from Kodeks, a company supplyinglegal information. Second, the disputes were matched with sizes of �rms thatparticipated in them. The size data are from the Gnozis/Alba database andthe registry of the Federal Committee for Securities20 . Third, the method ofappointment for each judge was obtained from the documents published by theSupreme Soviet and the president of Russia.The sample contains 2633 decisions made by 759 regional judges working
in 56 out of 81 regional commercial courts. The study is limited to thesecourts because the rest do not supply names of judges in texts of decisions21 .All of these decisions had been reviewed by an appellate court. For eachdispute, the following information was coded by reading the text of the decision:the identities and types of plainti¤ and defendant, the nature of the dispute,
18See, for example, McFall (2002) for a historical analysis of how Russia�s 1993 constitutionwas created.19 In 1995, an amendment was introduced requiring the judicial candidates to be approved
by regional assemblies prior to a �nal presidential selection.20A more detailed description of Gnozis/Alba data can be found, for example, in Bessonova
et al (2003). Federal Securities Commission, (FKTsB) is Russian security market regulator.For access to its data see http://www.fcsm.ru.21Withholding of names may be associated with certain characteristics of courts. I will hope
to minimize contamination of my results by such selection by using of variation in appointmentprocedures within each court.
13
locations of regional and appellate courts, the winner in the regional court, thewinner in the court of appeals, and several other variables22 . A �reversal�wasde�ned as an instance when a regional court decision in favour of the plainti¤was overruled by a decision in favour of the defendant by the appellate court,and vice versa23 .Table 1 provides the breakdown of court cases in the sample by the type of
dispute. Breach of contract cases dominate with 40% of the sample, followedby tax disputes (25%), regulatory disputes (14%), ownership disputes (8%),and other disputes which include enforcement of court decisions by baili¤s,liquidation of �rms, etc.For each type of dispute, table 1 also presents the mean reversal rate of the
lower court decision by the court of appeals. With the exception of regulation,reversal rates do not signi�cantly di¤er across dispute categories, staying closeto the 19% sample average.Table 1 also shows win rates of plainti¤s in lower courts for each category of
dispute. In four out of the �ve categories, as well as in the sample overall, theaverage probability that the plainti¤ wins the case is not signi�cantly di¤erentfrom 50%. This matches the predicted plainti¤ win rates when selection intolitigation is characterized by divergent expectations, which I used to model thedecision to litigate in section 2.Table 2 shows the composition of all disputes in the sample according to the
types of litigants involved. These including government o¢ ces of the three levelsof Russian state hierarchy: federal, regional and local24 . Federal governmentcategory includes all state o¢ ces and departments which are a part of or reportto the executive branch of the federal government in Moscow. These includetax authorities, anti-trust regulators, environmental regulators, etc. Litigantsclassi�ed as regional government are those which are a part of or report to theexecutive branch of a regional government of each of 89 Russia�s regions. Theseinclude regional governor�s o¢ ce, regional ministries of �nance and economy,regional registration chambers, etc. Similarly, the local government category isde�ned to include all o¢ ces of city or village authorities.I divided all �rms that participated in disputes into small, medium and large.
This was done by matching the disputes to enterprise employment data drawnfrom Gnozis/ Alba database, which covers medium and large size enterprises,and �nancial reports for publicly traded companies published by the FederalSecurities Commission. I was able to match approximately 1/4 of all businessesto employment data, with smaller �rms not covered by either source. Forthe rest of the �rms, I used their legal classi�cation to determine their size
22These include information on whether the parties were present at appeal, whether theytook the regional decision to a three judge panel of the same court prior to appeal, and whetherone judge or a three judges made the initial regional court decision.23Cases in which regional court awarded partial relief were classi�ed as reversed if the
appellate court substantially changed the award. The number of such partial relief cases isvery small.24 In cases where more than one litigant is involved on either side of the suit, only the �rst
litigant is listed in the table. Such cases account for approximately 1% of the sample.
14
category25 .The small �rm category includes all companies with less than 150 employ-
ees (the bottom quartile of employment distribution among litigants that werematched to employment numbers), as well as all individual entrepreneurs, soloproprietorships, cooperatives, farms and full partnerships. Large �rms includeall publicly traded companies except those with less than 250 employees as wellas all other �rms with more than 650 employees (the upper half of the sizedistribution). The rest of the �rms are classi�ed as �other�26 .Table 2 shows no signi�cant di¤erences in reversal rates between cases in-
volving di¤erent litigant pairs. There are very few disputes between small andlarge �rms, with the bulk of enterprise against enterprise cases being between�rms of similar size. Therefore, my empirical identi�cation of judicial selectione¤ects focuses on law suits involving government o¢ ces, whose disputes withsmall and large enterprises comprise 35% of the sample.Using the name of the regional court judge published in the text of each
decision, I matched the case data to the political o¢ ce which had selectedthe regional judge in each case. I gathered this information from documentspublished by the Supreme Soviet and the o¢ ce of the president. Three dummyvariables were created corresponding to each method of selection. L = 1 if theregional court judge was appointed by the federal legislature, and 0 otherwise;similarly, for judges selected by the executive (X) and by regional assemblies(RA). Table 3 presents the breakdown of reversal rates of lower court decisionsby the category of appointment and types of litigants who participated and whowon in the lower court.
