Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Wage Inflation, Electoral Uncertainty and the Exchange Rate Regime: Theory and UK Evidence Author(s): George S. Alogoskoufis, Ben Lockwood, Apostolis Philippopoulos Source: The Economic Journal, Vol. 102, No. 415 (Nov., 1992), pp. 1370-1394 Published by: Blackwell Publishing for the Royal Economic Society Stable URL: http://www.jstor.org/stable/2234795 . Accessed: 18/05/2011 09:50 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at . http://www.jstor.org/action/showPublisher?publisherCode=black. . Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. Blackwell Publishing and Royal Economic Society are collaborating with JSTOR to digitize, preserve and extend access to The Economic Journal. http://www.jstor.org TheEconomicJournal,I02 (NovemberI992), I370-I394 Printedin GreatBritain WAGE INFLATION, ELECTORAL UNCERTAINTY AND THE EXCHANGE RATE REGIME: THEORY AND UK EVIDENCE* GeorgeS. Alogoskoufis, Ben Lockwood andApostolisPhilippopoulos Recent developments in the theory of inflation stress the limited ability of governments to pre-commit to price stability. When governments can create unanticipated inflation in order to reduce unemployment, expected inflation will rise to the point where the marginal cost of inflation to the authorities is so high, that they will have no further incentive to create unanticipated inflation (Kydland and Prescott, I977; Barro and Gordon, I983a, b). Applications of this approach include the partisan model of inflation (Alesina, I987) and the exchange-rate-regime model of inflation (Giavazzi and Giovannini, I987), the latter being an extension of Rogoffs (I985) solution of appointing a conservative central banker. The partisan model emphasises that parties have different preferences for inflation and unemployment. The party that is relatively more averse to unemployment (say the 'socialists') will have stronger incentives to inflate. As these incentives will be anticipated by wage-setters, 'socialist' parties will be associated with higher expected and actual inflation than 'conservative' ones, although they will not be able to affect unemployment systematically, except in the period following an election. In such a period, with nominal wages having been set before the announcement of the election result, if there is a socialist (conservative) victory, inflation will be higher (lower) than wage setters anticipated, and thus unemployment will temporarily fall (rise)*1 The exchange-rate-regime model captures one of the institutional mechanisms that can help avoid the unpleasant outcome of high equilibrium inflation without any employment benefits. The first institutional solution, investigated by Rogoff (I985), is the appointment of independent inflation-averse central bankers. If wage setters know that monetary policy is in the hands of an independent anti-inflationary monetary authority, then their expectations will reflect this. As a result inflation will remain low. This solution has an almost perfect analogue in fixed exchange rate regimes (Giavazzi and Giovannini, * We have benefited from the comments and suggestions of two anonymous referees and participants in a seminar at Birkbeck. We would also like to thank the ESRC for financial support under the programme on 'The Macroeconomics of Imperfect Product and Labour Markets' at Birkbeck. Alogoskoufiswould also like to acknowledge financial assistance from the CEPR programme in International Macroeconomics, supported from the Ford Foundation, the Sloan Foundation, and the SPES Programme of the EC Commission. Ben Lockwood would also like to acknowledge financial support from the Leverhume Trust. 1 This is because in election years, inflationary expectations are a weighted average of the expectations under the 'conservatives' and the 'socialists', the weights being the respective probabilities of election victory. See below for further elaboration of this point. Surveys of recent theories of the interaction between macroeconomics and politics can be found in Alesina and Tabellini (I988), Alesina (I988) and Persson and Tabellini(I990). I1370 ] [NOV. I992] ELECTORAL UNCERTAINTY AND INFLATION I37I I987; Giavazzi and Pagano, I988). Participation in such a regime, in which monetary policy is determined by an anti-inflationary foreign central bank, 'ties the hands' of domestic policymakers, who cannot determine their monetary policy independently. Again, this is reflected in the expectations of domestic wage setters, and thus the domestic economy ends up with the same average inflation as the rest of the economies participating in the system. The present paper attempts two things. First, it generalises the partisan model to allow for unemployment persistence. This extension is non-trivial because if unemployment persists, any reduction in unemployment will have future as well as current benefits for the party in power. Then, the time-horizon of the government, i.e. the time to the next election, will affect its incentives to create unanticipated inflation. It seems important to investigate this extension as unemployment persistence is a generalised feature of OECD economies (Bean et al., I986; Blanchard and Summers, I986; Alogoskoufis and Manning, I988a, b). Second, we investigate empirically the applicability of this 'extended' partisan model, and the related fixed-exchange-rate-regime model, to the postwar experience of the United Kingdom. The United Kingdom is particularly suitable for such an investigation as it is perceived to have had a rather polarised political system in the post-war period. In addition, it has had long experiences with both fixed and independently managed exchange rates. The ideological differences between the Conservative and the Labour Parties have been considered rather sharp, and both parties have held office for extended periods. Thus, the United Kingdom makes an ideal candidate for testing the implications of the partisan approach to inflation and the solution of entering a fixed exchange rate regime. The way we analyse these issues is to assume that wage setters set nominal wages for one period (in our case a year) in advance, in order to achieve an employment target (Gray, I976; Fischer, I977). We depart from Alesina (I987) and Rogoff (I985), in that we assume that this employment target is a combination of last period's employed (the 'insiders') and the unemployed (the 'outsiders'), as suggested by recent 'insider-outsider' theories (Blanchard and Summers, I986; Lindbeck and Snower, i986).2 Because of the dependence of the wage setters' target on last period's unemployed, changes in unemployment persist, and the unemployment rate only gradually returns to its equilibrium rate.3 This extension is analytically important, as it transforms the optimisation problem of governments into a genuinely dynamic one. The government knows that if it manages to reduce current unemployment, this reduction in unemployment will persist in the future. Thus, in deciding whether to inflate or not, they take into account the possible benefits from a reduction in 2 Models of this variety, however, cannot explain the outward shift of the European Beveridge curve, since they do not distinguish between the long-term and short-term unemployed. As a result, they do not show how the search behaviour of the unemployed outsiders may affect unemployment persistence. 3 Alesina and Rogoff implicitly assume that the wage-settler's employment target is fixed and less than the government's target. I372 THE ECONOMIC JOURNAL [NOVEMBER unemployment not only in the present, but in the future as well. This means that the incentives of political parties to create inflation differ according to whether they are near the beginning or the end of their term of office. We demonstrate two main results: First, if the 'socialists' care more about unemployment than the 'conservatives', they generate higher inflation at any point in their term of office. This is a generalisation of the basic Alesina (I987) result for the dynamic case. Our second main result is that, if the preferences of political parties for inflation and unemployment are close enough, inflation will be higher for both parties in the period following an election (which we call an election period) than in non-election periods. This latter result is very different from the case without unemployment persistence examined by Alesina (i 987) and Alesina and Sachs (i 988). In these papers, if the preferences of parties are the same, inflation will be the same in non-election and election periods. In our case inflation is higher in election periods because the persistence in the reduction in unemployment by unanticipated inflation will entail higher expected future benefits for the government near the beginning of its term of office than one near the end. As this is anticipated by wage setters, inflationary expectations are higher in the first part of a government's term. We test this model for the British economy for the period between I952-90, allowing for separate effects under fixed and managed exchange rates. The tests are based on structural wage equations. To the extent that we concentrate directly on structural wage equations, we provide an alternative to the reduced form testing strategy of Alesina and Sachs (i988) and Alesina and Roubini (I990). Our first empirical finding is that the exchange rate regime matters greatly for expected price inflation and hence nominal wage growth. One cannot reject the hypothesis that expected price inflation in the Bretton Woods period of fixed exchange rates has been independent of unemployment, the party in power and the timing of elections. In addition, in the post-Bretton Woods regime of managed exchange rates, a regime that has restored