Inferring hawks and doves from voting records

Transcription

1 Inferring hawks and doves from voting records Sylvester Eijffinger, Ronald Mahieu, Louis Raes This version: March 2017 Abstract: We analyze revealed policy preferences of the Monetary Policy Committee of the Bank of England. From the voting records we estimate the policy preferences with spatial models of voting. We find that internal committee members tend to hold centrist policy preferences, while more extreme policy preferences, both hawkish and dovish, are generally held by external members. An industry background is associated with more hawkish preferences. Keywords: Voting records; Central banking; Committees; Ideal points JEL Classification Numbers: E58, E59, C11 We would like to thank Willem Buiter, David Cobham, Jakob de Haan, Charles Goodhart, Bjørn Høyland, John Leahy, Bertrand Melenberg, Thomas Michielsen, Wieland Müller, Lucas Papademos, Peter Schotman, Christopher Spencer, Steven Verheyen, Nick Vivyan and Wolf Wagner for discussion and advice, as well as seminar discussants at the European Monetary Forum 2014, at the Tilburg Economic Workshop, at the VERC Annual Conference on Challenges for monetary policy actions and communication in uncertain times and at Heriot-Watt University. We also thank two anonymous reviewers for their input. Sylvester Eijffinger: CentER and EBC, Tilburg University, CEPR. Ronald Mahieu: CentER and Netspar, Tilburg University. Louis Raes: CentER and EBC, Tilburg University. (corresponding author). Address (all authors): Warandelaan AB Tilburg, NL. 1

2 1 Introduction Diversity of views drives the Committee to adopt an eclectic approach and thus serves to limit the risk that a single viewpoint or analytical framework might become unduly dominant. Bernanke (2007) Most central banks take monetary policy decisions by committee. These committees come in various forms. Committees differ in their decision protocol, their (in)formal hierarchy, size and composition. In this paper, we focus on the diversity of views and composition of the committee. Some committees consist mainly of career central bankers who have extensive experience at the central bank, whereas other committees consist mostly of external members recruited outside of the central bank. Besley, Meads, and Surico (2008) argue that the internal-external dichotomy could matter for at least two reasons. First, the selection procedure for internal and external members is often formally different. Second, career concerns may affect incentives. Externals leave after their terms, whereas internal members may be building a further career in central banking. Furthermore, the backgrounds of members differ. Experience in the financial industry, at a non-governmental organization (NGO) or for the government may influence how these committee members approach monetary policy making. We examine the voting record of the Bank of England. We use the observed votes to infer the policy preferences of the central bankers. We compare internals and externals along their policy preferences and analyze the impact of career backgrounds. The methodology we use builds upon a spatial voting model. A spatial voting model is a rational choice model of voting, where utilities are defined as functions of the distance between the preferred outcome and alternatives in a policy space. We use a unidimensional model to classify Monetary Policy Committee (MPC) members on a dove-hawk dimension. In this framework, a dove is a committee member who is more likely to prefer a lower interest rate out of two proposed policy rates, than a committee member labeled as a hawk. With this classification, we do not imply that MPC members target different inflation rates. Different ideal points could also arise because of different views on how the economy evolves or on different perspectives on the impact of a change in the policy rate. We present two main results. First, and in line with Hix, Høyland, and Vivyan (2010), we show that internals are not more hawkish than externals (or vice versa). Rather it seems that internals tend to hold more centrist preferences, whereas the most pronounced policy preferences (very dovish or very hawkish) are nearly always held by external members. This is related to a debate in the literature on voting at the Bank of England, where some researchers claim that internal members tend to be more hawkish, whereas other researchers argue that this is not the case. 1 Our second result is that policy preferences are only slightly influenced by career back- 1 Our results are in line with Besley, Meads, and Surico (2008), Hix, Høyland, and Vivyan (2010) but contradict Gerlach-Kristen (2009), Bhattacharjee and Holly (2010) and Hansen, McMahon, and Rivera (2014). 2

3 grounds. We compare career backgrounds by looking at the median voter per career background. An industry background is found to be associated with more hawkish preferences, whereas an NGO background is associated with more dovish preferences. The latter result does not hold when we compare NGO backgrounds with a background in the financial industry or with experience at the Bank of England. The methodology in this paper builds upon developments in other disciplines. We show how spatial voting models can easily be estimated with Bayesian methods. Throughout our empirical analysis, we demonstrate the flexibility of the methodology when analyzing voting data. It is our hope that economists will consider this methodology when analyzing voting data in the future. 2 Related literature This paper fits into the literature on decision taking at central bank committees. The institutional arrangements of monetary policy committees differ considerably across central banks and may have important implications for monetary policy making in practice. These institutional arrangements concern (i) the way decisions are reached, (ii) the transparency of procedures and (iii) the composition of such a committee. In this paper, we focus on the composition of a monetary policy committee and on the heterogeneity in policy preferences among the individual members. 2 We consider whether different types of members hold systematically different policy preferences. Existing empirical research on this topic mainly uses one of two approaches. On the one hand, researchers estimate aggregate and individual interest rate rules. Besley, Meads, and Surico (2008) estimate reaction functions for the individual committee members and assess the extent to which these capture heterogeneity in voting patterns. They group committee members according to career background (for example, academia vs non-academia) and according to their appointment within the committee (external or internal member). The parameters of the individual members are then compared across groups. The authors find no significant differences between members according to the background characteristics considered. Other examples of this approach, studying the Bank of England, are Riboni and Ruge-Murcia (2008) and Harris and Spencer (2009). The other dominant approach builds on a regression framework where the dependent variable captures the votes cast by members. This dependent variable is then regressed on relevant meeting characteristics (variables capturing economic conditions) and voter characteristics (backgrounds of the individual voters). As an example, Harris, Levine, and Spencer (2011) examine the frequency and type of dissenting votes in the MPC at the Bank of England. They find strong heterogeneity in voting patterns, but only find a weak role for career experience in explaining the decision to dissent. These findings stand in contrast to the large literature studying votes at the FOMC, suggesting that 2 See Dal Bo (2006), Riboni and Ruge-Murcia (2010) on voting procedures; Sibert (2003), Meade and Stasavage (2008) and Swank, Swank, and Visser (2008) are on the trade-offs of transparency. 3