4 Empirical model
I shall now investigate empirically whether judges appointed by the legislatorfavour small �rms relative to judges appointed by the executive. I take ad-vantage of the fact that appellate court judges in Russia are all selected by theexecutive.The model in section 2 of this paper suggests that the legislator�s judges are
more sympathetic towards small �rms than the executive�s judges. If this isthe case, proposition 3 makes three predictions. First, the probability that adecision of a legislator�s judge is reversed will be greater if this decision is infavour of a small �rm. Second, if selection is the only factor responsible fordi¤erent preferences of appellate and lower court judges, the probability thata decision of a lower court judge appointed by the executive is reversed willbe independent of who the decision favoured. Third, if the main impact ofselection methods is on policy preferences rather than on competence of lower
25Legal classi�cations were obtained from texts of court decisions.26Alternative de�nitions of �small�and �large�have also been tested: this does not signi�-
cantly a¤ect the results. The results are also robust to leaving only individual entrepreneursin the �small�category, which helps address a concern that large Russian �rms may hide behinda small enterprise facade.
15
court judges, the probability that a decision of a legislator�s judge in favour ofa small �rm is reversed will be greater than the probability that a decision ofan executive�s judge in favour of a small �rm is reversed. The opposite will betrue for a decision in favour of a government branch.Let R = 1 if the decision of a lower court judge is reversed on appeal, and
0 otherwise; W = 1 if a small �rm won in the lower court, and 0 otherwise;let L and X be dummy variables that capture appointment of the judge by thelegislator and the executive respectively, and L �W and X � W interactionterms between the appointment variable and W ; let x be a set of controls, and� a probability distribution. I estimate the following model
p(R = 1) = �[�+ �(L�W ) + �(X �W ) + �L+ �x]: (5)
If the �rst of the three predictions above holds, then � > 0: If the secondprediction also holds it further constraints the data to � = 0: The third onefurther implies that � + � > � and � < 0: The �rst two are necessary con-ditions for establishing that the legislator�s appointees favour small �rms morethan those of the executive. The third one is a su¢ cient condition, but not anecessary one.
5 Empirical results
Table 4 contains basic results of estimating the model in (5). Robustness checksare in table 5.Column 1 in table 4 reports the results of estimating (5) for disputes between
small �rms and government o¢ ces. First, � is positive and signi�cant at 1%level, indicating that judges appointed by the legislator have their decisionsreversed signi�cantly more often if they are in favour of small �rms rather thana government agency. This implies that lower court judges appointed by thelegislator favour small �rms more than appellate court judges appointed by theexecutive.Second, � is not signi�cantly di¤erent from zero, implying that when the
lower court and the appellate court judge are appointed by the same method,there is no di¤erence in their policy preferences. This con�rms that divergencein preferences between lower and appellate court judges captured by � is causedby selection procedures rather than some other di¤erences between the two tiersof courts.Third, � + � is signi�cantly greater than �; and � is signi�cantly less than
zero. This means that the lower court judges appointed by the legislator havetheir pro-small �rm decisions reversed more often the lower court judges ap-pointed by the executive, and the reverse is true for pro-government decisions.This con�rms once more the result that judges appointed by the legislator favoursmall �rms more than those selected by the executive. It further says that themain impact of selection procedures is on policy preferences of judges ratherthan their competence.
