The Impact of Political Uncertainty on Asset Prices: The Case of the United Kingdom's EU Membership Referendum
Author:
Mr. Niels-Jakob H Hansen
Search for other papers by Mr. Niels-Jakob H Hansen in
Current site
Google ScholarClose
and
Ms. Margaux MacDonald
Search for other papers by Ms. Margaux MacDonald in
Current site
Google ScholarClose

Contributor Notes

Author’s E-Mail Address: jmh378@georgetown.edu, NHansen@imf.org, MMacDonald@imf.org
Publication Date:
05 Feb 2021
eISBN:
9781513568645
Language:
English
Keywords:
Brexit; EU referendum; political uncertainty; high frequency data.; equity prices in the UK; EU membership
Download PDF (1.7 MB)

How did expectations of the outcome of the United Kingdom's (UK) referendum on European Union (EU) membership in 2016 affect prices in financial markets? We study this using high frequency data from betting and financial markets. We find that a one percentage point increase in the probability of "Leave" result caused British stocks (FTSE All-Share) to decline by 0.004 percent, and the Pound to depreciate by 0.006 percent against the euro. We find negative and significant effects for most sub-sectors, and negative spill-overs to other EU member countries. We show that the differential impact across sectors and countries can be explained by differences in the trade exposures.

Abstract

How did expectations of the outcome of the United Kingdom's (UK) referendum on European Union (EU) membership in 2016 affect prices in financial markets? We study this using high frequency data from betting and financial markets. We find that a one percentage point increase in the probability of "Leave" result caused British stocks (FTSE All-Share) to decline by 0.004 percent, and the Pound to depreciate by 0.006 percent against the euro. We find negative and significant effects for most sub-sectors, and negative spill-overs to other EU member countries. We show that the differential impact across sectors and countries can be explained by differences in the trade exposures.

I. Introduction

The United Kingdom’s (UK) referendum on European Union (EU) membership (“the referendum”) held on June 23rd 2016 provides an opportunity to study the impact of political uncertainty on asset prices. The the implications of a “Stay” result were relatively clear: the UK would remain in the EU and policies would be unchanged. The implications of a “Leave” result were more uncertain, both ex ante and ex post. As events following the referendum have shown, a “Leave” result did not give clear guidance on the specifics of an exit from the EU, for example the time for invoking Article 50 or trade policies following an exit. Thus, in expectation a “Leave” result meant much more policy uncertainty than a “Remain” result (Figure 1). There would also be spillovers to other EU countries. Therefore, a “Leave” result could have important implications for equity prices, both through the expected profitability of firms (and the UK economy more broadly) and through the added uncertainty. Using high frequency data from a political prediction market (i.e. an online betting platform), we examine how fluctuations in peoples’ expectations of the outcome of the referendum affected the market value of UK and European equities in the run-up to the referendum. The results are informative for how asset prices react to fluctuations in political uncertainty. However, the uncertainty we study is specific to the EU referendum result, which reflects the chance of leaving the EU and the political uncertainty that entails.

Figure 1.
Figure 1.

Policy uncertainty and the British EU Referendum in 2016

Citation: IMF Working Papers 2021, 027; 10.5089/9781513568645.001.A001

An alternative way of assessing the effect of a political event is through event studies. Some examples of such studies are Fisman et al. [2012], Jayachandran [2006], and Auerbach and Hassett [2007]. However, this method suffers from lack of variation as a given political event rarely happens repeatedly. In contrast, betting odds on a future political event are observable with high frequency, and can react almost instantaneously to new information as individuals place bets online. These odds thus display much more variation than the outcome of the actual political event. Our data contains prices for all trades made on Betfair (an online betting platform) during the run-up to the EU referendum, and the prices of various financial assets. By regressing the high frequency betting odds on financial market prices we get an estimate of how changes in the perceived probability of a “Leave” result, which implies substantially more policy uncertainty than a “Remain” result, affects asset prices.

We find that changes in the probabilities of a “Leave” result had a meaningful short run impact on asset prices. In the very short run (within 5 minutes), an increase in probability of a “Leave” result of one percentage point caused (i) British stocks (as measured by FTSE All-Share) to decline by 0.004 percent, and (ii) the Pound to depreciate by 0.006 percent against the Euro. Thus, a ten percentage point increase in the probability of a “Leave” result caused the stock market (FTSE all) to lose a value of around GBP 1 billion. We also find negative and significant effects for most sub-sectors in the FTSE all index. We also find evidence of negative spill-overs to other EU member countries. We show that the different impact across industries and countries can be explained by differences in the amount of trade exposures. Our results strengthen when we allow for 24 hour for adjustment. This could reflect lags in information processing of the market.

The validity of our results relies on several assumptions. First, we assume that betting odds in the run up to referendum reflected financial market participants’ view on the referendum outcome. We find this assumption plausible given the sheer volume of bets placed leading up to the referendum. This reassures us that betting on the referendum was not a niche industry for a small number of specialized participants, but rather a broader representation of market expectations. Betfair is the world’s largest online betting platform, and in 2016 its Brexit market had the largest number of transactions ever processed for a political event up to that point. Second, we assume that movements in equity prices which occur shortly after a change in the betting odds, are the price effect of a change in the expected outcome of the referendum. We test this assumption by varying the time-window we examine. Additionally, since the betting platform is easily accessible online, we believe it is reasonable to assume that the betting markets react quickly to events (such as speeches by politicians, publication of polling results, opinion pieces in major newspapers) in a similar manner as asset prices do.

Our work provides empirical support to a theoretical literature that has studied the impact of political uncertainty on financial markets [Pastor and Veronesi, 2012, 2013, Brogaard and Detzel, 2015, Kelly et al., 2016, Boutchkova et al., 2012, Beaulieu et al., 2005, Hil et al., 2019] and on economies more generally [Bloom, 2009, Baker et al., 2016]. In particular, our results are consistent with the leading theoretical model by Pastor and Veronesi [2012], whose show that asset prices should fall on the announcement of policy changes, more so when the associated political uncertainty is large, and with increasing magnitude as the probability of an event increases. In our case rather than formal policy announcements, we focus on indirect policy changes implied by changes in the probability of a “Leave” vote. We also provide empirical evidence consistent with Boutchkova et al. [2012], who find that political uncertainty in countries of trading partners and trade-dependent industries results in greater asset price volatility. Finally, it is a member of the growing body of literature on prediction markets summarized by [Horn et al., 2014, Wolfers and Zitzewitz, 2004].

