=Paper= {{Paper |id=Vol-1774/MIDAS2016_paper8 |storemode=property |title=Brexit or Bremain? Evidence from bubble analysis |pdfUrl=https://ceur-ws.org/Vol-1774/MIDAS2016_paper8.pdf |volume=Vol-1774 |authors=Marco Bianchetti,Davide Emilio Galli,Camilla Ricci,Angelo Salvatori,Marco Scaringi |dblpUrl=https://dblp.org/rec/conf/pkdd/BianchettiGRSS16 }} ==Brexit or Bremain? Evidence from bubble analysis== https://ceur-ws.org/Vol-1774/MIDAS2016_paper8.pdf

Brexit or Bremain ?
Evidence from bubble analysis

Marco Bianchetti, Intesa Sanpaolo, Financial and Market Risk Management,
and University of Bologna
Davide Emilio Galli, Università degli Studi di Milano, Physics Dept.
Camilla Ricci, Intesa Sanpaolo, Financial and Market Risk Management
Angelo Salvatori, Università degli Studi di Milano, Physics Dept.
Marco Scaringi, Università degli Studi di Milano, Physics Dept.

Abstract. We applied the Johansen-Ledoit-Sornette (JLS) model to detect
possible bubbles and crashes related to the Brexit/Bremain referendum
scheduled for 23rd June 2016. Our implementation includes an enhanced model
calibration using Genetic Algorithms. We selected a few historical financial
series sensitive to the Brexit/Bremain scenario, representative of multiple asset
classes.
We found that equity and currency asset classes show no bubble signals, while
rates, credit and real estate show super-exponential behaviour and instabilities
typical of bubble regime. Our study suggests that, under the JLS model, equity
and currency markets do not expect crashes or sharp rises following the
referendum results. Instead, rates and credit markets consider the referendum a
risky event, expecting either a Bremain scenario or a Brexit scenario
edulcorated by central banks intervention. In the case of real estate, a crash is
expected, but its relationship with the referendum results is unclear.
1 Brexit or Bremain ?

On Dec. 17, 2015 the UK Parliament approved the European Union Referendum Act
2015 to hold a referendum on whether the United Kingdom should remain a member
of the European Union (EU). The referendum will be held1 on Jun. 23, 2016, with the
following Q&A:

• Q: ”Should the United Kingdom remain a member of the European Union or leave
the European Union?
─ A1: “Remain a member of the European Union”
─ A2: “Leave the European Union”

The two scenarios above were called “Bremain” and “Brexit”, respectively. In case of
Brexit decision, there is no immediate withdrawal. Instead, a negotiation period
begins to establish the future relationship between UK and EU. The negotiation length
is two years, extendible upon agreement between the two parties. For example, the
agreements between EU and Switzerland took 10 years of negotiations.
Referendum campaigning has been suspended on 16th June 2016 following the
shooting of Labour MP Jo Cox. This event has had a strong impact on the public
opinion, rapidly changing the opinion polls and possibly the attitude of the country.

Forecasting the results of the 23rd June 2016 referendum, given the apparent parity
between Bremain and Brexit supporters and the high percentage of undecided voters
observed until the week before, is clearly a very challenging task, with a high error
probability. Nevertheless, there exist at least three sources of data supporting forecast
analysis: opinion polls [8], bookmakers betting odds [9], and market data [10].
In this paper we recur to a different approach, looking for possible bubble signals
in historical series of financial data, and interpreting them in terms of Brexit or
Bremain scenarios. We stress that such approach does not attempt to predict directly
Brexit or Bremain events, but rather looks for information on belief and expectations
of market participants about them.

2 Methodology

We applied a forecasting methodology based on the Johansen-Ledoit-Sornette
(JLS) model, developed since the 90s at ETHZ by D. Sornette and co-authors (see e.g.
[1]-[4] and refs. therein). The JLS model has been extensively applied to bubbles,
crashes and crisis analysis in many fields. For applications in finance see e.g. the
Financial Crisis Observatory [5].

We stress that this paper was delivered before the UK referendum scheduled
for 23rd June 2016.
The JLS model assumes that, during a bubble regime, the asset mean value follows
the so-called Log-Periodic Power Law (LPPL) function,

𝐿𝑃𝑃𝐿 𝑡 = 𝐴 + 𝐵 𝑡! − 𝑡 ! + 𝐶 𝑡! − 𝑡 ! 𝑐𝑜𝑠 𝜔𝑙𝑜𝑔 𝑡! − 𝑡 + 𝜙 , (1)