monetary sovereignty to UK governments, expected inflation has been both higher and more persistent, as it seems to have depended positively on the unemployment rate. This positive dependence of expected inflation on the lagged unemployment rate, due to the Barro-Gordon effect, fully offsets the traditional 'Phillips Curve-type' negative dependence of wage inflation on lagged unemployment. Our second empirical finding is that even in the post-Bretton Woods regime of managed exchange rates, there are very small differences between Labour and Conservative administrations in non-election years. The parties have been perceived to have the same incentives to create unanticipated inflation in order to reduce unemployment. On the other hand, elections are associated with higher expected inflation for both of them, as our theoretical model predicts, but Labour administrations seem to be slightly more inflationary in such periods. However, differences between Labour and the 'pre-Thatcher' Conservatives are not statistically significant. When we allow for the governments of Mrs Thatcher to constitute a different regime, we find the following additional I992] ELECTORAL UNCERTAINTY AND INFLATION I373 results: First, that there are no significant differences in expected inflation between Labour and pre-Thatcher Conservative governments, either in nonelection or in election years. As predicted by our model, immediate postelection years are associated with significantly higher inflation than nonelection ones. Our second finding, however, is that there is no separate effect of elections under the Thatcher governments. Yet, in non-election years these governments have been associated with as high and persistent inflationary expectations as the pre-Thatcher ones, Conservative and Labour alike. According to these findings, the anti-inflationary contribution of the Thatcher regime consisted in eliminating the electoral cycle in inflation that was pervasive-after the collapse of the Bretton Woods regime of fixed exchange rates. In conclusion, the results suggest strong support for the exchange-rateregime model of inflation, and our dynamic-partisan model. Under managed floating, expected price inflation and nominal wage growth were higher and more persistent than under Bretton Woods. They did not depend on the identity of the party in power, but, with the exception of the Thatcher governments, only on the point in the electoral cycle. Between i968 and i979 they were higher in immediate post-election periods, and lower before elections. After I979 they have been independent of the electoral cycle. The rest of the paper is as follows: Section I presents the theoretical model. This generalises the rational-partisan model of Alesina (I987), to allow unemployment persistence to affect the optimal choice of inflation rates by governments. It also draws on the exchange-rate-regime model of Giavazzi and Giovannini (I987). Section II presents econometric tests for the UK economy and the results are summarised and discussed in the final Section III. I. THE MODEL The main purpose of this section is to generalise the rational-partisan theory of Alesina (i 987), to allow for the effects of unemployment persistence. We then combine this dynamic rational-partisan-theory with the Giavazzi and Giovannini (i987) fixed-exchange-rates version of the conservative-central- banker theory of Rogoff (I985) .' We assume that there are two types of agent, wage-setters and political parties. Of the latter, there are two:' 'Labour' and 'Conservative'. Wagesetters unilaterally choose the nominal wage every time period, and, under managed exchange rates, the party in power controls monetary (or exchange rate) policy to determine the inflation rate. None of the parties can pre-commit to low inflation under managed floating.6 The two parties may differ in their 4 The alternative approach to political business cycles views politicians as solely office motivated so that they engineer pre-election booms to boost their electoral chances. See Rogoff and Sibert (1988). For a survey of the political business cycles literature, see Alesina (I988) and Persson and Tabellini (I990). 5 This seems a reasonable assumption for the United Kingdom: although there are of course other political parties, none of them has had a serious chance of forming a government in the post-war period. 6 Officially this is a floating and not a managed floating regime. However, in practice, pure floating has operated infrequently and for short periods, as the authorities in most cases seemed to be pursuing some form of exchange rate target through monetary policy. I374 THE ECONOMIC JOURNAL [NOVEMBER Election I -I| I t Policy set Policy set Policy set ~~ ~~~~~~~~~t ; t Wage Wage Wage set set set (non-election year) (election year) Fig. i. w oTime (non-election year) The timing of events. relative evaluation of unemployment and inflation: Labour are systematically at least as averse to unemployment as the Conservatives. Under fixed exchange rates, monetary policy and inflation are determined abroad, either in a cooperative fashion, or by a foreign central bank that acts as a leader in the system.7 The timing of events is as shown in Fig. i above. Time is divided up into discrete periods, or years. Wage setters sign annual nominal wage contracts at the beginning of each year, before monetary and exchange rate policy are chosen for that year. We suppose for simplicity that elections take place every second year (but none of the qualitative results below are sensitive to this assumption), and that the election takes place after wages have been set for that year. Thus, in election years nominal wage contracts are signed before the outcome of the election becomes known. In non-election years there is no such electoral uncertainty. Finally, the probability of any party being elected, q, is assumed to be independent of the government's choice of monetary and fiscal policy, although we initially allow q to depend both on the identities of the parties and the identity of the incumbent.8 Wage-setters know the exchange rate regime, and the identity and preferences of the incumbent party. They take this information into account in forming their expectations. They aim9 to achieve an employment target that depends asymmetrically on those who have been employed in the previous period ('insiders') and those who have been unemployed ('outsiders'). Finally, employment is determined by competitive firms. 7 Here we implicitly assume that the costs of abandoning an international agreement, like a fixed exchange rates regime, make the commitment to low inflation more credible than the unilateral fixing of the exchange rate, or other domestic commitments to non-inflationary monetary policy. Also, it is now widely accepted that fixed exchange rate regimes operate asymmetrically (see Eichengreen, I985, I989; Giavazzi and Giovannini, I987; Giovannini, I989 and others). For example during the classical gold standard (I890-I9I4) the Bank of England was to a large extent determining the aggregate stance of monetary policy, while a similar role was played by the Fed in the heyday of Bretton Woods, and is played by the Bundesbank in the EMS. In addition there is now considerable evidence that such fixed exchange rate regimes have been successful in containing average inflation and inflation differentials (see Alogoskoufis and Smith, I991; Alogoskoufis, I992). 8 The exogeneity of q to policy is a standard assumption in the RPT literature, and can be justified by rational expectations on the part of the private sector; with such rational expectations, electoral 'bribes' will be discounted fully and hence have no effect on q. We also assume, for simplicity, that the timing of elections is exogenous. 9 See Blanchard and Fischer, I989; Chapter 9, for more complicated objective functions of wage setters. I992] ELECTORAL UNCERTAINTY AND INFLATION I375 I. i. Wage and EmploymentDetermination At time t demand for labour, ld, is a decreasing function of real wages (all variables, with the exception of the unemployment rate will be in logs): ld 6 (Wt _Pt _AIt) n( where, e is the elasticity of labour demand, wt is the nominal wage, Pt the price level, and Jutis a productivity shock. The growth rate of the productivity shock is equal to g. Following Blanchard and Summers (I986); Lindbeck and Snower (I986) and Alogoskoufis and Manning (I 988 a), we assume that nominal wages are set by 'insiders', who aim to achieve an employment target lt. This is a weighted average of the labour force n, and those who are employed at the time the contract is being drawn up, It-,. Thus, It = Alt-,+ (i-A) n; o A I, (2) where A measures the extent to which the unemployed are disenfranchised from the labour market.10 Nominal wage contracts are signed at the beginning of each period so as to make the expected demand for labour (i), equal to the employment target of wage-setters (2). If the information set of wage-setters includes all values up to and including period t- i, then nominal wages are given by wt = g + lit1 + Et-lpt- I -A A n-e -It-1' (3) where Et_1 denotes a mathematical (rational) expectation conditional on the information set available at the end of t- i. Lagging (i) once, solving for IAtand substituting the solution in (3), after some rearrangement, we end up with, = g+Et_1ApAXwt p - ut-1, (4) where A is the first-difference operator, and ut-, is the lagged unemployment rate, defined as ut1_ n-it-1, where n is the log of the total labour force. Equation (4) defines the expectations augmented 'Phillips curve' in this model."l Ex post, employment is determined by labour demand (i). Using (i), (2) and (4), unemployment is given by, ut = Aut_1- e(Apt-EEt- Apt). (5) 10 If A equals one, only the employed (insiders) matter for wage setting. If A equals zero both the employed and the unemployed get the same weight (Blanchard and Summers, I986). " To avoid the theoretical possibility of negative unemployment, one could define n to be the labour force corrected for some measure of frictional unemployment. The analysis that follows would not change, the only difference being a slight change in the definition of the state variables. Such an amendment would, however, require the use of a particular model of the frictional unemployment rate, and in the interests of economy we have decided to stick to the definition in the text, which is equivalent to assuming a constant frictional unemployment rate. I376 THE ECONOMIC [NOVEMBER JOURNAL From (5), unanticipated inflation reduces unemployment, and the higher is A, the higher the persistence of this reduction."2 1.2. Inflation, Unemploymentand Political Parties We assume that both the Conservative party (denoted by the superscript c) and the Labour party (denoted by the superscript 1) prefer full employment and low inflation. In particular, when in power, the loss from unemployment and inflation to party i = c, I in period t is, Gi=(,pt -,Apt )+ a ut; aC al. (6) When out of power, the payoff to any party is zero. Note from (6) that the parties may differ only in their degree of aversion to unemployment. In what follows we allow the target level of inflation to depend linearly on the lagged domestic inflation rate. I > 0 > o. (7) ?+ Pt-,; In (7) we assume that the target inflation rate of the authorities depends positively on past inflation, as the higher the inherited inflation rate, the more costly the policy adjustment required to reduce it to an arbitrarily low level 7T0. Thus, governments are allowed to modify their target rate to take account of these costs. 0 is an indicator of the cost of adjustment of policy. Under a managed exchange rate regime, the party in power chooses current and future inflation to minimise the discounted expected present value of losses (6). We assume fully rational behaviour, in that each party takes into account the probability of losing (and subsequently regaining) power in election years. From (5) above, current unanticipated inflation affects future unemployment. Also, from (7) above, current inflation affects future inflation targets. Thus, the model has structural dynamics, unlike the theoretical models of Alesina (I987) and Alesina and Sachs (I988), who add lags only at the empirical stage. Our dynamic approach is preferable for at least two reasons. First, it is more realistic. Empirically, unemployment (and inflation under flexible exchange rates) is highly serially correlated. As we shall see below, unemployment dynamics imply that in the absence of pre-commitment, the optimal current policy of the government will depend on the 'state variables' of lagged unemployment and lagged inflation. Thus, the theoretical model will produce dynamic econometric equations directly, and there will be no need to add on extra dynamics at the estimating stage. Second, and equally importantly, the inclusion of unemployment dynamics changes some of the qualitative predictions of the RPT model. APt 1.3. A Dynamic Game betweenWage-Settersand Political Parties Because of the structural dynamics, it is convenient to interpret the model as a dynamic game, with three players, the two parties and the wage-setters (see Basar and Olsder, I982, for a general discussion of dynamic games). The party in power chooses Apt, and wage-setters 'choose' Et-, Apt. For wage setters, 12 In the extreme case when A = i, the unemployment rate stays at the new lower level until a new shock displaces it again. In other words, it displays 'hysteresis'. ELECTORAL I992] UNCERTAINTY AND INFLATION I377 forming rational expectations is then equivalent to minimising a loss function W = (Apt-Et-, Apt) 2.13 It appears at first sight that, at time t, we have two state variables, ut-1 and Apt-1.However, we can define (8) APt-, i t = Et-1 APt- To-A0 pt-, where 7i denotes the expected deviation of inflation from the target of the authorities. Substituting (8) in (6), the players' one period loss functions can be written as a function of current variables only: Gi = (7Tc)2?oxiut,and W = (gT-7T1)2. So, the problem is re-formulated into one with only one state variable, ut-1, whose change is described by the state equation from (5). gt = APt-7TO-0 (s') (-t t), To complete the description of the game, we must specify the exogenous reelection probabilities. One possibility is to assume that these probabilities are for Labour, where q' is independent of qG for the Conservatives, and q' = I the past history of the game. This is the assumption made in Alesina (i 987) . We call this the partisan advantage version of the model, as some party has an advantage over the other due to its general characteristics. However, independence of past history is a strong restriction, as incumbents have a number of advantages, so q mi-ght be higher for incumbents of either party.14 An alternative, therefore, is to assume that the qs are independent of party identity, but higher for the incumbent, so that I > q > 0o5 is the probability of re-election of the incumbent. We call this the incumbencyadvantageversion of the model. The two models are nested in a more general model in which election probabilities depend on both incumbency and the identity of parties. The incumbency model is somewhat easier to analyse, and so we concentrate on this model below. The analysis of the partisan advantage version of the model is in u= Aut_1 - qC Appendix i. It turns out that the empirical predictions of the incumbency advantage model are slightly more general than those of the partisan advantage version: we are thus able to test the additional restrictions of the latter on the data. We also present a direct test of the validity of the two processes for q against a more general process where q can depend on both the party identity and incumbent identity. The details of this test are in Appendix 2, and the results suggest that, given the election results in the United Kingdom since I952, the two special models have the same likelihood and cannot be rejected against the more general model where election probabilities depend on both partisan characteristics and incumbency.15 13 Lockwood and Philippopoulos (I99I) investigate a dynamic game between insiders and a single monetary authority, while Alogoskoufis and Philippopoulos (i 99I ) combine a static partisan model with an exchange rate regime model and investigate its applicability to Greece. 14 For example in the United Kingdom, the incumbent party has the option to choose the date of the next election within the constraint that the period between elections does not exceed five years. The fact that historically, the gap between elections has typically been significantly less than five years testifiesto the value of this option. 1 For the unrestrictedmodel, the estimated probability of a Conservativevictory when the Conservatives are the incumbent is equal to 67 % and the estimated probability of a Labour victory when Labour is the 48 ECS 102 I378 THE ECONOMIC JOURNAL [NOVEMBER 1.4. PoliticalEquilibrium We shall define political equilibrium to be a Nash equilibrium to the game described above, where the actions of the players depend only on the current value of the state variable. This is known as a Markov-perfect(Fudenberg and Tirole, i99i) or feedback Nash (Basar and Olsder, I982) equilibrium. The imposition of this restriction on the equilibrium rules out equilibria based on 'punishment' strategies as in Barro and Gordon (I983b). As the model is linear-quadratic, we may suppose that the actions of the players are linear functions of the state variables, i.e. = 7t g, t-1; e, n; 7Tt= 7TUt-1 I=C) ilj (9) where, e and n stand for election and non-election years, respectively. To characterise equilibrium inflation in this setting, note first that optimal behaviour for the wage-setters (i.e. rational expectations) is to set 1Tt= 7. From (g) this implies that, 7 = 7(i) = n (non-election year with incumbent i), qT?+ (i -q) 7Ti (election year with incumbent i). (ioa) (iob) Now we turn to the equilibrium behaviour of the political parties. Again because of the linear-quadratic structure of the model, we can write the discounted present value of losses from unemployment and inflation from time t onwards as a quadratic function of the unemployment rate inherited from t- , i.e. i = c, 1, j = e, n, where /?'> o is an, as yet, undetermined 1 u2_/2, coefficient. Note that /J?depends on the time to go to an election but not on calendar time. The f, and the associated inflation rates 7T, must satisfy the following recursion equations: U = min [(ir ut_)2+ (x1?&/3n) u2] (election year), (IIa) e ,n u_1 = min [(iT u_1) 2 + (c? + q6/e) u2] (non-election year), (II b) 'Tn where in both cases the minimisation is subject to (5') and (9), with the wagesetters expectations Tt taken as exogenously given. These require the present value of losses at time t (the left-hand sides of (i i a) and (i i b)) to be equal to the sum of the current loss and the expected value of future losses from t+ i onwards, when current inflation is chosen to minimise this sum (the right-hand sides of (iia) and (iib)).16 incumbent is equal to 50 %. For the incumbency advantage model the estimated probability of a victory for the incumbent is 6b0%. For the partisan advantage model, the estimated probability of a Conservative victory is 60 %, 16 One of our referees has objected strongly to our assumption that the parties do not care at all about inflation and unemployment when out of power. He finds the assumption unreasonable in the context of pure partisan models, pointing out that the Alesina model does not make this assumption. This is true, but whether one uses this assumption or not in the Alesina model does not matter, as this model contains no state I992] ELECTORAL UNCERTAINTY AND INFLATION I379 We can now solve (i i a) and (i i b) in several stages. First, calculating the optimal inflation rates from (ii a) and (ii b) conditional on fixed /7T, , and substituting in the equilibrium conditions (ioa) and (iob), we find that the 7rT are; lT =yA (I 2a) -d -e?