Our work also contributes to the relatively sparse literature on the EU referendum’s impact on global asset prices. As far as we are aware, only two other studies use betting data to examine the implied market costs of the referendum Coleman et al. [2018] use betting data to study how the equities of US banks responded to a possible “Leave” vote in the referendum. They find that leading up to the vote, the equity prices of US banks fell by 0.003 to 0.0054 percent in response to a 1 percent change in the probability of Brexit. The US banks with more extensive UK operations were affected the most. In contrast, we do not find significant results for the UK based banks and financial firms in the short run. This is potentially attributable to the expected regulatory changes following a “Leave” vote or that investors did not expect UK banks to face the same geographic and legal organizational issues as their US counterparts. Belke et al. [2018] uses betting data in addition to policy uncertainty measures to estimate the implied cost of the referendum on equities, currencies, and sovereign debt in the UK, and some EU/OECD countries. They find that an increase in the probability of Brexit caused an depreciation in the British pound and has strong negative effects on the European stock markets, which is in line with our results. Importantly, both Coleman et al. [2018] and Belke et al. [2018] use betting data at daily frequency. In contrast, we use betting data at much higher frequency to estimate the impact of changes in expectations to the referendum result on currencies and equities across industries. Pairing high frequency betting and financial data allows us to make a more precise identification of these relationships.

Other papers have studied the impact of the referendum result on equity and currency volatility a few days after referendum [Ehler et al., 2017, Quaye et al., 2016] and over a longer horizon [Kurecic and Kokotovic, 2018, Caporale et al., 2018]. Relatedly, Hohlmeier and Fahrholz [2018] studies the consequences of the “Leave” result on on regulations governing financial activities between the UK and EU. There is also a growing literature examining how prolonged uncertainty about post-Brexit policy is impacting the broader UK economy [[Lapin-ska and Orak, 2020, Ries et al., 2020, 2017, Bloom et al., 2019]] and other countries around the world Hassan et al. [2019]. Unlike our work, these papers focus on the post-referendum financial or real outcomes and not on the impact of perceptions on markets before the vote.

The paper proceeds as follows. Section II provides background on the referendum and explains the mechanics of the betting platform Betfair. Section III describes our data. Section IV presents our baseline estimations. Section V contains the robustness checks of our baseline specifications and some additional analysis. Section VI concludes.

II. The Referendum and Betting Platform

On February 20, 2016, the British government announced a referendum on the future of British membership in the European Union (EU), to be held on June 23, 2016. On the referendum day, the question posed to voters was “Should the United Kingdom remain a member of the European Union or leave the European Union”? Voters chose from two answers, “Remain a member of the European Union”, or “Leave the European Union”.

The announcement of the vote was not unexpected, as Prime Minister David Cameron had pledged to hold such a vote during his re-election campaign in 2015. In response to this promise, Betfair, an online exchange where people can place bets against each other rather than a traditional bookmaker, created a market for people to bet on the outcome of the referendum. The question presented to betters was, “What will be the EU Referendum Result?” Betters could choose between two outcomes, “In favor of staying in EU” and “In favor of leaving EU”. In this two-sided market, Betfair takes on no risk, it provides a platform for customers to match bets against each other and in turn takes a small commission on the winnings. Betters with a placed bet could either wait until after the referendum to cash out any winnings, or sell their bet to take advantage price movements in the bet before the election.

The exchange works in the following way. Betters can either match an existing bet placed by another person, or place a new bet with their desired odds for other betters to match. There are two types of bets a person can place, a back bet and a lay bet. A back bet pays out if the outcome occurs, while a lay bet pays out if the event does not occur. For example, a better can put in an offer to pay out 300 GBP if a certain outcome materializes, against the opportunity to receive 100 GBP if the outcome does not materialize. This is a back bet. Conversely, the better can put in an offer to payout 100 if the outcome does not materialize, against the opportunity to receive 300 if the outcome does materialize. This is a lay bet. Other betters can then buy these back/lay bets with 3:1 odds that the event materializes. Thus, every matched bet on the exchange involves both a lay bet and a back bet.

At a given point in time, a range of back/lay bets will be on sale on the Betfair platform, giving rise to a bid-ask spread (Figure 2). The going odds at every point in time is the odds of the latest bet traded. These bets can be converted into probabilities. If a bet is trading at odds of 1:3 this reflects a probability of 75 percent of this event occurring.2 Figure 3 shows the probability odds of a “Leave” vote over the sample period and Figure 4 the cumulative distribution of trades placed on the platform leading up to the referendum day.

Figure 2.
Figure 2.

Betting Market Structure on the Betfair Platform

Citation: IMF Working Papers 2021, 027; 10.5089/9781513568645.001.A001

Notes: The figure shows the prevailing uncleared Referendum bets on the Betfair platform at a given point in time.
Figure 3.
Figure 3.

Market Perceived Probability of a Leave Result

Citation: IMF Working Papers 2021, 027; 10.5089/9781513568645.001.A001

Notes: The figure shows the perceived probability of a Leave result in the referendum on United Kingdom’s membership of the European Union on June 23, 2016 based on betting data. Source: Author’s calculation using Betfair.com data.
Figure 4.
Figure 4.

Trade Volume and Cumulative Distribution

Citation: IMF Working Papers 2021, 027; 10.5089/9781513568645.001.A001

Notes: The figure shows the by-minute trade volume and the cumulative distribution of trades on the Betfair.com platform leading up to the referendum on June 23, 2016. Source: Author’s calculation using Betfair.com data.

III. The Data

Our betting data comes from Betfair, an online betting platform. The data is at the tick-level, meaning that each data point represents a trade. The implied probability that the referendum would result in a “Leave” vote is derived from the price of a trade in the Referendum Market. The betting odds associated with each trade are presented as decimals, which we convert into a probability.3 As trades are made in the market the odds change, and simultaneously so does the implied probability. Each new implied probability is time-stamped with precision on the second. When Betfair first started taking bets, the implied probability would move only by the minute or hour. The frequency of bets increased significantly in the lead up to the referendum, and by the end of our sample probabilities were updated up to every second. The online market was open for bets 24 hours per day, seven days per week.