4 career backgrounds do matter, as well as political influence through appointment. 3 We use an alternative way to investigate the voting behavior of monetary policy committee members, spatial voting models. This approach builds on methodological advancements in other disciplines where researchers have investigated the voting behavior of legislators and judges. The spatial framework yields the revealed policy preferences (ideal points) of each MPC member, which can be depicted as points on a latent dovehawk dimension. The (Bayesian) methodology yields an intuitive and user-friendly measure of policy preferences and a joint posterior distribution of all the parameters. This makes our approach much more flexible, as it allows one to make inferences about derived quantities in a way that would be impractical with the classic approaches discussed above. While economists have greatly contributed to the development of the theoretical underpinnings of the spatial voting model since the pioneering work by Black (1948), empirical implementations of the spatial voting model remain scant within economics. 4 The empirical papers we are aware of nearly always consider applications in politics and the analysis of judicial votes. 5 One notable exception is Hix, Høyland, and Vivyan (2010), who use a spatial voting framework to empirically analyze voting at monetary policy committees. 6 Their analysis looks at the voting process through the lens of political economy. 7 Specifically, they explore the extent to which the (political) appointment process was able to move the median voter. Compared to Hix, Høyland, and Vivyan (2010) we use a substantially larger sample, covering seven additional years of monetary policy meetings and hence more monetary policy committee members. We use mostly voting data on the policy rate but we also provide results on asset purchases in the appendix to this paper. We use a robust version of the basic spatial voting model and discuss model checking tools. Having estimated the preferences of the monetary policy members, we analyze systematic differences between groups of monetary policy committee members. This focus follows a strand of the literature on central bank committees, such as Besley, Meads, and Surico (2008), Chappell, Havrilesky, and McGregor (1993) and many others. 3 See Chappell, Havrilesky, and McGregor (1993) for early evidence on the appointment channel. Adolph (2013) is a recent, extensive study suggesting that career concerns matter a great deal in case of the FOMC. In contrast, Harris, Levine, and Spencer (2011) suggest that career backgrounds and the political appointment channel have only a negligible effect on voting at the MPC. 4 Exceptions are Heckman and Snyder (1997) and Henry and Mourifié (2013) who test implications of the theoretical work developed in Degan and Merlo (2009). 5 Recently, the empirical analysis of voting data regained the attention of economists. Authors have proposed different equilibrium models of decision-making to study voting at the Supreme Court. These papers typically do not build upon a spatial voting model. Iaryczower and Shum (2012) examine voting behavior in the US Supreme Court and try to quantify the value of information. Iaryczower, Lewis, and Shum (2013) investigate the trade-off between politicians and bureaucrats. 6 Chang (2003) also analyzes monetary policy committees in a spatial voting framework. Econometrically speaking, however, Chang (2003) does not estimate spatial voting models. The reasons for doing so are technical and explained in the appendix to Chapter 3 of the book. Since the publication of that book, substantial progress has been made on the estimation of spatial voting models, see Clinton, Jackman, and Rivers (2004). 7 There are two books written by political scientists which touch upon themes related to this paper such as career concerns, the effects of appointments procedures, etc.: Chang (2003) and Adolph (2013). 4