(i) ie Ji= + d = (if incumbent i wins), i (I2b) (if non-incumbent i wins), (12C) where d= = q + y? (I - eq) j * i) y' = e[x + 8q1e(i)], = e((x?+ ) Note that the value of 7Te depends on who the incumbent party was; we have written 7T(j) for the inflation rate in election year when party i wins the election from partyj. For the same reason, the Pe also depend onj, so we have written fli(j) to express this dependence. Note, however, that the /?n do not depend on the identity of the previous incumbent; this is clear from (ii a), where the right-hand side only depends on parameters u,-1, q, 86,a , and fle(i) and 7Tare independent of the identity of the previous incumbent from (io a). We then say that the six pairs 7T4U),fe(), ' i, l-, i,j = c, 1, are a political equilibriumif they solve (i i a, b); (i 2 a-c). The equilibrium inflation rates can then be solved recursively. Substituting (I2 a, b) back into (ii a, b), and equating coefficients on u,-1, yields two pairs of simultaneous quadratics in coefficients (/fe (i), fnl), i = c, 1,known as Riccati equations in the dynamic game literature. In general, these Riccati equations can have unique, multiple or no solutions. Substituting solution values of fle(i), /?n,i = c, 1 (if they exist) back into (I 2 a-c) gives the equilibrium inflation rates in terms of the parameters.17 1.5. Inflation,ElectionsandPoliticalParties In this section, we investigate how inflation varies with the party in power and the time to the next election. In establishing these comparative statics results, we face the problem raised above, namely that there may be no, or multiple, political equilibria. Therefore, we focus attention on a particular equilibrium, the limit equilibrium,which can be shown to exist as long as 8.A2 is low enough. The limit equilibrium is the equilibrium whose inflation rates and valuation coefficients are limits of those of the finite horizon version of the game as the time-horizon becomes infinitely long. This is a rather attractive equilibrium to study, as the results obtained are robust to the introduction of a finite horizon. variables.In our case, however,the qualitativenature of our resultsis not affectedif one generalises somewhatby simplyassumingthatpartiescaresufficiently lessaboutinflationandunemployment whenout of powerthanwhenin power.Thus,ourmodelcan be seenas a combinationof a purepartisanmodelwith an 'opportunistic'model,whenpartiescannotaffecttheirre-electionprobabilities througheconomicpolicy. 17 It is then possibleto solvefor/e(1), flA(c) by substitutingbackinto (i I a), but we are not interestedin ,/Ie(l),/?e(c)as they do not affectthe Yland thereforedo not determinethe inflationrates. I380 THE PROPOSITION i. JOURNAL ECONOMIC [NOVEMBER A limit equilibriumalways exists if &A2is small enough.In the limit if acl> oci 7T1(i) equilibrium, > 7Tc(i), 7nT > 7nT, i.e. inflationis higherin non-election years if Labour is the party in power, and also in electionyears, conditionalon a given incumbent. Proof. See Appendix 3. Proposition I asserts that the predictions of the static RPT - namely, that the more unemployment-averse party will inflate more in every period than the less inflation averse party - carry over to the dynamic incumbency advantage model. We now focus on inflation in election years. In our empirical work, we are not concerned with inflation per se, but only in the wage-setter's expectations of inflation in election years, which is a weighted average of the two party's inflation rates and so depends only on the previous incumbent, and i.e. T(l) = q7T(l) + (i -q) 7nc(l) if Labour was the incumbent, then We incumbent. the were if Conservatives = 7T1(c) q) AT(c) q7T(c) + (i have: iT(l) > ii(c), i.e. expected PROPOSITION2. If cc'> occ and q > o inflationin election years is higher if Labour was the previous incumbent. Proof. From (I 2 a-I 2 c) and proof of Proposition I, the definition of di, AT(i) = Ad,/ (i -di). From the which implies I > d, > dC so the result y/ > yce follows. I Finally, we turn to the question of how inflation varies over the term of office for a particular party. We start with a 'benchmark' result: = o, 7T(i) > nT,71(i) < 7Tni = c, 1, PROPOSITION 3. With myopicpreferences,i.e. 8 i.e. whoeverthe previous incumbent,inflation is higher in the post-electionperiod for Conservativegovernments,but the reverseis truefor Labourgovernments. = cci, so this follows immediately from Proof. If 8 = o, I (I 2 a-c). The intuition for this is as follows: when the Conservative (Labour) party is in a post-election year, wage settlements have been based on a higher (lower) expected rate of inflation than in a non-election year. Thus, taking these expectations as fixed, to achieve a given reduction in unemployment, they must inflate more (less) than in a non-election year. Thus, even in the static RPT model, inflation does vary over the electoral cycle. (This point has been made by Alesina and Sachs (I988), p. 70.) This can be contrasted with the following result: PROPOSITION 4. If acc= acc,then7Te > Tn) i = C)1. Inflationis higherforbothparties in the periodfollowing an electionthan beforean election. Proof Let acc= cc = c. Then from (I 2 a-c), 7Tn = e(aC + q8l/e), n7e ( /3n) q/Jn which implies f/n < fle, as q < i. But then after substituting for inflation into the recursion equations, this implies Now suppose on the contrary that ne e2(X +q&l3e)2+ (cc+qq/3e) A2 =/fn < 7n. < fie = This requires f/3n 62(cX?+8y3 )2? (cc+8ykn) A2 I992] ELECTORAL UNCERTAINTY AND INFLATION But this inequality implies that q/3, </3n, a contradiction. I38I I The intuition for this result is as follows: As from (5') the incumbent party faces a perceived negative tradeoff between inflation and current employment, and there is a positive link between present and future employment, the party in power will have a greater incentive to raise inflation, the more it cares about the future. The closer it gets to an election, the less it cares about the future, as the probability of losing office comes closer. This is clear from the recursion equations above. A unit reduction in unemployment has a future return of &/?fn just after an election (when the next election is two years away), and only q&/3le when the election is one year closer. Propositions 3 and 4 suggest that when 8 is high and ac'is close to aC, both parties will inflate more immediately after elections than before, whereas when 8 is low and ac' is much larger than aC, the Labour party will behave in the opposite way when in power. Thus, when the preferences of the parties over inflation and unemployment are not too different, the effect identified in Proposition 4 is large enough to more than offset the force, outlined in Proposition 3 above, which lowers the post-election inflation rate for Labour governments in an environment without forward-looking behaviour. Analogues of Propositions I-4 for the partisan advantage version of the model are established in Appendix i. Proposition 3 continues to hold, and so do Propositions i and 4 if q' and qc are sufficiently close. However, Proposition 2 becomes rather different if incumbency has no effect on the election outcome. 7T(i) becomes independentof i.18 1.6. Inflationand Political Parties underFixed Exchange Rates The next question that needs to be resolved is the determination of expected inflation during fixed exchange rate regimes. In our sample, the only such regime in which the United Kingdom participated was the Bretton Woods system of fixed exchange rates. In this regime, UK administrations did not have the independence of determining the average domestic inflation rate. Up until I968, the average OECD inflation rate was determined by the monetary policy of the Federal Reserve System. Under the rules of Bretton Woods all countries with the exception of the United States were pegging their exchange rates to the US dollar, while the Fed undertook to maintain a fixed price of gold at $35 an ounce. This precluded accommodation of price shocks by the Fed, and resulted in lack of inflation persistence (see Alogoskoufis and Smith, I99I, and Alogoskoufis, I 992). Thus, given the characteristics of this system we shall assume that expected inflation in the United Kingdom during the Bretton Woods system of fixed exchange rates was constant and, given the obligations 18 Our dynamic 'partisan model' implies a clear time pattern for inflation (see Proposition4): policymakersgo for higher inflationimmediatelyafter electionsthan beforeelections.'Opportunistic models'withsluggishpriceadjustmenthavea similarimplication,aspre-electoral expansionary policieslead firstto lowerunemployment(beforethe election)and laterto an increasein inflation(afterelections).In principle,one shouldbe able to distinguishbetweenthe two models,by usingunemploymentas well as inflationequations. I382 ECONOMIC THE [NOVEMBER JOURNAL of the Fed, lower than expected inflation under managed floating. Thus we assume that,19 3) Et-, Aptf = '7Ton('I where, 7ro is a constant. Participation in a fixed exchange rates regime in order to gain the antiinflationary credibility of a foreign monetary authority that determines monetary policy for the whole system is the central hypothesis of Giavazzi and Giovannini (I987) and Giavazzi and Pagano (I988). Their models are applications