Financial prices are collected from Bloomberg at minute intervals. We collect price series for exchange rates and equity indices:

  • The exchange rates for the British pound sterling (GBP) to Euro and GBP to Unites States Dollar (USD). The currencies are traded over the counter 24 hours day, Monday through Friday globally.

  • The Financial Times Stock Exchange 100 Index (FTSE 100) index and the FTSE All-Share index. The FTSE 100 index is comprised of the 100 companies listed on the London Stock Exchange (LSE) with the highest market capitalization, which is approximately 80 percent of the total market capitalization of the LSE. The FTSE All-Share, is a capitalization-weighted index, comprising of 613 companies (as per August 31, 2020) representing 98 percent of of the UK’s market capitalization. The LSE is open from 8am to 4:30pm Monday through Friday.

  • The FTSE Industry Classification Benchmark (ICB) 19 sub-indices. In aggregate, these 19 sectors make up the full market capitalization of the FTSE100 and FTSE All-Share (these include, for instance, sectors as broad as construction, banking, chemicals, household and personal goods, among others).

  • The major stock indices for other EU member countries. Stock markets across the EU open and close at various times throughout the day, but all markets typically stay open for about 8 hours.

All financial and betting data is harmonized to Greenwich Mean Time (GMT). The data covers the period from January 4th, 2016 to June 24th, 2016. The referendum was held on June 23rd, and the results were released gradually on the evening of June 24th. Note that the period we consider in our baseline empirical model is shorter, beginning in April 23rd and ending on June 24th, which is to focus on the period when the betting market saw a greater volume of trades, more turnover, and increased volatility. The full sample period is used in the robustness tests which are presented in Section V.

Observations in the final data set are recorded at the minute level, so that the implied probabilities match the financial data. In the instances where multiple probabilities are recorded per minute, the last recorded probability is used. For example, if the last recorded probability between 12:15pm and 12:16pm occurred at 12:15:45pm (hh:mm:ss) at 33 percent, then this probability is assigned to 12:15pm. In instances where no probability was recorded for a minute(s), the last recorded probability is used. For example, 33 percent would be assigned to 12:16pm and 12:17pm if there was no change in probability until 12:18pm.

Our data also includes the timing of poll results leading up to the referendum. Details on 88 polls conducted by eleven companies and released from January until June were downloaded from Twitter. We describe in V.B. how variation just around the release of these poll results can be used as a robustness check of our main empirical methodology. Figure 7 indicates when these polls were released and how the implied probability of Leave responded.

Figure 5.
Figure 5.

Probability of Leave Result and GBP/EUR

Citation: IMF Working Papers 2021, 027; 10.5089/9781513568645.001.A001

Notes: The figure shows the high frequency (by minute) changes in the perceived probability of a Leave result in the referendum and the GBP/EUR exchange rate.Source: Author’s calculations using data from Betfair.com and Bloomberg.
Figure 6.
Figure 6.

Probability of Leave Result and FTSE100

Citation: IMF Working Papers 2021, 027; 10.5089/9781513568645.001.A001

Notes: The figure shows the high frequency (by minute) changes in the perceived probability of a Leave result in the referendum and the FTSE100 stock index.Source: Author’s calculations using data from Betfair.com and Bloomberg.
Figure 7.
Figure 7.

Poll Result Releases

Citation: IMF Working Papers 2021, 027; 10.5089/9781513568645.001.A001

Notes: The figure presents the perceived probability of a Leave result, and release date-times for 88 polls conducted about the public’s support/opposition for the Brexit Referendum leading up to the vote on June 23, 2016. Source: Author’s calculation using Betfair and Twitter data.

Finally, trade data are taken from the OECD’s Trade in Value Added (TiVA) database which provides bilateral sectoral-level trade data at an annual frequency. The sectoral classifications in TiVA are matched to the broad FTSE categories using detailed sectoral definitions.

For further details on the data we use, see Annex VII.C.

IV. Analysis

A. Impact of referendum expectations on nominal exchange rates

We begin by estimating the impact of changes in the expected probability of a “Leave” vote on the value of the British pound sterling (£, GBP). To do so, we regress the log of the nominal sterling exchange rate vis-à-vis the euro and the U.S. dollar, sht,£/l where l ∈ {$,€}, on the perceived probability of leave derived from betting odds, P(leave)ht, measuring financial market’s belief of the probability of leaving the EU:

Δ ⁢ ln ⁡ s h t , £ / l = α h + β Δ P ( l e a v e ) h t + e h t ( 1 )

where h denotes the hour and t the exact time and date (minute, day, month, and year), αh are hourly fixed effects that capture all other factors affecting the pound during a given hour, and eht is the residual term. We estimate the model in differences, expressed by the ∆ term, which we define as either 5 or 10 minute difference of the dependent and independent variables.4

Estimating equation (1) involves a choice of time unit. This needs to be sufficiently long for news related to the referendum to be passed on to betting markets, but sufficiently short to avoid that other news influence the exchange rate during the same time period. Below we look at two possibilities, 5 and 10 minute intervals. These intervals are short, yet they allow time for market adjustment to new information. In both cases, we also include time fixed effects to allow for other possible that also could affect the exchange rate. In order for the model to be identified the fixed effects must be at a lower frequency than our variables of interest–which are at the minute frequency. We choose one hour time fixed effects, which we believe provides a reasonable window for news to be transmitted to exchange rates. In robustness exercises, we show that time fixed effects at a lower frequency do not affect the results in a substantive way (Section V).

The results are reported in Table 1 for 5 minute difference increment and in Table 2 for 10 minute. In columns (1) and (2) of both tables we include as our sample the full period for which Betfair’s markets were taking bets on the probability of the referendum outcome, 00:00 (midnight) on 3 January 2016 until 16:29 on 24 June 2016. In columns (3) and (4) we restrict the sample period to start two months before the vote, at 00:00 23 April 2016. The shorter sample period gives us an indication of whether or not markets were more responsive to changes in the perceived probability of a leave vote as the date of the referendum approached. In addition, as Figure 5 shows, there was little variation in the perceived probability of leaving in the early part of the sample, making identification in the full sample possibly difficult.