16
In column 2, I estimate the same model controlling for changes over timeusing year dummies, for the location of the court in a non-autonomous region,and for attendance of appellate court hearing by the litigants. Year dummieshelp control for over time increases in the share of executive�s appointees amongthe lower court judges. Autonomous regions had never experienced legislativeappointments. Controlling for this helps rule out a possibility that specialcharacteristics of these regions are driving comparison between executive�s andlegislature�s judges.The results show that all three �ndings obtained in the �rst estimate are
robust to inclusion of these controls. Column 2m reports marginal e¤ects ofexplanatory variables. When a lower court judge selected by the executive isreplaced by that selected by the legislature, the probability that his decision infavour of a small �rm is reversed at appeal increases by 29 percentage points(�+��� in column 2m). This di¤erence in policy preferences of judges selectedby the two methods is substantial, given the average probability of a reversal inthe sample of 19%.The mix of judges selected by di¤erent methods varies by court. Although
judicial selection procedures changed exogenously and simultaneously for all re-gions, I am concerned that regional characteristics may have a¤ected the com-position of judges. To check whether such characteristics rather than selectionmethods are generating my results, I introduce dummies controlling for regional�xed e¤ects.Column 3 and 4 report estimates of conditional logit and linear probability
models, both with regional �xed e¤ects. These regressions con�rm the greatertendency among judges selected by the legislator to favour small �rms relativeto judges selected by the executive. This is despite the source of identi�cationhaving been reduced to the di¤erences in reversal rates for judges appointedby di¤erent methods that work within the same court. The magnitude of thee¤ect of legislative selection is similar in the linear model to that estimated bythe probit regression earlier.A word on control variables. In all regressions, regional as well as year
e¤ects are highly signi�cant, the latter picking up a strong downward trend inoverall reversal rates, possibly because lower court judges improve with time.Regression in column (2) shows that courts located in non-autonomous regionshave lower reversal rates. Controlling for case categories does not contributesigni�cant explanatory power27 .The results of table 4 unambiguously indicate that judges selected by the
legislature behave very di¤erently from judges selected by the executive govern-ment. In disputes between small �rms and government o¢ ces, the legislator�sjudges tend to favour small �rms more than the judges appointed by the exec-utive. Moreover, this di¤erence in preferences is quite large.
27The results of regressions with case category dummies are not reported here, but areavailable from the author.
17
5.1 Other explanations?
5.1.1 Anti-government judges
So far, the results have been based on the subset of disputes between small�rms and government agencies. Thus, my �ndings could also be consistentwith the hypothesis that the legislator�s judges favour government less than dothe executive�s judges, regardless of the type of �rm that litigates with it. Todistinguish between these two conjectures about the preferences of legislator�sjudges (pro-small �rms vs. anti-government), I estimate the same regressionmodel as before on disputes between large �rms and government o¢ ces, lettingW = 1 when a large �rm wins.Table 5 has the results in columns 2 (conditional logit) and 3 (linear prob-
ability). For comparison, column 1 repeats the last regression of table 4 forsmall �rms. All regressions include regional �xed e¤ects.In the subsample of disputes between large �rms and the government, both
� and � are not signi�cantly di¤erent from zero. This indicates that judgesappointed by the legislature do not favour large �rms any more than judgesappointed by the executive. Thus, the �ndings here indeed show that thelegislator�s judges are more favourable to small �rms, rather than against thegovernment in general, compared to the executive�s judges.
5.1.2 Time of appointment
Although I observe the legislator�s and the executive�s judges at the same time,the former had been selected in 1991-1992, while the latter �in 1993 or after.If the pool of judicial candidates changed over time, selection variables usedabove may proxy for such change. In particular, suppose that older Sovietschool judges are more oriented towards redistribution of resources from haveto have-nots. If their share among job candidates fell over time and some ofthis fall coincided with the switch in selection procedures, this could generatemy results even if selection procedures did not a¤ect judicial preferences.I take advantage of the history of Russian judicial appointments to identify
the e¤ect of time from the e¤ect of the method of appointment. Recall, thatin 1992-1993 the legislature appointed judges to 51 of 81 regional courts, whileregional assemblies appointed judges in remaining 25 regions. If time ratherthan method of appointment is responsible for my �ndings, judges appointedby regional assemblies will exhibit the same preferences as those selected by thelegislature.Unlike the federal legislature, regional assemblies in Russia do not legislate
in the area of commerce and economics28 . The theory developed in this study(section 2) predicts that when the o¢ ce in charge of selection does not write
28 In rare cases that regional assemblies do pass economic legislation, arbitrazh courts arecalled upon to check its legality relative to the federal law. This has resulted in a number ofcases in which laws of regional assemblies were disregarded by courts as contradictory to thefederal law. This makes the relationship between regional assemblies and commercial courtsin Russia similar to that between the executive o¢ ce and the courts.