to open economies of the Rogoff (I985) idea, that appointing an independent, inflation-averse central banker will affect the inflationary expectations of wage-setters, and will thus result in an equilibrium with low inflation. In our model, the assumption is that this also does away with the dependence of equilibrium inflation of past-unemployment, as it is highly unlikely that the Fed took British unemployment into account in determining its monetary policy. We can now move on to empirical testing. II. ECONOMETRIC ESTIMATES AND TESTS: THE UNITED KINGDOM I952-90 In this section we present the econometric specification of the model and discuss estimation and a number of testable hypotheses. Subsequently we discuss the data and present econometric estimates and tests without and with allowance for a separate Thatcher regime. II. I. EconometricSpecification and (ioa, b) and (I3) into (4) gives us estimable wage Substituting (I2a-c) equations for the incumbency advantage model. They can be written in general form as =g+rTO+ /t g + 7TO+ OA\pt-1 /\Wm AW+Of I Ut-,+PnUt-, OApt-i 6ut-1 +) = g U++7TO- Pe lt-1 ct-1) (i = i c,l), (I4a) 4 b) c l ( (I4C) where i is an index of the government in power, j an index of the incumbent in an election, and the ps satisfy (see (i o a, b) and (I 2 a-c)), = 4; Pn4~ (i = c, 1),pe = qi4(j) + (i -q) 7Te(j); i,j = c, lij. ( ) Equation (I4a) refers to wage inflation under a managed exchange rate 19 Equation (i8) iS probably too restrictive, as under Bretton Woods governments could have inflation rates that were temporarily different from American rates, but not persistently different. Allowing for temporary differences in inflation rates would not affect the nature of our results. I992] ELECTORAL AND UNCERTAINTY INFLATION I383 regime, in a non-election year, when party i is in power. Equation (I 4 b) refers to wage inflation under a managed exchange rate regime in the aftermath of an uncertain election, where the incumbent was party j. Equation (I 4 c) refers to wage inflation under the fixed exchange rate regime of Bretton Woods.20 Estimation of the model is straightforward. Since all the right-hand-side variables are lagged, OLS is consistent. To allow for the differences under alternative regimes one can estimate a single wage equation with multiplicative dummy variables to capture the regime shifts. Given Propositions I and 2 in section I, we have the following predictions on the ps for the incumbency advantage model: 1C 1Pn P (i6a, P b) For the partisan advantage model, we have the additional restriction that, (I7) Pec= pe. Equation (I 7) reflects the property that if the probability of election is party specific, expected inflation will not depend on the identity of the incumbent, as it will be a simple weighted average of expected inflation under either party (Proposition A2). So, the main difference between the partisan advantage and incumbency advantage models is that for the latter, the coefficient on ut_1 in (i 6 b) should be higher (lower) if the Labour (Conservative) party was the incumbent party at the election, whereas for the former, the two should be the same. It is also worth considering the additional predictions of the model in the special case where there are no differences between the two parties in their relative aversion to unemployment (similar preferences). In that case, if we assume the incumbency advantage model, from Proposition 4, the relevant restrictions will be, Pn =Pn = 7Tnh Pe = Pe = 7Te Te > 7Tn. (I8a-c) Equations (i8a, b) suggest that there are no party-specific differences in expected inflation in either non-election or election years. Equation (i 8 c) reflects our Proposition 4, that, because of the longer horizon of an election winner, the present value of reductions in current unemployment through unanticipated inflation is higher, and expected inflation rises to reflect that. These predictions are also generated by the partisan advantage model in the case where qc = ql = o05. 20 While the theoretical model examines two periods (election and non-election), in fact an election period is followed by up to four non-election periods. A correct specification would therefore affect the values of the ps (depending on how many periods are left before the next election). However, this would not change our qualitative results since electoral uncertainty exists only in the election period. I384 11.2. THE ECONOMIC JOURNAL [NOVEMBER The Data Before we move on to estimation and testing, a brief presentation and discussion of the data is in place. Fig. 2 depicts wage and consumer price 030 - 002s 20 - ............................ 015- 0:10 0.05.- ''.................... '''''''''''' . Wag in,aiio n -4........... / o.oo - 2. , -~~~~~~~ / ~~~~Price rate Unemployment inflation 1965 1960 1970 1975 1980 1985 Wage and price inflation and the unemployment rate. 1955 Fig. 1;C& 1990 inflation and the unemployment rate in the United Kingdom from I952 tO I 99O. The wage series (hourly earnings in manufacturing, I 985 = i), the price series (the Retail Price Index, I985 = i) and the unemployment rate are from various issues of Economic Trends. The basic dummy variables used are reported in Table i. In constructing the dummy variables for exchange rate regimes we assumed that sterling was expected by wage setters to remain fixed to the dollar until the devaluation of sterling in November I967. Thus, in our empirical work we assume that the Bretton Woods system credibly applied only between I952 and i967. In any case the anti-inflationary credibility of the system was undermined in the late i96os not only by the devaluation of sterling, but also by the abandonment of the gold pool and the establishment of two-tier gold arrangements on I6-I 7th March, I968 (see Solomon, I982) 21 In constructing the party political and post-election dummies we divided particular years according to the proportion of total quarters in a year that each party spent in power, and the proportion that belonged to immediate post-election year (see note to Table i). We only used information relating to the quarter of the year in which an election was held, as the exact timing of an election in the United Kingdom is usually not known sufficiently in advance to justify finer distinctions.22 21 In any case, our empirical results are not particularly sensitive to the exact choice of year for the abandonment of Bretton Woods. 22 We have toyed with the idea of using quarterly rather than annual data. We have eventually rejected it, against the usual argument that we would have more degrees of freedom, on the following grounds: First, the increase in degrees of freedom would have been apparent rather than real, since the episodes that matter for our purposes are elections, and using quarterly data does not give more of these. Second, the United Kingdom is characterised by an annual wage round, so a year is the natural time period. Third, quarterly observations generate additional statistical problems, related to the question of seasonality and spurious ELECTORAL I992] UNCERTAINTY AND I385 INFLATION Table I Dummy Variables Used in the Regressions I 952 I 953 1954 I955 I956 1957 1958 a5 d I I I I 0 0 0 0 0 0 0-875 0 0 0 0 0 o0I25 0 I I 0 0 0 0 0 o625 0 375 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0375 o-625 I I 0 0-I25 0 0 0-875 0 o875 0 0 0 0 0 01I25 0 I 0 0 I dedc del dnc dnt d0il 1959 I I I I960 I I96I I 0 0 I 962 I 0 0 0 0 I I 963 I 964 I 965 I I I 0 0 0 0 0 I 0 0-I25 0 0 0-875 0 I966 I I967 I 968 I 969 I 0 0 0 0 0 0 o-875 0 0-I25 0 0 o-625 0?375 0 0 0 0 0 I I o-625 0?375 0 0 0 0 0 I 0 0 0?375 0 0 0 I I I 0 0 0 0-375 0 0 0 0 0 o-875 0 0 0 0 o-625 I I o0I25 0 0o875 01I25 0 I 0 0 0 0 0 0 0 I I I o0625 0 375 0 0 0 0 0 0 0 0 0 0 o-625 I I 0 I I I970 I97I I972 0 0 1973 0 I974 0 0 I 0 0 I I I I975 1976 I 977 I 978 1979 I 980 0 0 0 I I o-625 0 0?375 0 0 0 I I98I 0 0 I 0 0 I 982 0 I 0 0 0 I 0 I983 I984 0 I 0 0 I 0 0 0 0 0 0 0o375 o-625 I I 0 0 0 0 0 o-625 0 375 0 0 0 0 0 0 I 985 I986 I987 I988 I 989 I 990 I I I 0 0 I I o-625 0 375 0 0 I 0 0 0 0 0 0o375 o-625 I 0 I 0 0 0 0 0 0 0 Notes: adf denotes the Bretton Woods system of fixed exchange rates before the devaluation of sterling in I 967, and d' Bretton Woods after the devaluation of sterling and the post-Bretton Woods managed floating regime. deCand de' denote the proportion of the year that fell within the four quarters immediately after an in which the incumbent election was the conservatives and labour dn' and dfc denote respectively. of the year under a Labour and a Conservative the proportion respectively the administration, excluding four post-election quarters. Finally d05'denotes a zero-one dummy variable for the oil shocks. II.3. Estimates and Testsfrom Wage Equations, 1952-90 In Table 2 we present evidence relating to the wage equations in (I4a-c). We start with a fairly unrestricted wage equation, which is reported in column I. In this we allow for separate dummy variables for each political party, for each exchange rate regime, and for election and non-election years. Thus we have dynamics. On balance, we therefore decided that the costs of using quarterly data would outweigh the benefits. An additional investigation with quarterly data may be useful at a later stage, but it does not appear to us that this is how one should start. 