Table 1.

Impact of Perceived Probability of Leave on Exchange Rate (5 Minutes)

article image
Notes: Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. Variables are differenced over 5 minute intervals.
Table 2.

Impact of Perceived Probability of “Leave” on Exchange Rates (10 Minutes)

article image
Notes: Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. Variables are differenced over 10 minute intervals.

As expected, we observe a larger impact on exchange rates in the period leading up to the referendum and we see that when we allow more time for information process there is a larger effect on the exchange rate (compare Table 1 to Table 2). The explanatory power of changes in the perceived probability of leave of exchange rates almost double with a longer time window (compare (1) to (3) and (2) to (4), respectively, in Table 1 and Table 2). While the impact of a one percentage point increase in the probability of leaving the EU lead to a 0.004–0.007 percent decline in the value of the pound sterling against the euro (column (1)) and a 0.006–0.100 percent decline against the U.S. dollar (column (2)) over the full sample, during the shorter period leading up to the vote the impact of changes in the perceived probability of leaving were about 0.002–0.003 percentage points higher against both.5 This suggests that while market participants negatively priced in the event of a Leave vote throughout the period leading up to the referendum, they became considerably more sensitive as the date approached.

B. Impact of the referendum on UK equity markets

Next, we examine the impact of the perceived probability of a “Leave” result on the Financial Times and Stock Exchange index (FTSE), FTSEAll,ht, which comprises the most highly capitalized companies in the UK listed on the London Stock Exchange. We also look separately at each of the sub-sectors of the index, FTSEi,ht . We estimate an analogous regression in 5-minute and 10-minute differences to (1) for these equities:

Δ ⁢ ln ⁡ F T S E i , h t = α h + β Δ P ( l e a v e ) i , h t + e i , h t ( 2 )

More specifically, we start with the entire FTSE market, FTSEAll,ht, which represents the performance of all eligible companies listed on the London Stock Exchange main market, which pass screening for size and liquidity, and captures 98% of the UK’s market capitalization.6 We then narrow the sample to the FTSE 100, FTSE100,ht, which consists of the 100 companies with the highest market capitalization listed on the London Stock Exchange, see Figure 6. Finally, we look separate at the FTSE’s specific sectors, FTSEi,t , which include: (i) real estate, (ii) industrial, (iii) utilities, (iv) banks, (v) oil and gas, (vi) insurance, (vii) technology, (viii) construction, (ix) housing, (x) media, (xi) finance, (xii) retail, (xiii) chemistry, (xiv) telecommunications, (xv) travel, (xvi) food and beverages, (xvii) health care, (xviii) basic resources, and (xix) automotive. The sector-level analysis allows us to investigate which sectors market participants expected to be most affected by losing free access to EU markets. For example, by breaking down equity markets in this way, our estimate will provide insight to whether UK sectors whose firms are more dependent on the EU market upstream (sectors whose customers are individuals or firms in the EU) or downstream (firms whose suppliers are in the EU) are expected to be most affected by leaving.

We will also be able to test whether those sectors that were more strictly regulated by EU law (for e.g. food and beverage, telecommunications) were more affected than those that fell more under domestic regulation (for e.g. construction and healthcare), since there was more uncertainty in EU regulated markets.

Our results are reported in Tables 3 and 4 for the 5 minute differencing specification and 10 minute differencing specification, respectively.7 Comparing Table 3 and 4 we see that the impact from changes in in the probability of a “Leave” result on equity markets in the UK is substantially larger for the specification with 10 minute differencing. This is likely because the longer time window allows markets to better process the information. For the total market the impact of a one percentage point increase in the “Leave” probability is -0.0039 percent for 5 minute differencing, and -0.0073 percent for 10 differencing. For the 100 largest firms the impact is -0.0045 and -0.0083 percent, respectively. This result is likely driven by an expected fall in future earnings resulting from a “Leave” result. On the other hand, some firms in the indices have substantial earnings from abroad, and will thus get an offset from the associated sterling depreciation (Section IV.A).

Table 3.

Impact of Perceived Probability of “Leave” on FSTE Sectors, (5 Minutes)

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. Variables are differenced over 5 minute intervals. All regressions include time fixed effects (hourly frequency).
Table 4.

Impact of Perceived Probability of “Leave” on FSTE Sectors, (10 Minutes)

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly frequency). Variables are differenced over 10 minute intervals.

The impact of the perceived probability of “Leave” is mixed across industry indices, with some indication that more EU-exposed industries take greater hit than domestic industries. For instance, the impact of chemicals, basic resources and construction are not statistically significant (in the 5 minute difference specification), while the impact on other sectors is highly significant and large. It is noteworthy that the results are not significant for either banks nor finance, which potentially could be explained by the intangible, and thus mobile, nature of capital in these sectors. The results could also be driven by expected regulatory changes following a “Leave” vote.

C. Impact of the referendum on European equity markets

The UK market is highly integrated with European markets through trade in goods and services and movement of people. As such, the impact of a “Leave” vote should not only affect UK listed companies but also those firms based in other EU countries who do business in the UK, with UK customers or UK based-businesses, or conduct transactions in pound sterling. In order to determine the extent to which businesses based in other EU countries were affected by the perceived probability of a “Leave” result, we estimate the country-specific impact of changes in the probability on other EU equity markets:8

Δ ⁢ ln ⁡ E q u i t y j , h t = α j , h + β Δ P ( l e a v e ) h t + e h t ( 3 )

where Equityj,t is country j’s aggregate equity market index. Results for the 5 and 10 minute differencing specifications are reported in Tables 5 and 6.9 Not surprisingly, the impact of changes in the perceived probability of a Leave result in the referendum is much smaller in foreign equity markets than in the UK market, particularly when we look at the 5 minute differencing specification. Yet, a number of countries did see their equity markets fall significantly in response to a one percentage point rise in the probability of a “Leave” result. In particular, the French and Irish equity markets, whose economies are two of the most highly integrated with the UK are estimated to have fallen by about 0.004 percent in response to a one percentage point increase in the perceived probability of a “Leave” result (allowing for the longer, 10 minute, information processing time). The Netherlands and Belgium are also estimated to have been significantly affected by changes in the perceived probability of “Leave”, which could be related to fears of disruption in import activities between the UK, Netherlands (Rotterdam, Amsterdam), and Belgian (Antwerp) ports, which are the some of the principal points of entry for foreign imports into the EU. 10 We investigate the role of trade exposures in the following section (IV.D).