5 3 Voting records and ideal point estimation The approach presented in this paper uses voting records to estimate the policy preferences of each voter. We do so by estimating spatial voting models. 8 The data we analyze consist of voting records of monetary policy deliberations. For a given committee, we observe the votes cast on a policy rate. The data consist of MPC members to whom we refer as voters n = 1,..., N voting on policy choices t = 1,..., T. Each policy choice t presents voters with a choice between two policy rates. The lower policy rate is the dovish position, the higher policy rate is the hawkish position. A voter n choosing the hawkish position on policy choice t is denoted as y nt = 1. If voter n chooses the dovish position, we code this as y nt = 0. The canonical spatial voting model is then: P (y nt = 1) = logit 1 (β t x n α t ), (1) where x n are ideal points of the voters. 9 The parameters α t and β t are vote parameters. To understand these, consider first a situation where β t = 1. Figure 1 provides an illustration with two voters and two meetings. For each meeting t there is a corresponding vote parameter α t. This parameter captures the environment (for example, the state of the economy) in which voting takes place. A higher α t reflects a higher inclination to vote dovishly. Consider now voter 1. Voter 1 has an ideal point (x 1 ) equal to α 2 which implies that voter 1 is indifferent in meeting two between the dovish and the hawkish policy choice. 10 Voter 2 has a larger ideal point on the other hand, implying that in both meetings he has a higher probability of voting hawkishly than voter 1. Now consider the effect of β t or the discrimination parameter. This parameter captures the extent to which preferences in the dove-hawk dimension determine the choice between two competing policy rate proposals. Say we find that for a certain meeting t, β t equals zero. Then β t x n equals zero and the preferences in the underlying dove-hawk dimension do not have an impact on the choice between competing policy proposals. Analogously, a negative β t implies that doves (hawks) have a higher probability of choosing the hawkish (dovish) policy choice. A voter n is as likely to choose the dovish as the hawkish choice if his ideal point x n equals α t /β t. The model presented described by Equation 1 has quickly become the standard approach for estimating ideal points of legislators. However, the data we have in the 8 See Black (1948) and Enelow and Hinich (1984). 9 A derivation of the spatial voting model is provided in Clinton, Jackman, and Rivers (2004). Note that one can also specify different link functions, such as the cdf of normal distribution as in Clinton, Jackman, and Rivers (2004). The canonical spatial voting model presented here is isomorphic with a two-parameter item-response theory (IRT) model in which ideal points are latent traits and the parameters α t and β t are item parameters. One major difference with the IRT literature is that ideal points models focus on the ideal points, whereas as the IRT literature is mostly concerned with the item parameters. IRT models have a long history in psychometrics and educational testing and many extensions have been proposed, see Lord (1980). Foundational work on the estimation of such models was provided by Bock and Aitkin (1981) and by Albert (1992) in an MCMC setting. van der Linden and Hambleton (1996) and Embretson and Reise (2013) are comprehensive overviews. 10 Because logit 1 (x 1 α 2 ) = logit 1 (0) =

6 Figure 1: Ideal points on a single latent dimension Ideal points x1 x2 dovish hawkish a1 a2 Vote parameters This figure illustrates Equation 1. On the latent dove-hawk dimension two ideal points x 1, x 2 (voters) and two vote parameters α 1, α 2 (meetings) are shown. If the ideal point of voter n is larger than the vote parameters α t, then it is more likely that voter n will vote for the hawkish policy choice. In this example, voter 1 is as likely to vote hawkishly as to vote dovishly on the policy choice represented by α 2. context of monetary policy committees are more limited than the roll call data available to researchers investigating votes in the U.S. Senate or Congress for which these methods were developed. Consider, for example, the seminal article by Clinton, Jackman, and Rivers (2004), where the authors fit the standard spatial voting model to the roll calls from the 106th U.S. House of Representatives. This gives the authors 444,326 individual voting decisions. Compare this with our sample of only 1,121 individual voting decisions and the substantial variation in number of observed votes for each voter (see section 4). But there is another problem. Logit and probit models are not robust to outliers. This was already shown by Pregibon (1982) and more recently by Liu (2004). In this context, the term outlier refers to an observation of an outcome that is highly unexpected given the linear predictor. Bafumi, Gelman, Park, and Kaplan (2005) provide the following example. Say we have estimated a logit model P (y i = 1) = logit 1 (X i β) and we have for a particular observation i, X i ˆβ = 10. Then logit 1 (10) = , so the observation 6

7 y i = 0 would be an outlier. The many missing entries in our votes matrix and the fact that we observe only a limited number of votes for some voters potentially aggravate the problem of outliers. Only a few outliers could substantially bias our parameter estimates. A modification to the standard voting model to become more robust against outliers (in the sense explained earlier) is proposed in Bafumi, Gelman, Park, and Kaplan (2005). The idea is to add a level of error ɛ 0 and ɛ 1 to the model presented by equation (1): P (y nt = 1) = ɛ 0 + (1 ɛ 0 ɛ 1 )logit 1 (β t x n α t ). (2) Now, every voter has an immediate probability of voting hawkishly ɛ 0 and of voting dovishly ɛ 1. The initial item-response model applies then to the remaining outcomes. This simple modification makes the standard spatial voting model more robust and is straightforward to implement. In the appendix to the paper we report model checks suggesting that the robust modification is preferable over the standard model in our application. 3.1 Priors Our approach is Bayesian, so we specify priors as shown in Table 1. The priors on the ideal points serve to identify the parameters. The parameters in the spatial voting model are not identified. As can be seen in Figure 1, the probabilities depend on the relative position of ideal point and vote parameters. We could add a constant to β t x n and to α t and the predictions would not change. Analogously, we could multiply β t by a constant and divide x t by the same constant. A popular approach (see Clinton, Jackman, and Rivers (2004)) to overcome this is to restrict the ideal points to have mean zero and a standard deviation of one. Now we only need to fix the dove-hawk direction. Remember that we assumed that votes coded as zero were votes favoring the most dovish (the least restrictive) interest rate of two proposed policy rates. Restricting β t to nonnegative values then makes sense. It implies that we explicitly model the directionality of each vote, which is clear in this application. We restrict β t to be nonnegative by choosing for each β t a diffuse normal prior with a positive mean, but which is truncated at zero. 11 For the vote-difficulty parameters α t we also choose diffuse priors. The prior choice for ɛ 0 and ɛ 1 follow the recommendations of Bafumi, Gelman, Park, and Kaplan (2005). This prior choice restricts the values of these parameters to lie in the interval [0, 0.1]. This is not restrictive because if we were to find values which are close to 0.1, suggesting an immediate chance of success or failure of 10%, then a logit-type model should not even be used as an approximation, see Bafumi, Gelman, Park, and Kaplan (2005). In the online appendix to this paper we demonstrate that alternative reasonable prior specifications do not meaningfully affect the results presented in this paper. 11 Cromwell s rule states that if a particular region of the parameter space has zero prior probability then it also has zero posterior probability, see Jackman (2009). So by restricting the support of the prior, we have in fact restricted the β t parameters to be nonnegative. 7