1386 THE ECONOMIC [NOVEMBER JOURNAL Table 2 WageInflation,Electionsand theExchangeRate RegimeEstimatesand Testsfor the UnitedKingdom (Dependent variable: Awt, sample: Constant APe-1 method of estimation: OLS.) (I) (II) (III) (IV) oo96 0-094 0-93 I 0I (O-OII) (O-OI2) (00-I 3) Ap^t1 I952-90, 02 (0-0I 2) -0-28I I (0o3I9) ?0403 (0'I07) ut-l -2.004 utf_l 1 (0-973) -4 57I 0-4I I (00I 4) 0-322 (0-089) -2-2I5 -2-277 (o-835) 0-3I9 (0-095) -2.784 (o.853) (o0876) (2.938) u-tf1 o-8og (I'I 19) a-tfe1 4-426 I38) I.56i (3 utm-1c mnl utm-11 um-elc umMel t-1 do" (0-924) I o036 (o-983) I-8I4 (0o938) 3'430 (I-OIO) 0-052 (0-020) I*847 (0-767) I1323 (o-833) I__8 (0-785) i-_822 I838 (0-785) 2-2q4 (0-814) 2-24 (0-8I4) 2-099 2-o96 2-723 (0o783) 3'743 (o-802) 3-668 3'2 (o-8io) 2723 (o-837) (o-85s) (0-8 I0) 0-05 I (0-020) o0o63 o-o84 (00 I9) (o-oi8) R2 o-86 i o-843 o-83 I o-8o i 17 0-020 0-020 0-020 0-022 I-883 i-8i i I-678 I-796 0-203 0-367 I*o83 0-496 i-i i6 0333 0455 DW AR (i) JB (2) ARCH (i) F-test 1-950 I-283 O1I84 2-495 o-8o5 (4, 27) 2-54I (, 3I) I-IOI 5-576 (I, 32) Notes: Ayvand Apm denote the product of Ap and dummies df and dmrespectively. uijdenotes the product of the unemployment rate and dummies with superscript i and j. Asymptotic standard errors are in parentheses below estimated parameters. R2 is the coefficient of determination, o- the standard error of the regression and DW the Durbin-Watson statistic. AR (i) is a Lagrange multiplier test for first-orderresidual autocorrelation,JB (2) is the Jarque-Bera test of non-normality and ARCH (i) is a Lagrange multiplier test for autoregressive conditional heteroskedasticity, based on an auxiliary regression of squared residuals on their first lag. These are x2 tests. In each column the underlined or bold coefficients have been imposed to be equal. The F-test (degrees of freedom are reported below them) tests the validity of the restrictions that the estimates in each column imply for the estimates of the preceding column. For variable definitions see Table I and the text. multiplied the party political and election dummies by the exchange rate regime dummies. We also allow for a non-zero inflation persistence during Bretton Woods. We have included an additional dummy variable for the two oil-shocks. This equation has an impressive fit (recall that the dependent variable is a rate or change), and there is no evidence of statistical misspecification in the reported diagnostics. Thus, we can treat it as a reasonable encompassing statistical model, against which we can test the predictions of our model. From the estimates in column I one would not at conventional significance I992] ELECTORAL UNCERTAINTY AND INFLATION I387 levels be able to reject the hypothesis that during Bretton Woods lagged inflation does not help predict price and, therefore, wage inflation. The asymptotic t-ratio is o9go. A second hypothesis, that expected inflation during Bretton Woods did not depend on the identity of the party in power, or elections does not seem to be rejected either, as neither of the coefficients of Ufnl, Ufec, Ufet seem to be statistically different from zero. These restrictions are imposed and tested in column II. The relevant F-test is much lower than its 5 0 critical value. Thus, one would not be able to reject the hypothesis that expected price inflation, and therefore wage inflation, during Bretton Woods was the same independently of lagged inflation, the party in power, or elections. However, it is also worth noting that Labour was in power only for 3 out of the i6 relevant years. The results in columns (I) and (II) of Table 2 fail to provide evidence against the exchange-rate-regime theories of inflation, as the additional lagged unemployment coefficients under managed floating are strongly positive and significant, as predicted by our theoretical model. Managed floating seems to eliminate almost completely the negative dependence of wage inflation on lagged unemployment prevalent in the Bretton Woods system of fixed exchange rates. According to our model this is because of the anticipation by wage-setters of the incentives of political parties to inflate. As we showed in section I these incentives are a function of lagged unemployment. We next move on to testing the predictions of the RPT. The estimates of column (II) are based on the incumbency advantage electoral model. The partisan advantage model implies the further restriction that the identity of the incumbent should not matter for expected inflation after an election (equation (I 7)). The Wald-test for this restriction in the estimates of column II is equal It cannot be 3284. = to 6's. This restriction is rejected at 500 as XO.95(') rejected at i % though (%0.99(I) = 6 63). The two restrictions implied by the assumption that the parties have the same preferences for inflation and unemployment (equations (i8a) and (i8b)) yield a Wald statistic of 8-47. These restrictions are rejected at 500 but not at i % (X%95(2) = 5-99, X%.99(2)= 92I). However, it appears from the estimates of column II that, during managed floating, the differences between political parties have been very small in nonelection years. In column (III) we impose the restriction that the two political parties are seen by wage setters to have the same incentives to create surprise inflation in non-election years. The F-test for this restriction indicates a very high likelihood that this restriction holds. However, according to our model, if the two parties have the same preferences in non-election years, expectations of inflation will be the same in election years as well. In column (IV), we present estimates under this additional restriction. The rejection of the joint restrictions implied by similar preferences seems to be due to election years only. The F-test for the restriction of equality of expected inflation for the two political parties in election years is equal to 5y8, against a 95 0 critical value of F(I, 32) equal to 4-2. Thus, the restriction of equality of coefficients in election years is rejected at 5 0. However, this restriction is not rejected at I %, since F0.99(I, 32) = 7-5. I388 THE ECONOMIC JOURNAL [NOVEMBER When we impose the equality of post-election coefficients, as in column (IV), our proposition 4 is borne out. Expected, and actual, inflation is higher in postelection years, as the positive reaction of inflation to lagged unemployment is higher. The X2(i) Wald-test for the statistical significance of the difference between the two coefficients is equal to 2-9. Thus, the difference between nonelection and election years is statistically significant only at I o %, but not at the customary 5 / . Overall, the results in Table 2 suggest that the exchange-rate-regime model cannot be rejected, but the evidence on the dynamic rational-partisan model is weaker. There are no party-related differences in inflationary expectations in non-election years, inflationary expectations are higher in post-election years but not by much, and there is a stronger post-election inflationary effect when Labour is the incumbent. In the light of these results we move on to investigate possible separate effects for the Thatcher administrations, which are widely regarded to have been a change in political regime in themselves (see Buiter and Miller (i 983)) . 11.4. The ThatcherRegime and UK Inflation Did Mrs Thatcher make a difference for expected inflation in the United Kingdom, and what was the nature of this difference? This has been a question that has intrigued macroecoyiomists and political commentators in Britain and abroad from early on. The estimates in Table 3 attempt to examine this question in the context of our model. Column (I) presents the unrestricted estimates of column II of Table 2 in a different format. The superscripts mn and me refer to non-election and election years during managed floating. The estimated coefficients of lagged unemployment for these periods measure the response to unemployment that was common to all Conservative and Labour governments. Then, the coefficients of variables with the additional superscript I measure the additional effect from the Labour governments. One can directly see in Table 3 that there is no statistically significant additional Labour effect in non-election years, but that there is a large positive one in election years which is significant at 5 00. Column II presents estimates with additional Thatcher dummies (Ig80-go). The format is as in column I, the dummies with the additional superscript t denoting differences of the Thatcher governments from the rest. One notes immediately that the differences between pre-Thatcher conservatives and Labour in election years become statistically insignificant. Also, there do not seem to be significant differences between Mrs Thatcher and the rest in non-election years. These restrictions are imposed in column III and cannot be rejected by the F-statistic at the conventional 5 0. In column IV we also impose the restriction that the dependence of expected inflation on lagged inflation was no different under Mrs Thatcher. This restriction cannot be rejected either. The estimates in column (III) suggest that there are no differences between Labour, pre-Thatcher Conservatives and Thatcherite Conservatives in the weighting of unemployment relative to inflation. All governments have been perceived to care the same about unemployment, and, given past un- ELECTORAL I992] UNCERTAINTY I 389 INFLATION AND Table 3 Wage Inflationand the ThatcherRegime (Dependent variable: Awt, sample: APt-1 o-083 0o094 (O-OI 0'41 I (O' I 04) (o I I g) - umn -I -2277 (o0835) I-847 ?