Table 5.

Impact of Perceived Probability of “Leave” on EU Equities, (5 Minutes)

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly). Variables are differenced over 5 minute intervals.
Table 6.

Impact of Perceived Probability of “Leave” on EU Equities, (10 Minutes)

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly). Variables are differenced over 10 minute intervals.

D. Role of Trade Exposure

Does trade exposure explain the sensitivity of equity prices to changes in the probability of a “Leave” vote across UK industries (Section IV.B) and across other European equity markets (Section IV.C)? To investigate this connection further, we augment our model specification to include foreign trade exposure by industry (for the regressions using UK industry-level data) and trade exposure by EU country (for the regressions using EU country-level data).

We define industry-level trade exposure as the sum of industry i exports UK trading partner j (including all trading partners globally), over total UK exports in 2016:

E X i I n d u s t r y = Σ j E x p o r t s i j , 2016 Σ j Σ i E x p o r t s i j , 2016 ( 4 )

We define EU-country level trade exposure, EXjCountry in a similar fashion, summing the UK’s exports to country j across all industries in 2016:

E X j C o u n t r y = Σ i E x p o r t s i j , 2016 Σ j Σ i E x p o r t s i j , 2016 ( 5 )

We then, in turn, interact each of these trade exposure variables with our perceived probability of “Leave” variable. This allows us to include sector (or country) fixed effects, and thus control for time-invariant sector (or country) specific characteristics. Specifically, we estimate the following equations which correspond to our FTSE industry and EU baseline specifications, respectively:

Δ ln F T S E i , t = α i + α t + β 1 Δ P ( l e a v e ) t ⋅ E X i I n d u s t r y + β 2 Δ P ( l e a v e ) t + β 3 E X i I n d u s t r y + e t ( 6 )
Δ ln E q u i t y j , t = α j + α t + β 1 Δ P ( l e a v e ) t ⋅ E X j C o u n t r y + β 2 Δ P ( l e a v e ) t + β 3 E X j C o u n t r y + e t ( 7 )

Table 7 reports our estimation results. Columns 1–2 report the estimated equation (6) with 5 and 10 minute differencing, respectively, and columns 3–4 report the estimated equation (7). All columns includes fixed time and industry/country fixed effects, and robust standard errors.11 The results suggest that trade linkages are an important determinant of the extent to which asset prices were affected by changes in the perceived probability of a leave vote. Specifically, for each one percent increase in a sector’s share of total UK exports, the price of that sector’s equity index would decline by 0.01 percent more relative to other sectors. Our results are consistent with Hil et al. [2019] who find policy uncertainty has a differentiated impact on industries in the UK. We also find similar results at the EU country level. For each one percent greater an EU country’s trade was with the UK, they equity prices would decline by 0.02 percent more relative to other countries.

Table 7.

Estimated Impact of Perceived Probability of on UK and European Equities, Controlling for Trade

article image
Notes: Standard errors in parentheses, ∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. Standard errors robust. All regressions include time fixed effects. Exports are defined as total exports in sector (country) i (j) as a share of total global exports from the U.K. in 2016.

V. Robustness

A. Alternative fixed effects

The previous section provided evidence that foreign exchange and equity markets (in the UK and EU) were significantly affected by changes in the perceived probability of a “Leave” vote in the EU referendum, under the assumption that events triggering changes in the perceived probability were unique in the hour that they occurred and no other news during that hour would affect financial markets. We now test whether that is a reasonable assumption.

First we estimate the model with shorter time fixed effects, setting them to half hour intervals. That is, we shorten the time period during which we absorb any variation in the data that could affect financial markets. Results for the estimation of equations (1) – (3) are reported in Tables 12, 13, and 14. Results for the nominal exchange rate and UK equity regressions are highly robust to our baseline results, both in terms of the magnitude of the estimated coefficients and their statistical significance. The results on European equity markets are also broadly robust, except for on French equities which is no longer statistically significant. The robustness of these results suggests there was indeed little additional information during the second half of each hour on financial markets.

Next we extend the fixed effects to daily intervals. This is a less strict specification, which absorbs daily news and allows the coefficient estimate to capture more variation within days. The results, reported in Tables 15, 16, and 17, are highly robust for the exchange rate and UK equity market regressions. Results on European Equity markets tend to be more statistically significant, not surprisingly because we allow for more variation in our perceived probability of “Leave” variable. The results of both exercises broadly suggest our choice of hour-fixed effects in our baseline regressions are reasonable, but that the results are not sensitive to that choice.

B. Restricted sample

A final test of the degree to which we are capturing the correct news regarding the transmission of information about the probability of a “Leave” result rather than other news that may have moved markets in the same direction is to restrict our sample to those 5-minute intervals when news regarding polling results was released. We assume that it is during these intervals when individuals are more likely to place informed bets on the probability of a “Leave” result, and is a truer proxy for the actual probability of the UK voting in favor of leaving the EU. Wall et al. [2017] provides evidence for this assumption. They find that betting market prices were quite sensitive to key events leading up to the 2014 Scottish Independence Referendum, and were particular responsive to poll releases. Results are reported in Tables 18 – 20. We show that by restricting the sample to those short intervals where concrete news about the public’s preference for leaving the EU, in the form of poll results, the estimated impact of changes in the perceived probability of a “Leave” result had a much stronger impact on exchange rates. The impact on equities is more mixed, with only those sub-sectors that had the largest effect in our baseline regression that show a statistically significant effect in this restricted sample.