8 Table 1: Prior specification Parameter Baseline Prior ɛ 0 Uniform(0, 0.1) ɛ 1 Uniform(0, 0.1) α t N(0, 4) β t N(1, 4) truncated at 0 x n N(0, 1) This table provides an overview of the priors we specified. Note that we assume independence across voters and meetings. We acknowledge that voting could depend on a whole range of influences, including personal and group preferences (for example, through an organizational consensus or varying reputational concerns). Identifying each of these requires considerably more data and/or assumptions. The measures of revealed policy preferences we propose in this paper, represented by the ideal points, are therefore a mix of these influences on monetary policy voting rather than a literal measure of policy preference. 12 In our opinion, these serve as a useful summary of policy preferences and could aid researchers analyzing monetary policy votes. 4 The data We study the voting records of the Monetary Policy Committee (MPC) of the Bank of England. This committee is classified as individualistic by Blinder (2007) which implies that the vote of an individualistic committee conveys genuine information. An individualistic committee does not preclude that members listen to and learn from each other, but we do assume that voters vote according to their preferences. It is commonly accepted that this is true in the case of the Bank of England. 13 The individualistic nature of the MPC is reflected in the voting record. In about 60% of the MPC meetings held between June 1997 and November 2015, a decision was taken by non-unanimous votes. The raw data contain the votes of 38 MPC members at 223 meetings. We drop unanimous votes, as these are uninformative for our purposes. 14 The remaining votes are coded as decisions over two alternatives. Table 2 clarifies the coding with two examples. Example 1 is the situation where there were only two policy choices to vote on in a given meeting. Nickell voted in that particular meeting for a lowering of the interest 12 See also the discussion in Jackman (2009) p See for example Jung (2013): In committees with individual accountability (Bank of England s MPC, Riksbank s Executive Board), members may have a strong incentive to reveal accurate information on their preferences within the voting records. Chappell, McGregor, and Vermilyea (2014) furthermore show that the governor has little influence over the votes of other committee members at the MPC. 14 Ideal Point estimation boils down to classifying voters relative to each other. A meeting where all voters vote in the same manner does not provide information regarding the differences between the ideal points of MPC members. 8

9 rate with 25 basis points, whereas the other voters preferred to keep the interest rate unchanged. In this case, Nickell selected the dovish option so his vote is coded as 0, whereas the others chose the hawkish choice and therefore their vote is coded as 1. If a meeting involved a choice with more than two interest rates, we coded these as a series of choices over pairwise alternatives, as shown in example 2 in Table 2. We also removed Howard Davies from our data set as we did not observe any vote from him after having dropped the unanimous meetings. This left us with 126 meetings and 37 MPC members, henceforward referred to as voters. 15 Table 2: An example of the coding scheme Example 1: April Example 2: May Name Vote Code 1st time Vote Code 1st time Code 2nd time King Lomax Tucker Bean Barker Nickell Walton Gieve This table explains how the data were coded. Example 1 shows the situation where there were only two alternatives favored. In Example 2, votes were split among three policy choices. In Figure 2, we provide a summary of the individual voting records of the MPC members with two graphs. The left graph shows the total number of votes we have for each MPC member after recoding the dataset as described above. This graph shows a wide variation in the number of votes we observe for the different voters. We have 117 votes of Mervyn King in our data set, while we have only 3 votes of Vlieghe. The right graph shows the fraction of votes of each MPC member, coded as dovish. 5 Ideal points at the Bank of England We start by estimating the ideal points of the MPC members. Figure 3 presents the ideal points of the 37 voters along with the uncertainty in the estimates. The Figure provides a historical ranking of the MPC members, taking all members into account. This allows us to compare the policy preferences of, for example, King and Carney, although they never served at the same time. We would roughly classify MPC members with negative ideal points and 95% uncertainty intervals who do not overlap zero as doves, and members with positive ideal points and 95% uncertainty intervals 15 From here onwards we use the term meeting for one recorded voting session in our dataset. This means that a real monetary policy meeting could occupy two meetings in our dataset when three policy choices were considered. The uncoded data did not contain records of votes where four or more policy choices were considered. 9