𝐿𝑃𝑃𝐿 𝑡 = 𝑙𝑛 𝐸! 𝑝 𝑇 = 𝑙𝑛 𝑝 𝑡 ,

where 𝑝 𝑡 is the asset price and 𝐸! 𝑝 𝑇 denotes the conditional expectation of the
future value 𝑝 𝑇 at present time 𝑡 < 𝑇, given all information available up to time t.
In eq. (1) above, 𝐴 is the value 𝑙𝑛 𝑝 𝑡! at the critical time, 𝐵 < 0 is the increase in
𝑙𝑛 𝑝 𝑡 over the time unit before the crash if C were to be close to zero, 0 < 𝑚 < 1
should be positive to ensure a finite price at the critical time 𝑡! and lower than one to
quantify the super-exponential acceleration of price 𝑝 𝑡 , 𝐶 ≠ 0 is the proportional
magnitude of the oscillations around the exponential growth 𝜔 is the frequency of the
oscillations during the bubble, and finally 0 < 𝜙 < 2𝜋 is a phase factor. Note that the
seven JLS parameters 𝐴, 𝐵, 𝐶, 𝑚, 𝜔, 𝜙, 𝑡! are all free parameters that must be
calibrated to fit the asset’s historical series, without imposing a known critical time 𝑡! .
Extensive backtesting of the JLS model on past bubbles allowed to identify more
stringent parameters constraints, namely 0.1 < 𝑚 < 0.9, 6 < 𝜔 < 13, and |C| < 1 [3].

Overall, the JLS model describes the dynamics of a system with a growing
instability, generated by behaviors of investors and traders creating positive feedback
in the valuation of assets leading to unsustainable growth and culminating with a
finite-time singularity at some future critical time 𝑡! , which is interpreted as the
forecast of a possible crash. A voluminous literature has applied this model (and
slightly different versions) to various financial data, detecting many historical cases to
which the log-periodic apparatus could be applied. We refer the reader to [1]- [4] and
to references therein for more details.

3 Numerical solution

Our implementation of the JLS model is based on the original JLS version [1]-[4],
enhanced with robust global optimization methods, i.e. Genetic Algorithms, for model
calibration [6],[12]. The JLS model calibration requires the optimal fit of the
historical series with the LPPL function. The fit is optimal if the set
℘ = 𝐴, 𝐵, 𝐶, 𝑚, 𝜔, 𝜙, 𝑡! of LPPL parameters minimizes the root mean square error
between the historical series and the LPPL fit function,

𝑅𝑀𝑆 ℘ = 𝑝 𝑡! − 𝐿𝑃𝑃𝐿 𝑡! , ℘ !,

!!!
where 𝑡! ⋯ 𝑡! and 𝑝! ⋯ 𝑝! are the historical dates and prices, respectively. The
calibration problem above is computationally hard, since the oscillating term in the
LPPL function produces many local minima in the RMS error function, where a local
minimization algorithm gets trapped. This is the reason why different calibration
strategies have been proposed in the literature [3]. In particular, Sornette et al.
adopted a taboo search algorithm, based on multiple local optimizations, enhanced by
certain assumptions on the landscape of the RMS cost function.

Our global optimization approach is based on genetic algorithms, and attacks the
problem without any assumption on the shape of the LPPL hyper-surface. Our genetic
algorithm is based on the MatLab implementation2. We modified the uniform
crossover and gaussian mutation functions such that they are applied serially, giving
better performance. We also scaled the mutation intensity according to the behavior of
the optimization process, such that mutations are less important when the cost
function is decreasing and more important when no significant progress occurs.
We observed that this set-up allows a stable convergence to the global minimum,
since several runs of the same optimization problem lead to the same result. We were
able to successfully replicate the results by Sornette et al., and, in a few cases, we
were also able to find slightly better solutions.
However, such approach is much more computationally demanding, and required
appropriate parallel computing facilities [7]. In particular, it may be applied when just
a few historical series are examined, as in the present case.

We calibrated the JLS model as described above to the historical series described
in the next section. For each series, we run multiple model calibrations with different
calibration windows, and detected possible bubble signals, corresponding to possible
critical times 𝑡! . In particular, we used different window lengths, with final date equal
to the most recent data (17 June 2016), and initial date ranging between 12 February
and 1 April 2016, with one business day step. The candidate critical times 𝑡! were
accepted or rejected according to the constraint discussed above. This procedure
ensures the stability of the observed results.

4 Results

We selected a sample of financial data sensitive to the current Brexit/Bremain
scenario, representative of equity (BBRXEQT), currency (Gold, GBPUSD and
GBPEUR fx), rates and credit (FTSE ORB, GBP and EUR Libor – OIS basis), and
real estate (UK HPI) asset classes.
The data and the JLS model results are reported in the following Figures 1- 8. The
description of the market data are included in their corresponding captions. The
comments on the results and their interpretations are given below the figures. Each

2
Matlab release R2012a [6].
figure shows, on the left hands scale, the historical series (blue line), and one single
representative fit with LPPL function in eq. (1) (red line), chosen among the many
calibrations run with different calibration windows. The histogram reported on the
right hand scale counts the bubble signals (if any), coming from these calibrations. In
case of no bubble signals, no histogram appears.