C. Non-linearities

Given the important differences between the result of a “Stay” vote in the referendum–that the UK would enter a regime of almost absolute certainty with regards to it’s relationship to the EU–and the result of a “Leave” vote–that the UK would enter a regime of high uncertainty– it’s possible that the impact of an increase in the perceived probability of a “Leave” result has a significantly different effect on markets than the impact of a decrease in the perceived probability of a “Leave” result. We test whether this is the case and report the results for our exchange rate model specification in Tables 21, for the FTSE index in Tables 22 and 23, and for European equities in Tables 24 and 25. The impact of a rise in the probability of a leave vote appears to be much more important for the UK than the impact of a fall in the probability– given the certainty surrounding the effects of staying relative to leaving the EU, it’s natural that the impact on the exchange rate would be greater. Yet the asymmetries are much less apparent in equity markets.

D. Effects beyond immediate impact

So far, we have been investigating the immediate impact of changes in the perceived referendum result. How does the impact change over time? To investigate this, we re-run the regressions for equity (1) and exchange rates (2) while we increase the time window over which the estimated impact on exchange rates and equity prices (the left hand side variable) is calculated. The change in probability (on the right hand side) remains calculated over 10 minutes. The results from this exercise is show in Table 26 for exchange rates and Table 27 for equity prices. For exchange rates, Table 26 shows that the estimated impact is persistent up to 15 minutes after the the change in probability. The estimated effect falls from 0.009 percent after 10 minutes to 0.008 after 15 minutes. After 15 minutes the estimated effect turns insignifi-cant.12 For equity prices, Table 27 the estimated effect is more persistent. The estimated impact increases from 0.008 percent after 10 minutes to 0.013 percent after 30 minutes. Even after 1440 minutes (24 hours) the estimated effect is 0.011 percent and significant.

VI. Conclusion

In this paper we study how expectations of the outcome of the UK’s referendum on EU membership in 2016 affected exchange rates and equity prices in the UK and EU, and in doing so provide empirical support for the leading theoretical predictions of the impact of political uncertainty on asset prices (see Pastor and Veronesi [2012]). Since equity prices are a reflection of expected future profits of cash flows from firms, the impact should be heterogeneous across sectors based on their exposure to EU-based regulations, to the EU-single market, to foreign competition, and to trade. We use high frequency betting data to assess the revealed expectations of market participants, and show that expectations did indeed affect asset prices. In the very short run (within 5 minutes), we find that an increase in probability of a “Leave” result of one percentage points caused (i) British stocks (as measured by FTSE all) to decline by 0.004 percent–the equivalent of a GBP 1 billion loss for a ten percentage point increase in the probability of “Leave”–and (ii) the Pound to depreciate by 0.006 percent against the Euro. We find negative and significant effects for most sub-sectors in the FTSE aindex, and more so the more exposed each industry is to international markets. We also find evidence of negative spill-overs to other EU member countries, and more so the more exposed each country is to the UK market.

VII. Annexes

A. Equity regressions, full sample period

Table 8.

Impact of Perceived Probability of “Leave” on FSTE Sectors, (5 Minutes, Full Sample)

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 03 January 2016 until 16:29 24 June 2016. Variables are differenced over 5 minute intervals. All regressions include time fixed effects (hourly frequency).
Table 9.

Impact of Perceived Probability of “Leave” on FSTE Sectors, (10 Minutes, Full Sample)

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 03 January 2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly frequency). Variables are differenced over 10 minute intervals.
Table 10.

Impact of Perceived Probability of “Leave” on EU Equities, (5 Minutes, Full Sample)

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 03 January 2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly). Variables are differenced over 5 minute intervals.
Table 11.

Impact of Perceived Probability of “Leave” on EU Equities, (10 Minutes, Full Sample)

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 03 January 2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly). Variables are differenced over 10 minute intervals.

B. Robustness tables

Table 12.

Impact of Perceived Probability of “Leave” on Exchange Rate, half hour FE

article image
Notes: Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. All regressions include time fixed effects (half hour frequency). Variables are differenced over 5 minute intervals.
Table 13.

Impact of Perceived Probability of “Leave” on FSTE Sectors, Half Hour FE

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. Variables are differenced over 5 minute intervals. All regressions include time fixed effects (half hour frequency).
Table 14.

Impact of Perceived Probability of “Leave” on EU Equities, Half Hour FE

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. All regressions include time fixed effects (half hour frequency). Variables are differenced over 5 minute intervals.
Table 15.

Impact of Perceived Probability of “Leave” on Exchange Rate, Daily FE

article image
Notes: Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. All regressions include time fixed effects (daily frequency). Variables are differenced over 5 minute intervals.
Table 16.

Impact of Perceived Probability of “Leave” on FSTE Sectors, Daily FE

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016.Variables are differenced over 5 minute intervals. All regressions include time fixed effects (daily frequency).
Table 17.

Impact of Perceived Probability of “Leave” on EU Equities, Daily FE

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. Variables are differenced over 5 minute intervals.
Table 18.

Impact of Percevied Probability of “Leave” on Exchange Rates, Restricted Sample

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. Sample period start data is 23 April 2016. All regressions include time fixed effects (hourly frequency).
Table 19.

Estimated Impact of Perceived Probability of Leave on FSTE, by Sector, Restricted Sample

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23apr2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly frequency).
Table 20.

Impact of Perceived Probability of “Leave” on EU Equities, Restricted Sample

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly frequency).
Table 21.

Impact of Percevied Probability of “Leave” on Exchange Rates, Asymmetries

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. P(L) is shorthand for P(Leave). Sample period state date is 23 April 2016. All regressions include time fixed effects (hourly frequency).
Table 22.

Impact of Perceived Probability of “Leave” on FSTE Sectors, P(Leave)<0

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23apr2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly frequency).
Table 23.

Impact of Perceived Probability of “Leave” on FSTE Sectors, P(Leave)>0

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23apr2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly frequency).
Table 24.

Impact of Perceived Probability of “Leave” on EU Equities, P(leave)<0

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly frequency).
Table 25.

Impact of Perceived Probability of “Leave” on EU Equities, P(leave) >0

article image
Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. All regressions include time fixed effects (hourly frequency).
Table 26.