10 who do not overlap zero as hawks. We can also compare MPC members directly. We can explicitly construct hypotheses such as Was Mervyn King more hawkish than Mark Carney?. The Figure provides a visual aid to quickly deduce the answer. If two 95% intervals do not overlap, then the ideal points are significantly different from each other. 16 We see that the ideal points of King and Carney do not overlap, suggesting that Carney is more dovish than King. An explicit calculation verifies that this is indeed the case, P (x Carney < x King ) > We will explore various hypotheses in Section 6 and Model checking Before we proceed, we need to verify that the estimation results pass some checks. These checks give an insight into how well the model fits the data and the impact of certain assumptions. 17 We conduct three checks. First, we compare the performance of the robust model we use with the standard spatial voting model. In the appendix to this paper, we show that the robust model provides more stable estimates, i.e. estimates which are not sensitive to the priors we impose to identify the model. In particular, the ideal point of Willem Buiter becomes sensitive to the prior choices when using the standard spatial voting model. 18 Second, we assess the impact of the identifying constraints (the constraints on the discrimination parameters). The results of this check are presented in the appendix. We find that the constraints are reasonable and do not contrast with the data. Finally, we perform a check through the calculation of prediction errors. 19 Define a prediction error pe nt as: pe nt = 1 if E(y nt ) > 0.5 and y nt = 0, or E(y nt ) < 0.5 and y nt = 1 = 0 otherwise. (3) Using the prediction errors, we can quickly check the error rate or the proportion of times the prediction is wrong. The error rate for our preferred model is less than 8%. This can be compared to the error rate of the null model, where we assign the same probability to each y nt. 20 The error rate of the null model is close to 50%, so our model fits considerably better. We can also use the prediction errors to consider the excess error rate i.e. the proportion 16 This is just a rough check. Even when 95% uncertainty intervals overlap, it could be the case that the difference in ideal points is distinguishable from zero. 17 We draw on Gelman, Goegebeur, Tuerlinckx, and Van Mechelen (2000), Gelman, Carlin, Stern, and Rubin (2003) and Bafumi, Gelman, Park, and Kaplan (2005). Gelman and Shalizi (2013) provide additional references. 18 See also the appendix for additional discussion hereon. 19 This model check closely follows the suggestions in Bafumi, Gelman, Park, and Kaplan (2005). 20 See (Gelman and Hill (2006) chapter 5. 10

11 Table 3: Prediction errors Voter Excess Voter Excess Voter Excess error rate error rate error rate George King Lomax Large Tucker Bean Barker Nickell 0.04 Allsopp Bell Lambert Budd Buiter 0.25 Goodhart Vickers Julius Wahwani Plenderleith Clementi Walton 0.25 Gieve Blanchflower Besley 0.03 Sentance Dale Fisher Miles Posen Weale Broadbent McCafferty Carney Cunliffe Forbes Haldane Shafik Vlieghe This table reports the excess error rates for the voters as defined in the text. An excess error rate refers to the proportion of errors beyond what would be expected given the parameter predictions of the model. The largest excess error rates (Buiter and Walton) are in bold. of errors beyond what is expected. If the model is true, the probability of error is: E(e nt ) = min(ɛ 0 +(1 ɛ 0 ɛ 1 )logit 1 (β t x n α t ), 1 (ɛ 0 +(1 ɛ 0 ɛ 1 )logit 1 (β t x n α t ))) (4) which is the minimum of the model s prediction and 1 minus the prediction. The excess error ee nt is then: ee nt = pe nt E(e nt ). (5) To be able to interpret the prediction errors, we consider averages of errors, as these offer 0 as a baseline, see Bafumi, Gelman, Park, and Kaplan (2005). In Table 3, we show the excess error rate for the voters. The ideal points for Buiter and Walton have the highest excess error rates. We find for both an error rate which is 25% higher than what we would expect if the model were true. For a reference on the size of error rates we can expect, see Bafumi, Gelman, Park, and Kaplan (2005), where the authors report excess error rates varying from to with outliers over 0.5. So, taken together, the excess error rates are low, suggesting that we can estimate the ideal points well. Additionally, we could use the estimated parameters to generate reference excess errors based on replicated data. Such an exercise is reported in the appendix to this paper. The results of the comparison of the realized errors to the reference errors confirm the conclusion that only the ideal points of Buiter and Walton have a remarkable excess error rate. 11

12 5.2 Data on asset purchases Since March 2009, the MPC of the Bank of England also votes on asset purchases financed with central bank reserves. We can integrate these voting data into our framework by coding the votes in a similar way as we did for the policy rate votes. These new vote data provide additional information to identify the ideal points of the voters. However, we only have voting data on asset purchases for a limited period, hence only for a subset of voters. Moreover, we need to be willing to assume that a vote on the asset purchase program and the policy rate can be used on an equal footing in our spatial voting model. We provide the results of the combined dataset in the appendix of this paper. Here we limit ourselves to a comparison of the point estimates of the ideal points (median of the posterior). The first thing we notice in Figure 4 is that a few ideal points, that were close together when considering only votes on the policy rate, diverge a lot when including data on asset purchases. Posen becomes much more dovish whereas Fisher and especially Broadbent become more hawkish. Miles is an example of an ideal point that remains more or less stable. The reason for these shifts lies in the voting record on asset purchases. Broadbent was always on the hawkish side when voting on asset purchases whereas Posen was most of the time on the dovish side. 21 The shifts of Dale and Weale can be explained in a similar way. Asset purchases (and recently corporate bond and goverment purchases) fall in the realm of unconventional monetary policy. Whether we should include votes on asset purchases to determine ideal points seems a matter of taste. Media coverage certainly seems to factor in votes on asset purchases to classify MPC members. Adam Posen was routinely portrayed as an arch-dove, mostly because of his stance on asset purchases and quantitative easing. Our results with asset purchase votes supports this viewpoint. We have chosen however, to present results in this paper which are only based on the votes regarding the policy rate. We deem this choice to be more conservative. One could worry that since the lower bound was hit, decision making at the MPC changed. We believe that this is less likely to be an issue with votes on the policy rate. We refer the interested reader to the online appendix to this paper, where we discuss the results with votes on asset purchases in more detail. 21 We coded the voting records on asset purchases in the same way as we coded votes on policy rates. We only consider meetings where votes on asset purchases were not unanimous. The lower proposal was then coded as a zero (the dovish choice) and the higher proposal as a one (the hawkish choice). 12