The interpretation of the occurrence or not of the JLS bubble signal deserves some
attention. The theory behind the JLS model states that if investors in some asset
expect a future event (e.g. the UK Referendum) leading to a possible negative
scenario for that asset (e.g. Brexit), this may trigger an asset dynamics leading to a
bubble regime, possibly followed by a crash. Thus, reversing the argument, if one
detects bubble signals for an asset and knows how a future event will affect the asset
price, then one can state that the investors expect a negative scenario for that asset.
Translating into the Brexit context, if one detects bubble signals for an asset with a
critical time 𝑡! around June 23th, and knows that Brexit/Bremain are negative/positive
scenarios for that asset, respectively, one can conclude that investors are expecting
Brexit. The specular argument also holds: if one knows that Bremain/Brexit are
negative/positive scenarios for that asset, respectively, one can conclude that investors
are expecting Bremain.

Fig. 1. Brexit Equity Index (Bloomberg BBRXEQT Index), basket of 10 UK stocks designed
to reflect British exposure to the EU across different sectors. Data up to Friday 17th June 2016.

• Comments: the historical series shows a decreasing trend, but no super-exponential
behaviour and instabilities typical of bubble regime. In fact, the JLS model (LPPL
fit) does not propose valid bubble and crash signals.
• Interpretation: market participants are currently suspicious about UK stock market,
but do not actually fear either a crash following Brexit or a sharp rise following
Bremain.
Fig. 2. Gold prices (Bloomberg XAU BGN Crncy). Data up to Friday 17th June 2016.

• Comments: the historical series shows an increasing trend, but no super-
exponential behaviour and instabilities typical of bubble regime. In fact, the JLS
model (LPPL fit) does not propose valid bubble and crash signals.
• Interpretation: market participants are currently refuging into gold, but do actually
fear neither a sharp rise following Brexit nor a crash following Bremain. This
result is consistent with the BBRXEQT and GBPUSD FX rate observations.

Fig. 3. GBP/USD FX rate (Bloomberg GBPUSD BGN Crncy).
Data up to Friday 17th June 2016.

• Comments: the historical series shows an erratic trend, no super-exponential
behaviour and instabilities typical of bubble regime. In fact, the JLS model (LPPL
fit) does not propose valid bubble and crash signals.
• Interpretation: market participants but do not actually fear either a crash following
Brexit or a sharp rise following Bremain. This result is consistent with the
BBRXEQT and GBPUSD FX rate observations.

Fig. 4. GBP/EUR FX rate (Bloomberg GBPEUR BGN Crncy).
Data up to Friday 17th June 2016.

• Comments: as for GBP/USD
• Interpretation: as for GBP/USD.

Fig. 5. FTSE ORB Total Return GBP Index (Bloomberg TFTSEORB Index), includes GBP
fixed coupon Corporate bonds trading on LSE across different industry sectors and maturity
bands. Data up to Friday 17th June 2016.
• Comments: the historical series shows an upward trend (due to the overall lowering
discount rates, driven by lowering GBPLibor w.r.t. increasing GBP credit spreads)
and super-exponential growth and instabilities typical of bubble regime. In fact, the
JLS model (LPPL fit) propose several valid crash signals around 23th June.
• Interpretation: market participants consider the referendum a risky event for
corporate bonds, expecting either a Bremain scenario or the BoE intervention in
case of Brexit.

Fig. 6. GBPLibor3M vs GBP OIS 3M (Bloomberg BP003M Index – BPSWSC Crncy).
Measures the London interbank credit and liquidity risk on 3M time horizon relative to
overnight horizon. Data up to Friday 17th June 2016.

• Comments: the historical series shows super-exponential behavior and instabilities
typical of bubble regime. In fact, the JLS model (LPPL fit) does propose valid
bubble and crash signals around 24th June.
• Interpretation: market participants expect that the basis spread will crash back to
lower values, corresponding to lower credit and liquidity risk in the London
interbank market. This result is consistent with the FTSE ORB observations.
Fig. 7. Euribor3M vs EUR OIS 3M (Bloomberg EUR003M Index – EUSWEC Crncy).
Measures the EUR interbank credit and liquidity risk on 3M time horizon relative to overnight
horizon. Data up to Thursday 16th June 2016.