18
laws for the judges, the incentives to choose judges that favour small �rms dis-appear. Therefore, if the observed di¤erences between legislative and executiveappointees are due to the method of appointment, judges appointed by regionalassemblies should not favour small �rms relative to executive�s judges.Let RA = 1 if the judge was appointed by a regional assembly. I estimate
the model in (5), modi�ed to include RA alone and interacted with the indicatorof a small �rm win W: I include regional dummies, restricting my identi�cationto comparison of judges that were appointed by di¤erent methods but workwithin the same court.Columns 4 and 5 present the results of a conditional logit and a linear prob-
ability estimations. The coe¢ cient on the interaction term RA�W is not sig-ni�cantly di¤erent from zero. The sum of the coe¢ cients on RA and RA�Wis not statistically di¤erent from zero and from �. This indicates that judgesappointed by regional assemblies in 1992-1993 have policy preferences that donot di¤er from those of judges appointed by the executive from 1993 onwards.Therefore, I con�rm that it is not the time, but the method of appointmentthat has the e¤ect on judicial preferences. The size of the gap between thepreferences of the legislator�s and the executive�s appointees is similar to thatobtained before: decisions in favour of small �rms made by the legislator�sjudges are 21 percentage points more likely to be reversed than decisions madeby the executive�s judges29 . For further comparison, column 6 reports the es-timate of this model on disputes between large �rms and government agencies.Again, I do not �nd that selection method has any e¤ect there.Thus, the e¤ect of selection procedures on judicial preferences is robust to
these two tests. These results lend further con�dence �rst, to the fact thatjudges chosen by the legislator favour small �rms more than do judges ap-pointed by the executive, and second to the fact that these di¤erences in policypreferences of judges are indeed the result of how these judges were selected,rather than some other coincidental change.
6 Concluding remarks
The paper examines the impact of judicial selection procedures on law enforce-ment. Around the world, the process of selecting judges typically involves eithera legislative or an executive branch of the government. I develop a model whichcompares these two types of procedures by contrasting incentives of the legisla-tor to those of the executive when appointing judges. It is based on a premisethat both political o¢ ces collect rents from �rms. The di¤erences in incentivesbetween the two o¢ ces arise because the executive�s policy is a subject to ju-dicial review, while the legislator�s is not. The model shows that, compared
29Autonomous regions, where regional assemblies appointed judge, are less developed andmore �pro-soviet� than the rest of Russia (see table 6). Thus, it is unlikely that regionalcharacteristics were responsible for the less �soviet� judges in these regions compared to therest in 1992-1993.
19
to the executive, the legislator selects judges who are more favourable towardslitigants that cannot pay rents.The implications of the model are tested on performance of Russian com-
mercial court judges, exploiting variation in judicial appointment proceduresacross time and regions, and using the reversal rate method developed in anearlier paper.Empirical �ndings show that judges appointed by the legislature favour small
�rms in disputes with government o¢ ces more than judges appointed by theexecutive. These results are consistent with the predictions of the theoreticalmodel. They are robust to controlling for court �xed e¤ects and to tests of twoimportant alternative hypotheses.The theory and empirical evidence presented here are a step towards under-
standing how institutional arrangements governing the judiciary can shape theprocess of law enforcement. A rare natural experiment in Russia has allowedme to compare the two judicial selection methods most commonly used aroundthe world, and to show that they can produce very di¤erent results.At the same time, the example of Russia is highly speci�c. In particular, the
results of the model here rely on the assumption that politicians are su¢ cientlyinterested in rents.Therefore, one broad implication of this study is that the e¤ect of a so-
ciety�s judicial institutions depends on institutional arrangements that governits politicians. This suggests that there is scope for further research on howdi¤erent types of political, judicial and regulatory institutions interact and thee¤ects of these interactions on economic performance.
20
Appendix A
Proof of proposition 1
Here I show that the expected payo¤ of the executive monotonically declineswith z: To do so, I argue that @�x=@z < 0 8 z:Recall that the executive�s expected payo¤ is
�X =1
2V
Z 1
�1
Z 1
�1p(Y > D j �)h( )g(Y )d dY � c� e(z): (6)
Denote p̂ �R1�1
R1�1p(Y > D j �)h( )g(Y )d dY: Di¤erentiating �X with
respect to z gives@�X@z
=@�X@e
@e(z)
@z+@�X@p̂
@p̂
@z: (7)
To sign this, �rst consider the �rst term of (7). Since the executive chooses eoptimally, it is zero. Second, the executive�s expected pro�t is increasing withthe probability that he will win in court, i.e.
@�X@p̂
=1
2V > 0 (8)
To di¤erentiate expected probability of winning at appeal, expand the latter to
p(Y > D j �) =Z ���z�(1��)Z+e�+P
�1h( )d : (9)
Thus,
@p̂
@z= ��
Z 1
�1
Z 1
�1h(�� �z � (1� �)Z + e� + P )h( )g(Y )d dY < 0 (10)
for all z: That is, the estimated probability that the executive wins the courtcase falls when the lower court favours the small �rm more. Therefore,
@�X@z
= �12V �
Z 1
�1
Z 1
�1h(���z�(1��)Z+e�+P )h( )g(Y )d dY < 0 (11)
for all z: Similarly, @�X=@Z < 0: QED.