13 Figure 2: Overview votes by MPC members Total votes by MPC members Percentage dovish votes by MPC members George King Lomax Large Tucker Bean Barker Nickell Allsopp Bell Lambert Budd Buiter Goodhart Vickers Julius Wadhwani Plenderleith Clementi Walton Gieve Blanchflower Besley Sentance Dale Fisher Miles Posen Weale Broadbent McCafferty Carney Cunliffe Forbes Haldane Shafik Vlieghe % 20% 40% 60% 80% 100% Total votes Percentage dovish Here we present the voting records of MPC members in two graphs. These graphs are based on the cleaned and recoded data which is used in the subsequent analysis. The left graph shows the total number of votes we have for each MPC member. The right graph shows the percentage of total votes coded as dovish. Eleven MPC members have voted for the dovish policy proposal in every non-unanimous meeting they attended while two MPC members have voted for the hawkish policy proposal in every non-unanimous meeting they attended. 13

14 Figure 3: Ideal points at the Bank of England: overview Revealed Preferences in the MPC Sentance Besley Large McCafferty Budd Buiter Weale Vickers Dale Walton King Gieve Goodhart Tucker Lambert Barker Lomax Clementi George Bean Plenderleith Vlieghe Nickell Bell Fisher Posen Forbes Carney Shafik Cunliffe Haldane Broadbent Allsopp Miles Julius Wadhwani Blanchflower 95% credibility interval Dove Hawk This figure is a graphical representation of the estimated ideal points of the MPC members. A point indicates the estimate of the ideal point and the thin line represents the 95% interval. A triangle indicates an external MPC member, a circle indicates an internal MPC member. 14

15 Figure 4: Comparison of ideal points Posen Miles Fisher Broadbent Dale Weale 3 3 This figure provides a comparison of the estimated ideal points (medians of the respective marginal posterior densities). Ideal points on the bottom of the figure correspond to the ideal points in Figure 3. The ideal points at the top of the figure are estimated using a dataset which also comprises votes on asset purchases. 15

16 6 Internals and externals In this section, we investigate whether groups of members systematically differ in their policy preferences. First, we consider the differences between internal members who have a full-time executive position at the Bank and external members who have no executive responsibilities. In Table 4, we provide an overview of all MPC members we consider in our analysis as well as some information on career backgrounds. The information in this table comes from Harris, Levine, and Spencer (2011), which we updated for the MPC members who joined after In Figure 3 externals are indicated by a triangle while internals are marked by a circle. At first sight, it seems that externals and internals do not seem to be easily classified as either dovish or hawkish. As can be seen in Figure 3, outspoken doves such as Blanchflower and Wadhwani as well as clear hawks such as Sentance and Besley both belong to the external group. We notice that the internals tend to take the centrist position while the positions of externals are more spread out. The extreme positions tend to draw additional attention in media reports on the MPC. MPC members who are considered to be very dovish or hawkish are often referred to as arch-hawk or arch-dove. 23 It is enticing to look at Figure 3 and declare Sentance and Blanchflower the all time arch-hawk and arch-dove. However such an approach ignores the fact that we have uncertainty regarding the ideal points which should translate into uncertainty regarding the ranks. To verify that indeed these two external members have the most outspoken ideal points, we want to infer the ranks of the estimated voting scores. Our Bayesian simulation results facilitate this analysis. As explained in Jackman (2009), given the joint posterior density over the N ideal points x = (x 1,..., x N ) we can induce a posterior density over any quantity of interest that is a function of the x. 6.1 The most dovish and hawkish voters To obtain a posterior density over the order statistics for each voter, we use the following procedure. 24 For each MCMC draw k = 1,..., K, we assign a rank r to the N sampled ideal points x (k) n = x (k) 1,..., x(k) N. We denote these ranks at each iteration as r(k) = (r (k) 1,..., r(k) N ). Each element of r (k) is an integer r n (k) {1,..., N}. The probability that voter n occupies rank r is thus 1 K K k=1 1 {r}r n (k). Computing these ranks over the iterations of our sample yields a posterior mass function over the possible ranks. We now consider only rank 1 and rank N to obtain the probability of being the most 22 We thank Christoper Spencer for providing background on the original classification and advice on the update. 23 The top search results for simple web search on arch-hawk or arch-dove are media reports on the MPC. For example a profile of Andrew Sentance: co.uk/finance/economics/ /andrew-sentance-profile-of-the-mpcs-arch-hawk.html or an article on Adam Posen inflation-worries-see-arch-dove-posen-drop-support-for-qe html. 24 See also Jackman (2009), chapter 9. 16