-523 (0-33 -252 ) 526 (o 856) 1-288 (0-715) -o-642 0'I04 ume1 mel (0583) 420 4 202 2 099 umel I *644 (o0645) u1 -t do"l 005 I (0-020) R2 o0843 CT 0-020 DW I-8I I 0-367 AR (i) JB (2) ARCH (i) F-test I1283 0077 o0o83 o.o85 (O-OI I) (O-OI I) 0-327 (oo8o) (0-079) 0'312 (0-103) - I531 (0'791) I-363 (0-712) - I679 (0o778) I'425 (0-709) (0?459) ut-1 (0 784) (IV) _01I03 (0138) (0o767) u mnll 2) 0-47I - (III) (O-O I I) Apt-1 Ut_1 method of estimation: OLS.) (II) (I) Constant 1952-90, (*1.53) -o 983 (II73) -2-8I7 36 3-6I I376 (0o737) --21I97 3-763 (0-72I) -2242 (i *i 85) o0o62 (o0o I 8) (o-o65) o-o65 (0-0 I 7) 0-059 (o o I 6) o0894 o-871 o-867 O I7 o0oI8 ooI8 I-873 01I 20 0-229 2-264 I1073 0?495 2-093 (0-562) I-833 2-125 I1764 2605 2042 2-036 (3, 28) I*oI8 (I, 3I) Notes: See Table 2. Superscripts n and e refer to all non-election and post-election years, and superscript t refers to the Thatcher years. employment, any of these governments would, in non-election years, generate the same expected and actual inflation. The estimates also confirm the electoral predictions of our model for Labour and pre-Thatcherite Conservatives. In election years inflationary expectations are higher than in non-election ones, and, as a result, wage inflation is higher by almost two and a half times the rate of lagged unemployment. This difference is statistically significant at 5 00. The exception is Mrs Thatcher. On the basis of our estimates, the Thatcher governments were not expected to behave any differently in non-election and election years. There are two possible explanations for this difference between Mrs Thatcher and the rest. One is that the probabilities of re-election of the incumbent changed after I 980, in view of the crisis in the Labour Party. If the probability of re-election of the incumbent was estimated at i, there would be no difference between election and other years for the Thatcher governments. This explanation is one possibility, although maybe not the most plausible one. A second explanation is that the behaviour of the Thatcher government once I390 THE ECONOMIC JOURNAL [NOVEMBER it was elected in I979 managed to establish a credible anti-inflationary reputation. Even if money growth rules, the Medium Term Financial Strategy and the like were above their target ranges, the recession in the immediate aftermath of the election may have convinced wage-setters that the electoral cycle was gone for good, and this may be why no resurgence of inflation was expected in the aftermath of the I983 and I987 elections.23 In conclusion, under managed exchange rates, and before Mrs Thatcher, there is strong support for the 'pure' version of partisan models for the United Kingdom. The preferences of different parties between inflation and unemployment do not seem to differ. Given lagged unemployment, UK political parties were expected to generate the same rate of inflation in nonelection years. In election years they were expected to generate significantly higher inflation, with the exception of Mrs Thatcher whose behaviour in her first two years in office may have dispelled the expectation that future elections would be followed by monetary relaxation. III. CONCLUSIONS This paper extends the basic partisan model of inflation and unemployment to allow for unemployment dynamics. The extension turns out to be important because if unemployment persists, any reduction in unemployment through unanticipated inflation will have future as well as current benefits for the party in power. Thus, the distance from the next election will affect the incentives of governments to create unanticipated inflation to reduce unemployment. We demonstrate two main results: First, that if socialists care more about unemployment than conservatives, they would be associated with higher inflation at any point in their term of office. This generalises the basic Alesina (I987) result to the dynamic case. Our second main result is that if the preferences of political parties are close enough, inflation would be higher under both parties in the period following an election than in non-election periods. This prediction of our dynamic model is in sharp contrast to the case without unemployment persistence examined by Alesina (I987) and Alesina and Sachs (I988). We test the predictions of the dynamic model for the British economy, combining the political model with a model of exchange rate regimes, which predicts that inflation should not systematically differ with the party in power or elections under credible fixed exchange rate regimes (Giavazzi and Giovannini, I987). This is because fixed exchange rates take away the ability of governments to create surprise inflation. Our empirical findings suggest that the exchange rate regime matters greatly for expected price inflation and thus nominal wage growth. One cannot reject the hypothesis that wage inflation in the Bretton Woods system of fixed exchange rates has been independent of the party in power and the timing of elections. 23 The possible effects of the I 980 deflation on the anti-inflationary credibility of Mrs Thatcher have been investigated by, among others, Buiter and Miller (I98I, I983), Backus and Driffill (I985), Begg (I987), Goodhart(I989) and Beanand Symons(I990). I992] ELECTORAL UNCERTAINTY AND INFLATION I39I In the period of managed exchange rates that has followed the collapse of Bretton Woods, there seem to be very small differences between Conservative and Labour administrations, especially in non-election years, and when one allows for a change in regime under Mrs Thatcher. Our prediction that postelection years are associated with higher wage inflation seems to be borne out in general. However, during the Thatcher years there has been no difference in wage inflation between election and non-election years, an indication that some form of anti-inflationary credibility was established by the governments of Mrs Thatcher. We close with some qualifying remarks. First, our empirical tests have concentrated only on wage equations, while the theoretical model has predictions for price and unemployment equations as well. The wage equations are particularly important in the context of this theory, as it is through them that expectations of inflation affect unemployment. One possible extension is to further test the full model using the cross equation restrictions between wage, price and unemployment equations. This may allow us to distinguish between our model and opportunistic models with sluggish price adjustment, and is left for future work. Second, it should be borne in mind, that as with almost any empirical results, our findings could be interpreted differently. As a referee has pointed out, it would be possible to interpret some of our dummy variables as reflecting shifts in the natural rate of unemployment, rather than shifts in expectations about inflation. Possible other omitted variables could also have biased our results. However, we feel that the pattern of our findings is consistent with the theory we have put forward, and we claim no more than that. Given the limited number of economic and political episodes in our sample, we have found that attempts to allow for too many different shifts, result in nearcollinearity among the regressors, and fail to provide additional information. BirkbeckCollege, Athens School of Economicsand CEPR Birkbeck College, University of Exeter and CEPR University of Essex Date of receiptoffinal typescript: June I992 APPENDIX I Analysis of the Partisan AdvantageModel Here, we establish analogues of Propositions I-4 in the text, but for the partisan advantage version of the model. The equations defining the equilibrium are as in (i o), (i i), and (I 2), except(i) ( Io b) is replaced withhT, = ql7?q + qci', (ii) Ze replacesorin (I2 b), (I 2 C); (iii) q is replaced by qtin ( II b). Because the incumbent party no longer has any effect on the outcome of the election, both 87T and /?e are now independent of the identity of the incumbent, and so the political equilibrium is now characterisedby four, rather than six pairs, 7i4,Pe, 74nfl , that solve (II) and (I2) with the appropriate modifications as described in (i)-(iii) above. We then have the following results: thenfor q',qc closeenoughto ? 5, 7Te> 7Te', Fn > 7Te o 5 in the proof of Proposition I: this establishes the result for PROPOSITION A I. If a' > a' Proof. Set q = I 392 THE ECONOMIC JOURNAL [NOVEMBER qC = q = 05. The result then follows by continuity parameters. I PROPOSITIONA 2. of the f and 7T in the Expectedinflationin electionyears is independent of the identityof the incumbent. PROPOSITIONA 3. With myopicpreferences, i.e. 8 = O, 7T' Proof. If 8 = o, 7 = oci, so this follows immediately modified following (i)-(iii) above. I > 7Tn, 7Te < 7T. from (I2a-c), PROPOSITION A 4. If Occ= ocx= ac, and qCis close to ql, then 7T' > iT4, i = appropriately c, 1. Proof. Set q = 0o5 in the proof of Proposition 4; this proves the result for qc_ ql = fl and 7T< in the parameters. I o05. The result then follows by continuity of the APPENDIX 2 Incumbencyor Partisan Advantage?A Direct Test The two special models of electoral uncertainty that we utilise, namely the incumbency advantage and electoral advantage model, are both special cases of a model in which both incumbency and the identity of parties matters for their probability of winning an election. Assume that p is the probability of a Labour victory if Labour is the incumbent, and q the probability of a Conservative victory if the Conservatives are the incumbent. Obviously the probability of a Labour victory if the Conservatives are the incumbent iS I - q, and the probability of a Conservative victory if Labour is the incumbent is I-p. Against this more general model, the partisan advantage model implies the restriction that q = I-p, and the incumbency advantage model implies the restriction that q = p. To test the restrictions implied by each of the models, we have used the data on all UK elections since I952 to estimate the election probabilities with and without the restrictions implied by the incumbency advantage and partisan advantage models. Given the UK election results, the unrestricted likelihood function takes the form, L = kq4(i_ y)2( q I _ p)2p2, (A 2. I) where k is an arbitrary constant, reflecting the unconditional probability of the Conservatives being the incumbent in the first post-I952 election of I955. The