Effects beyond Immediate Impact of Perceived Probability of “Leave” on GBP/BC

article image
Notes: Robust standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. Variables are differenced over 10 minute intervals.
Table 27.

Effects beyond Immediate Impact of Perceived Probability of “Leave” on FTSE 100

article image
Robust standard errors in parentheses, ∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 28.

Impact of Perceived Probability of “Leave” on FTSE Index, Controlling for Trade Exposure

article image
Standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. DK is Driscoll-Kraay standard errors. Exports are defined as total exports in sector i as a share of total global exports from the U.K. in 2016.
Table 29.

Impact of Perceived Probability of “Leave” on EU Equities, Controlling for Trade Exposure

article image
Standard errors in parentheses, ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01. The sample period is from 00:00 23 April 2016 until 16:29 24 June 2016. DK is Driscoll-Kraay standard errors. Exports are defined as total exports to country j as a share of total global exports from the U.K. in 2016.

C. Data description

Brexit Probability Data:

Description: Tick-level data, meaning each data point represents a trade. The implied probability that the referendum would result in a “Leave” vote is derived from the price of a trade in the Referendum Market. The betting odds associated with each trade are presented as decimals, which we convert into a probability. Source: Betfair

Poll Results and Brexit Referendum District Level Results: Results of 88 polls about the public’s stance on the Brexit referendum, conducted by eleven companies and released from January until June.

Description:

Source: Twitter

Currencies:

Description: USD/Euro and GBP/Euro exchange rates.

Source: Bloomberg LLP

FTSE 100 Index:

Description: A share index of the 100 companies listed on the London Stock Exchange with the highest market capitalization.

Source: Bloomberg LLP

FTSEAll-Share Index:

Description: A capitalization-weighted index, that comprises of about 600 of more than 2,000 companies listed on the London Stock Exchange. It is aggregation of the FTSE 100, FTSE 250, and FTSE Small Cap Indexes, and represents 98–99 percent of UK market capitalization. Source: Bloomberg LLP

FTSE 350 Supersector Indices:

Description: Real-time industry sector indexes derived from companies in the FTSE 100 and FTSE 250 indexes. Companies are classified according to the Industry Classification Benchmark (ICB) classification taxonomy. It is used to segregate markets into sectors within the macroeconomy.

Source: Bloomberg LLP

European Union Stock Market Indices:

Description: The benchmark stock market indices for 26 of the 27 countries in the EU. Malta’s stock exchange is calculated on a daily basis, and therefore was not at a high enough frequency for our analysis.

Source: Bloomberg LLP

Bilateral trade data at sectoral level

Description: EU trade data are taken from the OECD’s Trade in Value Added (TiVA) database which provides bilateral sectoral-level trade data at an annual frequency.

Source: OECD’s Trade in Value Added (TiVA) database

Table 30.

FTSE 350 Supersector Indices

article image
Source: Bloomberg LLP and FTSE Russell
Table 31.

Euro Area Stock Market Indices

article image
Source: Bloomberg LLP and FTSE Russell

References

  • Alan J Auerbach and Kevin Hassett. Optimal long-run fiscal policy: Constraints, preferences and the resolution of uncertainty. Journal of Economic Dynamics and Control, 31(5):1451–1472, 2007.

    • Search Google Scholar
    • Export Citation

  • Scott R. Baker, Nicholas Bloom, and Steven J. Davis. Measuring economic policy uncertainty. The Quarterly Journal of Economics, 131, 2016.

    • Search Google Scholar
    • Export Citation

  • Marie-Claude Beaulieu, Jean-Claude Coasset, and Naceur Essaddam. The impact of political risk on the volatility of stock returns. Journal of International Business Studies, 36, 2005.

    • Search Google Scholar
    • Export Citation

  • Ansgar Belke, Irina Dubova, and Thomas Osowski. Policy uncertainty and international financial markets: The case of brexit. Applied Economics, pages 1–19, 2018.

    • Search Google Scholar
    • Export Citation

  • Nicholas Bloom. The impact of uncertainty shocks. Econometrica, 77(3):623–685, 2009.

  • Nicholas Bloom, Philip Bunn, Scarlet Chen, Paul Mizen, Pawel Smietanka, and Gregory Thwaites. The impact of brexit on uk firms. Stanford Institute for Economic Policy Research (SIEPR), 2019.

    • Search Google Scholar
    • Export Citation

  • Maria Boutchkova, Hitesh Doshi, Art Durnev, and Alexander Molchanov. Precarious politics and return volatility. The Review of Financial Studies, 25(4):1111–1154, 2012.

    • Search Google Scholar
    • Export Citation

  • Jonathan Brogaard and Andrew Detzel. The asset-pricing implications of government economic policy uncertainty. Management Science, 61(1):3–18, 2015.

    • Search Google Scholar
    • Export Citation

  • Guglielmo Maria Caporale, Luis Gil-Alana, and Tommaso Trani. Brexit and uncertainty in financial markets. International Journal of Financial Studies, 6(1):21, 2018.

    • Search Google Scholar
    • Export Citation

  • Nicholas Coleman, Friederike Niepmann, and Stefan Walz. What equity markets said about brexit-related costs to u.s. banks. IFDP Notes. Washington: Board of Governors of the Federal Reserve System, 2018.

    • Search Google Scholar
    • Export Citation

  • A. Ehler, M. Horn, and S. Wendt. Brexit: Short-term stock price effects and the impact of firm-level internationalization. Finance Research Letters, 22, 2017.

    • Search Google Scholar
    • Export Citation

  • David Fisman, Raymond J Fisman, Julia Galef, Rakesh Khurana, and Yongxiang Wang. Estimating the value of connections to vice-president cheney. The BE Journal of Economic Analysis & Policy, 13(3), 2012.

    • Search Google Scholar
    • Export Citation

  • Tarek Alexander Hassan, Laurence van Lent Stephan Hollander, and Ahmed Tahoun. The global impact of brexit uncertainty. NBER Working Paper No. 26609, 2019.

    • Search Google Scholar
    • Export Citation

  • Paula Hil, Adriana Korczak, and Piotr Korczak. Political uncertainty exposure of individual companies: The case of the brexit referendum. Journal of Banking and Finance, 100, 2019.