17 Table 4: Overview of backgrounds Name Internal or Source Finance Industry Government Academia Bank NGO Research Field External George, Eddie Internal HLS N N N N Y Y King, Mervyn Internal HLS N N N Y Y N Public Finance Clementi, David Internal HLS Y N N N N N Large, Andrew Internal HLS Y Y Y N N N Lomax, Rachel Internal HLS N N Y N N Y Gieve, John Internal HLS N N Y N N N Carney, Mark Internal Own Y N Y N Y N Cunliffe, Jon Internal Own N N Y N N N Shafik, Nemat Internal Own N N Y Y N N Development Plenderleith, Ian Internal HLS N N N N Y Y Dale, Spencer Internal Own N N N N Y N Fisher, Paul Internal Own N N N N Y N Vickers, John Internal HLS N N N Y N N Competition Policy Bean, Charles Internal HLS N N Y Y N N Macro (Monetary and Labor) Tucker, Paul Internal HLS Y N N N Y N Haldane, Andrew Internal Own N N N N Y N Buiter, Willem External HLS N N N Y N N Macro, Monetary, Public Finance Goodhart, Charles External HLS N N Y Y Y N Monetary Julius, DeAnne External HLS N Y N Y N Y Foreign investment, Governance Budd, Sir Alan External HLS Y N Y Y N N Monetary Wadhwani, Sushil External HLS Y N N Y N N Labor, Financial Nickell, Stephen External HLS N N N Y N N Labor Allsopp, Christopher External HLS N N Y Y Y N Fiscal, Reform, Monetary Barker, Kate External HLS N Y N N N N Bell, Marian External HLS Y N Y N N N Lambert, Richard External HLS Y N N N N N Walton, David External HLS Y N Y N N N Blanchflower, David External HLS N N N Y N N Labor Besley, Tim External HLS N N N Y N N Development, Political Sentance, Andrew External HLS N Y N Y N N Monetary, Environmental Weale, Martin External Own N N N Y N N Labor, Econometrics, Public Finance Miles, David External Own Y N N Y Y N Financial Posen, Adam External Own N N N Y Y N Monetary Broadbent, Ben External Own Y N Y Y N N Monetary McCafferty, Ian External Own Y Y N N N N Forbes, Kristin External Own N N Y Y N N International Finance Vlieghe, Gertjan External Own Y N N N Y N The information provided in this table draws upon tables provided by Harris, Levine, and Spencer (2011), which we updated for members who joined the MPC after May The column labeled Source indicates whether the information comes from Harris, Levine, and Spencer (2011), abbreviated as HLS or whether the classification is our own. Y (N) stands for Yes (No) and means that the monetary policy member does (not) have career experience in that sector. Some classifications required judgment calls. In the web appendix to this paper, we provide some clarifications to the classifications we added. 17

18 dovish and the most hawkish for each voter. Table 5 reports all the voters for whom we find a non-trivial probability of being the most dovish or the most hawkish (probability > 1%). If we consider the voters with a probability of being the most dovish, we see that the historical arch-dove is likely to be someone from the group Blanchflower, Wadhwani, Julius. Similarly we find that for the group of voters who have a non-trivial probability of being the most hawkish, only Large is an internal member. Taken together, our analysis re-emphasizes that the most dovish and most hawkish members at the MPC of the Bank of England so far were externals. Internal members tend to occupy the middle ground. This observation emphasizes what Hix, Høyland, and Vivyan (2010) found and sheds light on a topic which has been heavily discussed in the literature. Besley, Meads, and Surico (2008) could not classify external or internal members as either more dovish or more hawkish. Our ideal point estimates and rank estimations make clear why. Bhattacharjee and Holly (2010) and Gerlach-Kristen (2009) suggest that internals tend to be more hawkish than externals, but this could be driven by shorter samples. The evidence that externals tend to have members among their ranks with more outspoken policy preferences supports the idea that differences between internals and externals could arise because of an organizational consensus among internals as argued by Gerlach-Kristen (2003) and Harris and Spencer (2009). Related to this, there may be career concerns among internals which are less relevant for external members. Gerlach- Kristen (2003) suggests that externals may even be incentivized to gain media attention. Our analysis does not focus on the act of dissenting per se, but rather on the revealed policy preferences. 7 Career backgrounds Related to the differences between internals and externals, we are also interested in career background effects. The intuition is that career backgrounds may persistently influence the policy preferences of MPC members. This notion comes from the literature investigating voting at the FOMC, where such effects have been suggested. Adolph (2013) provides a detailed analysis linking career backgrounds, preferences and votes. He concludes that in case of the FOMC career effects matter a great deal and should be taken seriously. In contrast, the aforementioned study by Harris, Levine, and Spencer (2011) only finds weak (often counterintuitive) influences of career backgrounds when analyzing the records of dissents. Besley, Meads, and Surico (2008) consider fewer career background characteristics when comparing coefficients of reaction functions. They do not find a meaningful pattern. We investigate the differences in policy preferences by comparing voters with experience in (1) the finance industry (including banks), (2) industry in general (excluding the financial industry), (3) government (civil service or working for any government), (4) academia (only an appointment post doctoral education counts; most voters have 18