21
Appendix B
Proof of Proposition 2
I show here that the legislator maximizes his payo¤ (the payment received fromthe executive for P = v) by selecting the lower court judge with z � z�: Recallthat the legislator�s expected payo¤ is
�L =1
2[�X(P = v)� �X(P = 0)]: (12)
Di¤erentiating with respect to z gives
@�L@z
=1
2[@�X@z
(P = v)� @�X@z
(P = 0)] (13)
From appendix A we have that
@�X@z
= �12V �
Z 1
�1
Z 1
�1h(�� �z � (1� �)Z + e(z) + P )h( )g(Y )d dY < 0:
(14)Substituting this into (13) gives
@�L@z
=1
2V �f
Z 1
�1
Z 1
�1[h(�� �z � (1� �)Z + �e(z; P ))h( )g(Y )d dY
�Z 1
�1
Z 1
�1h(�� �z � (1� �)Z + v + �e(z; P ))]h( )g(Y )d dY g
By assumption 2.2, when z < z�,
e(z; P )
8<: = 0 if P = v
> 0 and < v=� if P = 0:
By optimality of e(z; P ); @�2X=@e2 < 0: This implies @h(���z�(1��)Z+
�e(z; P ))=@ < 0: Thus,
h(�� �z � (1� �)Z + �e(z; P )) > h(�� �z � (1� �)Z + v): (15)
Therefore,@�L@z
> 0: (16)
When z � z�; e(z; P = 0) = v=�+ e(z; P = v), and, therefore
@�L@z
= 0: (17)
Thus, the legislator�s payo¤ is maximized when z � z�: QED.
22
Appendix C
Proof of Proposition 3
Let �Y denote the mean of the distribution of Y: Below I show that if the lowercourt judge favours the small �rm more than the appellate court judge, theprobability that the decision of the lower court judge is reversed at appeal ishigher if this decision were in favour of the small �rm than if it were in favourof the executive, as long as the variance of and � not too large relative to1=jD � �Y j (this is based on Shvets (2005)).Let the distribution of " be denoted by f("): When the last condition of
the proposition is satis�ed, the selection process implies that the distribution ofY among disputes which are litigated is symmetric around D regardless of thedistribution of Y in the population30 . Let k(Y ) denote this distribution: For agiven value of Y; p(Y +" < d) =
R d�Y�1 f(")d". Denote F (d�Y ) �
R d�Y�1 f(")d":
Thus, to prove that p(Y > DjY + " < d) > p(Y < DjY + " > d); I need toshow that R1
DF (d� Y )k(Y )dYR1
�1 F (d� Y )k(Y )dY�RD�1 [1� F (d� Y )]k(Y )dYR1�1 [1� F (d� Y )]k(Y )dY
> 0: (18)
Inequality (18) is satis�ed i¤
Z 1
D
F (d� Y )k(Y )dY �Z D
�1k(Y )dY
Z 1
�1F (d� Y )k(Y )dY+
+ [
Z D
�1F (d� Y )k(Y )dY ]2 � [
Z 1
D
F (d� Y )k(Y )dY ]2 > 0
Since k(Y ) is symmetric around D;RD�1 k(Y )dY = 1=2; the inequality above
simpli�es to Z 1
�1F (d� Y )k(Y )dY > 1
2: (19)
Since d > D , the symmetry of k(Y ) around D is a su¢ cient condition forinequality (19) to be satis�ed. QED.
30When the distribution of Y in the population is itself close to being symmetric around D,i.e. 1=jD � �Y j is very large, then proposition 3 holds regardless of variances of � and :
23
References
Ashenfelter, O., T. Eisenberg and S. Schwab (1995) "Politics and the Judiciary:the In�uence of Judicial Background on Case Outcomes," Journal of Legal Stud-ies, Vol. 24Besley, T. and A. Case (2003) "Political Institutions and Policy Choices:
Evidence from the United States", Journal of Economic Literature, Vol. 1, No.41Besley, T. and A. Payne (2003) "Judicial Accountability and Economic Pol-
icy Outcomes: Evidence from Employment Discrimination Charges," mimeo,LSEBessonova, E., K. Kozlov, and K. Yudaeva (2003) "Trade Liberalization,
Foreign Direct Investment, and Productivity of Russian Firms," CEFIR workingpaperBiletsky, S., J. D. Brown, J. Earle, J. Khaleeva, I. Komarov, O. Lazareva, K.