maximum likelihood estimates can be derived by taking the first-order conditions of (A 2. I) with respect to p and q. The maximum likelihood estimates of the unrestricted model are p = 2 and q = 2 The maximum likelihood under the restrictions of the partisan advantage model (q = -p) yield q = 5. The likelihood ratio test for the restriction is equal to 028, well below the 5 0 value of x2(I) = 3-84. Thus, the restriction of the partisan advantage model cannot be rejected. On the other hand, the maximum likelihood estimates of the incumbency advantage model (q = p) also yield an estimate of q = 5. The incumbency advantage model has the same likelihood as the partisan advantage model. Of course, its restrictions cannot be rejected either. The results of the io elections held since I952 are not informative enough to allow us to discriminate between the two restricted models, and none can be rejected against the unrestricted one. Conditional on the incumbency model, the incumbent has a 6o0% probability of being re-elected, and conditional on the partisan model, the Conservatives have a 6o0 0 probability of being elected in any given election. I992] ELECTORAL UNCERTAINTY APPENDIX INFLATION AND I393 3 Proof of Proposition I We give here an outline of the proof; the full version is available from the authors on request. (i) Consider the following finite-horizon (T-period) version of the game between political parties and wage-setters. Label time backwards, and suppose without loss of generality that the last period, period i, is a non-election year, and that T is even, so that period T, the first period, is an election year. Then, fin and ,fl,(i)will be timedependent; redefine fl,, = l, /e t(i) = yt to avoid an excessive number of subscripts. Then, it is easy to check that the recursion equations (I I a), (i I b) become; 7T t) )2] i [(7Ti)2+ t (Ot+ 8_fi)t . (A_(7Ti_ t i.=-min = J' ue t even A 3.1) .I ~~~~(A iT1t = min[()2+ .) (at'+ q.8& . (A - (n-_ t) )2] i = l,c) t odd (A 3.2) lit where in both cases the wage-setters, expectations 7Ttare taken as exogenously given, and are defined by the analogues of (ioa) and (iob) i.e. lit = T, t odd, i incumbent at t+ I , and lit = q. 7Trt+ (I -q) .7et, t even, i incumbent at t+ I. Also, there is the initial condition that y1 = o, as in the last period, period i, future losses are zero. (ii) Given (A. 3.I) and (A. 3.2) and this initial condition, it is possible to show that; (a) At' 2 fly,t even, y' > at, t odd, for all t = i,. . T; (b) {fi}ti2t4.2.T and {Yi}t=1,3..T-1 c, are monotonically increasing bounded sequences, and so have limits as T-? co. Proof of property (a) and monotonicity is by induction and is straightforward; the details are omitted. To show boundedness, note that both P, t even, and yt, t odd, are bounded above by /3t, where flt.U2 iS the present value of loss over t periods, starting with unemployment ut, to a government which is perpetually in power, and has a weight on unemployment of xl. (The reason for this is that in the T-period game, for any given sequence of inflation rates, the higher ac, and the larger the probability of staying in power, the greater the present value of losses to the government; this can easily be proved formally by induction.) It is possible to show that lim ,Bt = /?exists i=1, t-oo and is finite as long as a.A2 ? [(I+xt)-2.(&)/112]/(I_al) = B (see Lockwood and Phillippopoulos (iggi) for the details). Also, flt is monotonically increasing. So, /'t, gty< Alt < fl, all t. and lim ?7' = i, i = 1,c exist and are (iii) From the resultsin part (ii), lim =T T ->,o T ->oo finite as long as d. A < B, as monotonic bounded sequences always converge. Furthermore, as the right-hand sides of (A 3. I), (A 3.2) are continuous in /At,y', i=1, C, taking limits as T-*oo in (A 3.1), (A 3.2) yields the recursion equations (II a) and ( I b) in the text. So, (fl, i), i = 1,c jointly solve these recursion equations. Therefore, 1,c we have shown that there exists making the identifications =lt= fn, Ie(,= a unique political equilibrium in the infinite horizon game which is the limit of finitehorizon equilibria, as long as <.A2? B. Furthermore, as flP_= B, t even, yt, t odd, for all t = i,. . T, these inequalities are preserved in the limit as T-* o, so ?l> I3'(1)> /Be(c). So, as a' > ca',it follows that in the limit equilibrium, y7 > yn, 0/> ye. Then, I > d1> dc, so from 7TC,as required. (I2 a)-(I2C) in the text, it follows that 7T4(i) > 7T'(i), 2T > REFERENCES Alesina, A. (I987). 'Macroeconomic policy in a two-party game with rational voters.' QuarterlyJournal of Economics,vol. I02, pp. 651-78. Annual (ed. S. Fischer). Alesina, A. (i988). 'Macroeconomics and politics.' In NBER Macroeconomics Cambridge Mass: MIT Press. THE 1394 ECONOMIC JOURNAL [NOVEMBER 1992] Alesina, A. and Roubini, N. (I990). 'Political cycles in the OECD economies', CEPR Discussion Paper no. 470. Alesina, A. and Sachs, J. (I988). 'Political parties and the business cycle in the United States, I948-I984.' Journalof MoneyCreditandBanking,vol. 20, pp. 63-82. Review,vol. 32, PP. 542-50. Alesina, A. and Tabellini, G. (i 988). 'Credibility and politics.' EutopeanEconomic Alogoskoufis, G. S. (I992). 'Monetary accommodation, exchange rate regimes and inflation persistence.' ECONOMIC JOURNAL, vol. I02, May, pp. 46I-80. Alogoskoufis, G. S. and Manning, A. (I988 a). 'Wage setting and unemployment persistence in Europe, Japan and the USA.' EuropeanEconomicReview,vol. 32, pp. 698-706. Alogoskoufis, G. S. and Manning, A. (I988b). 'On the persistence of unemployment.' EconomicPolicy, vol. 7, PP. 427-69. Alogoskoufis, G. S. and Philippopoulos, A. (i99i). 'Political parties, elections and exchange rate regimes: on the deeper determinants of inflation in Greece.' Discussion Paper in Economics no. 5/9I, Birkbeck College. Alogoskoufis, G. S. and Smith, R. (i99i). 'The Phillips curve, the persistence of inflation and the Lucas critique: evidence from exchange rate regimes.' AmericanEconomicReview,vol. 8I, pp. I254-75. Backus, D. and Driffill, J. (i 985). 'Inflation and reputation.' AmericanEconomicReview,vol. 75, PP. 530-8. Barro, R. and Gordon, D. (i983a). 'A positive theory of monetary policy in a natural rate model.' Journal of PolhticalEconomy,vol. 9I, PP. 589-6IO. Barro, R. and Gordon, D. (i 983b). 'Rules, discretion and reputation in a model of monetary policy.' Journal of MonetaryEconomics,vol. I2, Pp. I0I-2I. GameTheory.New York: Academic Press. Basar, T. and Olsder, G. J. (I982). DynamicNoncooperative Bean, C., Layard, P. R. G. and Nickell, S. J. (I986). 'The rise in unemployment: a multi-country study.' Econom:ca,Supplement, vol. 53, PP. SI-22. Annual(eds. 0. Blanchard Bean, C. and Symons, J. (i 990). 'Ten years of Mrs T.' In NBER Macroeconomics and S. Fischer). Cambridge Mass: MIT Press. of theBritishEconomy(eds. R. Dornbusch and R. Layard). Begg, D. (i 987). 'Fiscal policy.' In ThePerformance Oxford: Oxford University Press. Cambridge Mass: MIT Press. Blanchard, 0. and Fischer, S. (I989). Lectureson Macroeconomics. Blanchard, 0. and Summers, L. (I986). 'Hysteresis and the European unemployment problem.' In NBER Annual(ed. S. Fischer). Cambridge Mass: MIT Press. Macroeconomics Buiter, W. and Miller, M. (I98I). 'The Thatcher experiment: the first two years.' BrookingsPaperson EconomicActivity,vol. 2, PP. 315-8o. Buiter, W. and Miller, M. (i 983). 'The macroeconomic consequences of a change in regime: the UK under Mrs Thatcher.' BrookingsPaperson EconomicActivity,vol. 2, PP. 305-79. Eichengreen, B. (i 985). 'Editor's introduction', in his (ed.). TheGoldStandardin TheoryandHistory.London: Methuen. Eichengreen, B. (I989). 'Hegemonic stability theories of the international monetary system.' In CanNations Agree?(ed. R. C. Bryant et al.). Washington D.C.: The Brookings Institution. Serieson Fischer, S. (1977). 'Wage indexation and macroeconomic stability.' CarnegieRochesterConference PublicPolhcy,vol. 5, pp. I07-47. Fudenberg, D. and Tirole, J. (i99I). GameTheory.Cambridge Mass: MIT Press. Giavazzi, F. and Giovannini, A. (I987). 'Models of the EMS: is Europe a greater DM area?.' In Global (eds. R. C. Bryant and R. Portes). London: Macmillan. Policy Conftictand Cooperatzon Macroeconomics: Giavazzi, F. and Pagano, M. (1988). 'The advantage of tying one's hand: EMS discipline and central bank credibility.' EuropeanEconomicReview,vol. 32, Pp. I055-75. Giovannini, A. (I989). 'How do fixed-exchange-rate regimes work? Evidence from the gold standard, (eds. M. Miller, B. Eichengreen ExchangeRateManagement Bretton Woods and the EMS.' In Blueprintsfor and R. Portes). London: Academic Press and CEPR. JOURNAL, vol. 99, pp. 293-346. Goodhart, C. (I989). 'The conduct of monietarypolicy.' ECONOMIC Gray, J. (I 976). 'Wage Indexation: A Macroeconomic Approach.' Journalof MonetaryEconomics,vol. 2, pp. 22 I-35. Kydland, F. and Prescott, E. C. (I977). 'Rules rather than discretion: the inconsistency of optimal plans.' Journalof PoliticalEconomy,vol. 85, PP. 473-92. Lindbeck, A. and Snower, D. (i 986). 'Wage setting, unemployment and insider-outsider relations.' AmericanEconomicReview,Papers and Proceedings, vol. 76, pp. 235-9. Lockwood, B. and Philippopoulos, A. (i 99I). 'Insider power, employment dynamics and multiple inflation equilibria.' Discussion Paper in Economics no. 2/9I, Birkbeck College, London. Policy, Credibilityand Politics. New York: Harwood Persson, T. and Tabellini, G. (I990). Macroeconomic Academic Publishers. Rogoff, K. (I985). 'The optimal degree of commitment to an intermediate monetary target.' Quarterly Journalof Economics,vol. IOO,pp. II69-90. Rogoff, K. and Sibert, A. (I988). 'Equilibrium political business cycles.' Reviewof EconomicStudies,vol. 55, pp. i-i6. MonetarySystem,1945-i98i. New York: Harper and Row. Solomon, R. (i982). The Internatzonal