19 Table 5: Most dovish and most hawkish voters Voter Probability of being the most dovish Voter Probability of being the most hawkish Blanchflower 42% Sentance 59% Wadhwani 28% Besley 21% Julius 14% Large 10% Miles 2% McCafferty 9% Posen < 2% Fisher < 2% Shafik < 2% Cunliffe < 2% Carney < 2% Haldane < 2% Broadbent < 2% Forbes < 2% This table reports the probability of being the most dovish and being the most hawkish. Policy preferences are estimated with uncertainty. Small voting histories and lop-sided voting records induce uncertainty. Our ranking of voters incorporates this and for each rank we have a probability mass distribution. In this table we report the voters for whom we found a probability of more than 1% of being the most dovish or the most hawkish, based on our preferred model and draws from our MCMC algorithm. 19

20 obtained a Ph.D.), (5) at the Bank of England, (6) at an NGO. An overview is provided in Table 4. These groups overlap, so some voters belong to multiple groups. In Figure 5 the same ideal points as in Figure 3 are plotted but we have grouped the ideal points according to career backgrounds. If an MPC member has had several different careers, the corresponding ideal point appears multiple times. Figure 5 already suggests that Bank ideal points and NGO ideal points tend to be a bit more clustered and a bit biased towards the dovish direction. 25 We also notice that Academia ideal points have a wide spread. We can go beyond eyeballing Figure 5 and analyze the impact of career experiences by computing the predicted probabilities of voting hawkishly. Allison (1999) outlines how group comparisons in these models are complicated by confounding residual variation, an issue which does not arise with linear regression coefficients. Long (2009) suggests comparing predictive probabilities as a solution to the problems raised by Allison (1999). Such an approach requires us to compare groups at different levels of the variables in the model. To compare the predicted probabilities of voting hawkishly for the different groups, we proceed as follows. First, we fix ɛ 0, ɛ 1 in our ideal points model at their estimated median value. We fix β t at 1.5, which is the lower quartile of the point estimates of the discrimination parameters. Then we calculate the predicted probability of voting hawkishly for the median voter of each group l, for each meeting t and each iteration k, resulting in an , 000 array of predicted probabilities. To compare different groups, we then compute the difference in predicted probabilities between different median voters for the corresponding meeting-iteration combinations. We obtain an estimate of the difference in predicted probability for each meeting by calculating the median of all the calculated differences. Likewise, we obtain an estimate of the uncertainty by calculating the 2.5% and 97.5% quantiles. We present the results of these calculations graphically in Figure 6. Each graph shows the difference in predicted probability for two median voters. For example, the upper left graph titled Industry - Academia shows the estimated difference in probability of voting hawkishly between the Industry median voter and the Academia median voter. The grey area presents the 90% uncertainty level, whereas the dashed lines represent the 95% uncertainty level. The median voter with an industry background is more likely to vote hawkishly than any other median voter, regardless of the overall inclination to vote dovishly (the α parameter). We also notice that the difference between backgrounds is largest for intermediate values of the parameter α. Meetings with a large or small α parameter are meetings where circumstances are such that most voters are inclined to vote dovishly or hawkishly. The median voters with non-industry backgrounds are much harder to distinguish from each other and we do not find differences at 95% except for voters with an NGO background. The median voter with an NGO background is more likely to vote dovishly than voters with a background in the government, at the Bank of England or in the private industry. 25 The clustered ideal points in case of a Bank background could point towards an organisational consensus as argued by Gerlach-Kristen (2003). 20

21 Figure 5: Ideal points split according to career backgrounds NGO Bank Academia Government Industry Finance Dove Hawk In this graph we plot the same ideal point estimates as in Figure 3, but we now split the ideal points according to career backgrounds. If an MPC member has several different career backgrounds, then the corresponding ideal point appears multiple times. 21

22 Figure 6: Comparison of predicted probabilities Industry Academia Industry Finance Industry NGO Industry Government Industry Bank Academia Finance Academia NGO Academia Government Academia Bank Finance NGO Finance Government Finance Government NGO Government NGO Bank Government Bank In these graphs, we present a group by group comparison of predicted probabilities. Each graph shows the difference in the predicted probability of voting hawkishly between the median voter of two groups. The dashed line represents the 95% confidence interval, and the grey area represents the 90% confidence interval. 22

23 Harris, Levine, and Spencer (2011) found that industry experience and work experience at the Bank of England promote tightness dissents -both findings were deemed to be counterintuitive. Our findings suggest that while we find a more hawkish policy preference among those with industry experience, those with work experience at the Bank of England do not hold more hawkish policy preferences. Experience in government, academia or in finance does not seem to systematically shift the policy preferences. These results are in line with Harris, Levine, and Spencer (2011) and Besley, Meads, and Surico (2008), who could not find systematic differences in the estimated parameters of reaction functions when comparing voting members with academic experience and without. When we look at the background of the voters (Table 4) mentioned in Table 5, we find that these are often voters with an academic background. One explanation could be that academics may have developed their own, idiosyncratic view on what monetary policy should do and are subsequently more pronounced in their opinions and preferences. This might be related to the field of expertise. A labor economist and a monetary economist might place a different emphasis due to their academic work. In Table 4, we provide an overview of the different research fields. 26 The two most popular classifications are labor and monetary economics. However, in both categories we see people with dovish and hawkish ideal points. Most labor economists tend to be dovish, with Weale being the main exception. Among the monetary economists, the ideal points are more spread out. Broadbent, Allsopp, Posen are dovish, but Sentance and Budd are more hawkish. Compared with Academia, voters with other career backgrounds, be it in government, at a central bank or in the financial industry seem more homogenous. 26 Note that classifying academics into categories is complex, because many have worked on different issues. 23