Sabirianova, and C. Sprenger (2002), "Inside the Transforming Firm: Reporton a Survey of Manufacturing Enterprises in Russia," CEFIR working paperBlack, B., R. Kraakman and A. Tarassova (2000) "Russian Privatisation
and Corporate Governance: What Went Wrong?" Stanford Law Review, Vol.52, No. 5Canon, B. (1972) "The Impact of Formal Selection Process on the Charac-
teristics of Judges� Reconsidered," Law and Society Review, Vol. 6Demirguc-Kunt, A. and V. Maksimovic (1998) "Law, Finance and Firm
Growth," Journal of Finance, Vol. 52, No. 6Glick, H. and C. Emmert (1987) "Selection Systems and Judicial Charac-
teristics: The Recruitment of State Supreme Court Justices," Judicature, Vol.70Gray, C. and K. Hendley (1997) "Developing Commercial Law in Transition
Economies: Examples from Hungary and Russia," in The Rule of Law andEconomic Reform in Russia, K. Pistor and J. Sachs, eds., Boulder: WestviewPressHanssen, F. A. (1999) "The E¤ect of Judicial Institutions on Uncertainty
and the Rate of Litigation: The Election versus Appointment of State Judges,"Journal of Legal Studies, Vol. 1, No. 28Hanssen, F. A. (2000) "Independent Courts and Administrative Agencies:
An Empirical Analysis of the States," Journal of Law, Economics, and Organi-zation, Vol.16, No. 2Hanssen, F. A. (2004a) "Is There A Politically Optimal Level Of Judicial
Independence?" American Economic Review, Vol. 94Hendley, K. (1998) "Remaking an Institution: The Transition in Russia from
State Arbitrazh to Arbitrazh Courts," The American Journal of ComparativeLaw, Vol. 46Johnson, S., D. Kaufmann, and P. Zoido-Lobaton (1998) "Regulatory Dis-
cretion and the Uno¢ cial Economy," American Economic Review, Vol. 88, No.2
24
Johnson, S., J. McMillan and C. Woodru¤ (2002) "Property Rights andFinance," American Economic Review, Vol. 92, No. 5Keefer, P. and S. Knack (1995) �Institutions and Economic performance:
Cross-country Tests Using Alternative Institutional Measures,�Economics andPolitics, Vol. 7Kendall, C. (1997) "Appointing Judges: Australian Judicial Reform Propos-
als In Light Of Recent North American Experience," Bond Law Review, Vol.9Krueger, A. (2002) "Growth and Reform in Russia," Address, Conference
on Post-Communist Economic Growth, MoscowLa Porta, R., F. Lopez-de-Silanes, A. Shleifer and R. Vishny (1998) "Law
and Finance," Journal of Political Economy, Vol. 106, No. 6La Porta, R., F. Lopez-de-Silanes, A. Shleifer and R. Vishny (1997) "Legal
Determinants of External Finance," Journal of Finance, Vol. 52, No. 3Landes, W. and R. Posner (1975) "Independent Judiciary in an Interest-
group Perspective," Journal of Law and Economics, Vol. 18Maskin, E. and J. Tirole (2004) "The Politician and the Judge: Account-
ability in Government," American Economic Review, Vol. 94, No. 4Mauro, P. (1995) "Corruption and Growth," Quarterly Journal of Eco-
nomics, Vol. 110, No. 2McFaul, M. (2002) "Evaluating Yeltsin and His Revolution," in A. Kuchins
(ed.), Russia After the Fall, Carnegie Endowment for International PeacePersson, T. and G. Tabellini (2000) Political Economics: Explaining Eco-
nomic Policy, The MIT PressPistor, K., M. Raiser, and S. Gelfer (2000) �Law and Finance in Transition
Economies,�Economics of Transition, Vol. 8Priest, G. and B. Klein (1984) "The Selection of Disputes for Litigation,"
Journal of Legal Studies, Vol. 13Revesz, R. (1997) "Environmental Regulation, Ideology and the D.C. Cir-
cuit," Virginia Law Review, Vol. 83Shapiro, M. (1995) "Western Common Law Democracies: The United
States," in C.N. Tate and T. Vallinder (eds), The Global Expansion of Judi-cial Power, New York University Press, New York and LondonShvets, J. (2005) "Judicial Policy Preferences in Russian Commercial Courts,"
mimeo LSESiegelman, P. and J. Waldfogel (1999) "Toward a Taxonomy of Disputes:
New Evidence through the Prism of the Priest/Klein Model," Journal of LegalStudies, Vol. 28Skordaki, E. (1991) "Judicial Appointments," The Law Society, Research
Study 5Spitzer, M. and E. Talley (2000) "Judicial Auditing," Journal of Legal Stud-
ies, Vol. 28Thomas, C. (1997) "Judicial Appointments in Continental Europe," in Ju-
dicial Appointments Commissions: The European and North American Experi-ence and the Possible Implications for the United Kingdom, Lord Chancellor�sDepartment Research Series, No.6/97
25
Xu, C. and K. Pistor (2002) "Law Enforcement under Incomplete Law: The-ory & Evidence from Financial Market Regulation," STICERD Discussion paperTE/02/442
26
Cas
e ty
peN
Mea
nSE
Mea
nSE
Con
trac
t1,
084
0.19
(0.0
1)0.