• Comments: the historical series shows a decreasing trend but no super-exponential
behaviour and instabilities typical of bubble regime. In fact, the JLS model (LPPL
fit) does not propose valid bubble and crash signals.
• Interpretation: market participants but do not actually fear either a crash
following Brexit, also because the expected ECB intervention, or a sharp rise
following Bremain.

Fig. 8. UK house price index [12]. Data since July 2008 up to April 2016
(this data is available with monthly frequency with 2 months delay).
• Comments: the historical series shows an increasing trend with super-exponential
behaviour and instabilities typical of bubble regime. In fact, the JLS model (LPPL
fit) does propose valid bubble and crash signals around June.
• Interpretation: the trend remembers those observed during the 2008 subprime
crisis. Market participants expect a crash, but its relationship with the referendum
is questionable, since the growth regime started before the current Brexit/Bremain
context, and more recent UK HPI data would be needed.

In the following table 1 we summarize the findings for each historical series.

Table 1. Summary of JLS bubble signals (col. 4) from Figure 1- Figure 8.

5 Conclusions

We applied a forecasting methodology based on the Johansen-Ledoit-Sornette (JLS)
model, developed since the 90s by D. Sornette at ETHZ and co-authors [1], and
extensively applied to detect bubbles, crashes and crisis in many fields [5]. Our
implementation includes an enhanced model calibration using robust global
optimization methods, i.e. Genetic Algorithms [6].
We applied the JLS model to a selection of historical financial series sensitive to
the current Brexit/Bremain scenario, representative of equity (BBRXEQT), currency
(Gold, GBPUSD and GBPEUR fx), rates and credit (FTSE ORB, GBP and EUR
Libor – OIS basis), and real estate (UK HPI) asset classes.
We found the following evidence (see Table 1):

• equity and currency asset classes show no bubble signals,
• rates, credit and real estate show super-exponential behaviour and instabilities
typical of bubble regime, with the exception of Euribor-EUR OIS basis.
Our study suggests that, under the JLS model, the following interpretations can be
drawn:

• equity and currency: market participants coherently do not expect crashes or sharp
rises following the referendum results.
• Rates and credit: market participants coherently consider the referendum a risky
event for the London market, expecting either a Bremain scenario or a Brexit
scenario edulcorated by central banks intervention.
• In the case of real estate, market participants expect a crash, but its relationship
with the referendum results is unclear.

6 References
1. A. Johansen, O. Ledoit, and D. Sornette, “Crashes as critical points”, International Journal
of Theoretical and Applied Finance, vol. 3, no. 02, pp. 219-255, 2000.
2. D. Sornette, “Dragon-kings, black swans and the prediction of crises”, Swiss Finance
Institute Research Paper, no. 09-36, 2009.
3. P. Geraskin, D. Fantazzini, “Everything You Always Wanted to Know about Log Periodic
Power Laws for Bubble Modelling but Were Afraid to Ask”, 1 Feb. 2011, SSRN working
paper http://ssrn.com/abstract=1752115.
4. D. Sornette, R. Woodard, W. Yana, W. Zhou, “Clarifications to questions and criticisms
on the Johansen–Ledoit–Sornette financial bubble model”, Physica A 392 (2013) 4417–
4428.
5. ETHZ Financial Crisis Observatory
https://www.ethz.ch/content/specialinterest/mtec/chair-of-
entrepreneurial-risks/en/financial-crisis-observatory.html
6. A. Salvatori, “Stochastic Models for Self-Organized Criticality in Financial Markets “,
M.Sc. Physics Thesis, Università degli Studi di Milano, Mar. 2016.
7. Parallel Computing and Condensed Matter Simulations Lab., Physics Department,
Univesità degli Studi di Milano.
8. Opinion polls: see e.g. Wikipedia, Financial Times, or Bloomberg.
9. Bookmakers betting odds: see e.g. Oddschecker
http://www.oddschecker.com/politics/british-politics/eu-
referendum/referendum-on-eu-membership-result
10. Bloomberg, Brexit watch indicators
http://www.bloomberg.com/graphics/2016-brexit-watch
11. UK house price index
https://www.gov.uk/government/organisations/land-registry
12. E. Jacobsson, “How to predict crashes in financial markets with the Log-Periodic Power
Law”

Disclaimer and acknowledgments.
The views and the opinions expressed in this document are those of the authors and
do not represent the opinions of their employers. They are not responsible for any use
that may be made of these contents. The opinions, forecasts or estimates included in
this document strictly refer to the document date, and there is no guarantee that future
results or events will be consistent with the present observations and considerations.
This document is written for informative purposes only; it is not intended to influence
any investment decisions or promote any product or service.
The authors gratefully acknowledge Luca Lopez for fruitful discussion and analysis at
the early stage of this work.