    • Search Google Scholar
    • Export Citation

  • M. Hohlmeier and C. Fahrholz. The impact of brexit on financial markets – taking stock. International Journal of Financial Studies, 6, 2018.

    • Search Google Scholar
    • Export Citation

  • Christian F. Horn, Michael Ohneberg, Bjoern S. Ivens, and Alexander Brem. Prediction markets – a literature review 2014 following tziralis and tatsiopoulos. Journal of Prediction Markets, 8, 2014.

    • Search Google Scholar
    • Export Citation

  • Seema Jayachandran. The jeffords effect. The Journal of Law and Economics, 49(2):397–425, 2006.

  • Bryan Kelly, L’uboš Pástor, and Pietro Veronesi. The price of political uncertainty: Theory and evidence from the option market. The Journal of Finance, 71(5):2417–2480, 2016.

    • Search Google Scholar
    • Export Citation

  • P. Kurecic and F. Kokotovic. Empirical analysis of the impact of brexit referendum and post-referendum events on selected stock exchange indexes. South East European Journal of Economics and Business, 13:7–16, 2018.

    • Search Google Scholar
    • Export Citation

  • Anna Lapinska and Musa Orak. Real effects of uncertainty: Evidence from brexit. 2020. FEDS Notes. Washington: Board of Governors of the Federal Reserve System.

    • Search Google Scholar
    • Export Citation

  • Lubos Pastor and Pietro Veronesi. Uncertainty about government policy and stock prices. The Journal of Finance, 67(4):1219–1264, 2012.

    • Search Google Scholar
    • Export Citation

  • Lubos Pastor and Pietro Veronesi. Political uncertainty and risk premia. Journal of financial Economics, 110(3):520–545, 2013.

  • I. Quaye, Y. Mu, B. Abudu, and R. Agyare. Review of stock markets reaction to new events: Evidence from brexit. Journal of Financial Risk Management, 5:281–314, 2016.

    • Search Google Scholar
    • Export Citation

  • Charles P. Ries, Marco Hafner, Troy D. Smith, Frances G. Burwell, Daniel Egel, Eugeniu Han, Martin Stepanek, and Howard J. Shatz. After brexit: Alternate forms of brexit and their implications for the united kingdom, the european union and the united states. RAND Corporation, 2017.

    • Search Google Scholar
    • Export Citation

  • Charles P. Ries, Marco Hafner, Clement Fays, and Erez Yerushalmi. The end of the beginning: Assessing the potential economic implications of prolonged uk–eu trade policy uncertainty. RAND Corporation, 2020.

    • Search Google Scholar
    • Export Citation

  • Matthew Wall, Rory Costello, and Stephen Lindsay. The miracle of the markets: Identifying key campaign events in scottish independence referendum using betting odds. Electoral Studies, 46, 2017.

    • Search Google Scholar
    • Export Citation

  • Justin Wolfers and Eric Zitzewitz. Prediction markets. Journal of Economic Perspectives, 18, 2004.

1

The authors would like to thank Oya Celasun, Francesco Grigoli, Gita Gopinath, Heikki Hatanpaa, Mohamed Jaber, Divya Kirti, Antonis Kotidis, Malhar Nabar, Jan Moeller, Christian Mumssen, Jiri Podpiera, Hans Henrik Sievertsen, IMF colleagues especially those from the U.K. desk, seminar participants at the 2019 European Economic Association Meetings, the 2019 Association for Applied Econometrics Conference, and at the IMF for very helpful comments and suggestions. The authors also thank Betfair for providing data. All errors and misinterpretations are those of the authors alone.

2

This can be derived as the probability that would make a risk neutral person indifferent between placing both a bet or holding cash.

3

The formulation for conversion is (1/odds) *100 = implied probability.

4

Note that we multiple the dependent and independent variables by 100, so our coefficient of interest, β, is directly (approximately) interpreted as the percentage change in response to a one percentage point change in the probability of Brexit.

5

The magnitude of these coefficient can be compared to the average hourly standard deviation of changes in the sterling exchange rate, which was 0.036 vis-a-vis the euro and 0.04 against the dollar.

7

For the remainder of the paper we report results only for the sample period April 23, 2016 onwards, given the results from the exchange rate analysis suggests markets didn’t react to changes in the probability of Leave prior to that. Results for the full sample (from January 3, 2016) are provided in Appendix VII.A.

8

Note we do not include Malta, Slovakia and Luxembourg in our analysis because we do not have sufficient data on their equity market prices to identify the the model parameters.

9

Results for the full sample period are provided in Appendix VII.A.

10

The positive coefficients for Hungary and Bulgaria could be explained by pre-vote EU council agreement which allowed countries to limit in-work benefits for workers from other EU countries. Under the agreement EU countries would be allowed to lengthen the period required to qualify for in-work benefits up to 4 years for workers from other EU countries, and index child benefits to the cost of living conditions in workers’ home country. The implementation of this agreement was contingent on a “Stay” result in the UK referendum.

11

We also tried to estimate the same equations using Driscoll-Kraay standard errors and country fixed effects. This did not alter our conclusions (see Appendix Table 28–29).

12

It is however negative and significant after 1 hours. However, we do not attach much weight on this as the effect again is insignificant after 24 hours.

Same Series

Other IMF Content

Other Publishers

  • Collapse
  • Expand
Cover IMF Working Papers
The Impact of Political Uncertainty on Asset Prices: The Case of the United Kingdom's EU Membership Referendum
Author:
Mr. Niels-Jakob H Hansen
and
Ms. Margaux MacDonald
Volume/Issue:
Volume 2021: Issue 027
Publisher:
International Monetary Fund
ISBN:
9781513568645
ISSN:
1018-5941
Pages:
50
DOI:
https://doi.org/10.5089/9781513568645.001
Figure 1.

Policy uncertainty and the British EU Referendum in 2016
Figure 2.

Betting Market Structure on the Betfair Platform
Figure 3.

Market Perceived Probability of a Leave Result
Figure 4.

Trade Volume and Cumulative Distribution
Figure 5.

Probability of Leave Result and GBP/EUR
Figure 6.

Probability of Leave Result and FTSE100
Figure 7.

Poll Result Releases