52(0
.02)
Tax
673
0.20
(0.0
2)0.
51(0
.02)
Reg
ulat
ion
365
0.15
(0.0
2)0.
39(0
.03)
Ow
ners
hip
212
0.22
(0.0
3)0.
50(0
.03)
Oth
er29
90.
17(0
.02)
0.45
(0.0
3)
Tot
al2,
633
0.19
(0.0
1)0.
49(0
.01)
Dec
isio
n re
vers
al a
t app
eal
(1=r
ever
sed,
0=u
phel
d)Pl
aint
iff w
in ra
te
(1=p
lain
tiff w
on, 0
=def
enda
nt w
on)
Tab
le 1
Bre
akdo
wn
of c
ases
by
disp
ute
type
NM
ean
SEA
ll en
terp
rise
s1,
825
0.20
(0.0
1)Sm
all e
nter
pris
es50
70.
19(0
.02)
Larg
e en
terp
rise
s42
60.
20(0
.02)
Smal
l vs.
larg
e64
0.17
(0.0
5)Si
mila
r siz
e en
terp
rise
s74
40.
16(0
.01)
Tot
al2,
633
0.19
(0.0
1)
Ent
erpr
ise
vs. g
over
nmen
t
Ent
erpr
ise
vs. e
nter
pris
e
Dec
isio
n re
vers
al b
y ap
pella
te c
ourt
(R)
(1
=rev
erse
d, 0
=uph
eld)
Tab
le 2
Bre
akdo
wn
of c
ases
by
type
of l
itiga
nt
(1)
(2)
(3)
(4)
(5)
Judg
e ap
poin
ted
byA
ll ca
ses
All
case
s w
/ sm
all f
irm
sSm
all f
irm
won
Smal
l fir
m lo
st(3
) - (4
)
Fede
ral l
egis
latu
re (L
)0.
170.
160.
320.
056
0.26
(0.0
1)(0
.03)
(0.0
5)(0
.02)
(0.
06)*
*
Pres
iden
t (X
)0.
180.
180.
200.
160.
04(0
.01)
(0.0
3)(0
.04)
(0.0
3)(0
.05)
Reg
iona
l ass
embl
y (R
A)
0.23
0.24
0.25
0.23
0.02
(0.0
2)(0
.04)
(0.0
6)(0
.05)
(0.0
9)
N2,
633
571
240
331
571
Not
es:
* si
gnifi
cant
at 5
%;
** s
igni
fican
t at 1
%.
Tab
le 3
Jud
icia
l sel
ectio
n an
d re
vers
al r
ates
Per
cent
of d
ecis
ions
rev
erse
d by
app
ella
te c
ourt
Prob
itC
ondi
tiona
l lo
git
Line
ar
prob
abili
tyC
oeff
Coe
ffM
EC
oeff
Coe
ff(1
)(2
)(2
m)
(3)
(4)
LxW
(
λ)
1.26
**1.
32**
0.41
**2.
35**
0.29
**(0
.26)
(0.2
7)(0
.09)
(0.5
8)(0
.06)
XxW
(
ξ)
0.14
0.09
0.02
0.12
0.04
(0.2
0)(0
.21)
(0.0
5)(0
.40)
(0.0
5)
L
(β
)-0
.68*
*-0
.49+
-0.1
1+-0
.81
-0.1
0*(0
.25)
(0.2
7)(0
.06)
(0.5
9)(0
.05)
Non
-aut
onom
ous
regi
on-0
.42*
-0.1
1*
(0 =
aut
onom
ous,
1 =
not
)(0
.22)
(0.0
6)
Lose
r pre
sent
at a
ppea
l0.
190.
050.
550.
03(0
.19)
(0.0
4)(0
.36)
(0.0
5)
Win
ner p
rese
nt a
t app
eal
0.13
0.03
0.35
0.03
(0.2
0)(0
.05)
(0.3
7)(0
.05)
Con
stan
t-0
.99*
*-0
.87*
*0.
00(0
.13)
(0.3
4)(0
.26)
Yea
r dum
mie
sN
YY
Reg
iona
l fix
ed e
ffec
tsN
YY
N40
032
639
9
Not
es: S
tand
ard
erro
rs in
par
enth
eses
; +
sign
ifica
nt a
t 10%
, * s
igni
fican
t at 5
%; *
* si
gnifi
cant
at 1
%L
= 1
if ju
dge
sele
cted
by
legi
slat
or; X
= 1
if ju
dge
sele
cted
by
exec
utiv
e; W
= 1
if d
ecis
ion
in fa
vour
of s
mal
l fir
m
Tab
le 4
Sel
ectio
n of
judg
es: B
asic
res
ults
Pro
babi
lity
of a
rev
ersa
l in
disp
utes
bet
wee
n sm
all f
irm
s an
d go
vern
men
t
395
NY
Prob
it
Smal
lLa
rge
Larg
eSm
all
Smal
lLa
rge
Line
ar
prob
abili
tyC
ondi
tiona
l lo
git
Line
ar
prob
abili
tyC
ondi
tiona
l lo
git
Line
ar
prob
abili
tyLi
near
pr
obab
ility
(1)
(2)
(3)
(4)
(5)
(6)
LxW
(
λ)
0.29
**0.
940.
092.
28**
0.30
**0.
08(0
.06)
(0.6
1)(0
.06)
(0.5
7)(0
.06)
(0.0
6)
XxW
(
ξ)
0.04
-0.2
5-0
.05
0.18
0.03
-0.0
4(0
.05)
(0.4
8)(0
.06)
(0.3
8)(0
.05)
(0.0
6)R
AxW
-0.0
00.
02-0
.01
(0.5
1)(0
.07)
(0.0
9)
L
(β
)-0
.10*
-0.8
8-0
.11+
-0.8
5-0
.07
-0.1
0(0
.05)
(0.6
2)(0
.06)
(0.5
9)(0
.05)
(0.0
7)R
A0.
570.
03-0
.10
(0.5
5)(0
.07)
(0.0
9)
Lose
r pre
sent
at a
ppea
l0.
030.
50.
070.
65*
0.07
+0.
07+
(0.0
5)(0
.39)
(0.0
4)(0
.30)
(0.0
4)(0
.04)
Win
ner p
rese
nt a
t app
eal
0.03
0.13
0.01
0.25
0.05
0.03
(0.0
5)(0
.39)
(0.0
4)(0
.30)
(0.0
4)(0
.04)
Con
stan
t0.
000.
64**
0.07
0.41
+
(0.2
6)(0
.18)
(0.2
7)(0
.24)
Yea
r dum
mie
s
Reg
iona
l fix
ed e
ffec
ts
N39
922
134
242
650
642
5
Not
es: S
tand
ard
erro
rs in
par
enth
eses
; +
sign
ifica
nt a
t 10%
, * s
igni
fican
t at 5
%; *
* si
gnifi
cant
at 1
%,
Y Y
Tab
le 5
Sel
ectio
n of
judg
es: R
obus
tnes
sP
roba
bilit
y of
a r
ever
sal i
n di
sput
es w
ith g
over
nmen
t
Non
-aut
onom
ous
regi
ons
Aut
onom
ous
regi
ons
Judg
e ap
poin
ted
byFe
dera
l leg
isla
ture
Reg
iona
l ass
embl
y
(199
2-93
)
Out
put p
er c
apita
7843
6149
(R
oubl
es, 1
995)
(668
)(1
098)
Infa
nt m
orta
lity
16.4
19.8
(p
er 1
000
of p
eopl
e, 1
998)
(0.4
)(1
.3)
Free
dom
of m
edia
36
.925
.9
(ind
ex, 2
000)
(0.9
)(1
.4)
Vot
es in
favo
ur o
f com
mun
ist c
andi
date
s 12
.115
.6
(%, 1
993
parl
iam
enta
ry e
lect
ions
)(0
.6)
(2.4
)
Vot
es in
favo
ur o
f the
new
con
stitu
tion
58.8
55.8
(%
, 199
3 re
fere
ndum
)(1
.4)
(3.4
)
Reg
iona
l com
mer
cial
cou
rt d
ecis
ions
reve
rsed
16.6
22.1
(%
, pre
sent
sam
ple,
199
5-20
02)
(0.8
)(1
.3)
Num
ber o
f reg
ions
5625
Not
es:
All
diffe
renc
es b
etw
een
two
colu
mns
are
sig
nific
ant a
t lea
st a
t 5%
leve
l, w
ith th
e ex
cept
ion
of v
otes
for
new
con
stitu
tion.
Tab
le 6
Sel
ecte
d ch
arac
teri
stic
s of a
uton
omou
s and
non
-aut
onom
ous r
egio
ns