Examiners’ commentaries 2022

Examiners’ commentaries 2022

FN3142 Quantitative finance

Important note

This commentary reflects the examination and assessment arrangements for this course in the
academic year 2021–22. The format and structure of the examination may change in future years,
and any such changes will be publicised on the virtual learning environment (VLE).

Information about the subject guide and the Essential reading

references

Unless otherwise stated, all cross-references will be to the latest version of the subject guide (2015).
You should always attempt to use the most recent edition of any Essential reading textbook, even if
the commentary and/or online reading list and/or subject guide refer to an earlier edition. If
different editions of Essential reading are listed, please check the VLE for reading supplements – if
none are available, please use the contents list and index of the new edition to find the relevant
section.

General remarks

Learning outcomes

At the end of this course and having completed the Essential reading and activities, you should:

be able to demonstrate mastery of econometric techniques required in order to analyse

issues in asset pricing and market finance

be able to demonstrate familiarity with recent empirical findings based on financial

econometric models

understand and have gained valuable insights into the functioning of financial markets

understand some of the practical issues in the forecasting of key financial market variables,
such as asset prices, risk and dependence.

Aims of the course

This course is aimed at candidates interested in obtaining a thorough grounding in market finance
and related empirical methods. It introduces econometric techniques, such as time series analysis,
required to analyse empirical issues in finance. It provides applications in asset pricing, investment,
risk analysis and management, market microstructure and return forecasting. This course is
quantitative by nature. It aims, however, to investigate practical issues in the forecasting of key
financial market variables and makes use of a number of real-world datasets and examples.

1
FN3142 Quantitative finance

Reading advice

The subject guide is designed to complement, not replace, the listed readings for each chapter. Each
chapter in the subject guide builds on the earlier chapters, as is often the case with quantitative
subjects. Chapters should therefore be studied in the order in which they appear. Essential readings
for this course come from:

Christoffersen, P.F. Elements of Financial Risk Management. (Academic Press, Oxford,

2011) second edition [ISBN 9780123744487].
Diebold, F.X. Elements of Forecasting. (Thomson South–Western, Canada, 2006) fourth
edition [ISBN 9780324323597].

Further reading:

Bodie, Z., A. Kane and A.J. Marcus Investments. (McGraw–Hill Irwin, London, 2008)
eighth edition [ISBN 9780071278287] and (2013) [ISBN 9780077861674].
Brooks, C. Introductory Econometrics for Finance. (Cambridge University Press,
Cambridge, 2008) second edition [ISBN 9780521694681] and third edition [ISBN
9781107661455].
Campbell, J.Y., A.W. Lo and A.C. Mackinlay The Econometrics of Financial Markets.
(Princeton University Press, Princeton, NJ, 1997) [ISBN 9780691043012].
Clements, M.P. Evaluating Econometric Forecasts of Economic and Financial Variables.
(Palgrave Texts in Econometrics, England, 2005) [ISBN 9781403941572].
Elton, E.J., M.J. Gruber, S.J. Brown and W.N. Goetzmann Modern Portfolio Theory and
Investment Analysis. (John Wiley & Sons, New York, 2010) eighth edition [ISBN
9780470505847] and ninth edition [ISBN 9781118469941].
Granger, C.W.J. and A. Timmerman (2004) Efficient Market Hypothesis and Forecasting,
International Journal of Forecasting, 20(1).
Hull, J.C. Options, Futures and Other Derivatives. (Pearson, 2011) eighth edition [ISBN
9780273759072].
McDonald, R.L. Derivatives Markets. (Pearson, 2012) third edition [ISBN 9780321847829].
Taylor, S.J. Asset Price Dynamics, Volatility and Prediction. (Princeton University Press,
Oxford, 2007) [ISBN 9780691134796].
Tsay, R.S. Analysis of Financial Time Series. (John Wiley & Sons, New York, 2010) third
edition [ISBN 9780470414354].

Studying advice

In addition to reading, you are also expected to work through the learning activities and sample
examination questions provided in the subject guide. If you find it difficult to answer a given
learning activity, go through the readings to that learning activity again with a focus on resolving
the issues at stake. It is important to master the econometric techniques covered in the first part of
the course before moving onto more difficult topics.

Format of the examination

This year the format of the examination has been identical to last year’s and those of previous years,
i.e. candidates had to answer three out of four questions. The main difference compared to the
pre-2020 years, however, is that due to the Covid-19 situation, candidates completed the examination
online, over the course of 4 hours (there was an expected/recommended time of 3 hours).

2
 Examiners’ commentaries 2022

What the examiners are looking for

In a good answer to a quantitative question, you must provide rigorous derivations. Some
quantitative questions may furthermore ask for a numerical problem to be solved. With numerical
questions, it is important that answers and steps are carefully and clearly explained. Partial credit
cannot be awarded if the final numbers presented are wrong through errors of omission, calculation
etc., unless your workings are shown. In a good answer to a qualitative question, you are expected to
produce an answer which presents appropriate concepts and empirical evidence. You are
furthermore expected to get your points across in a direct, structured, and concise fashion.

Steps to improvement
You must study the subject guide thoroughly. If you manage to work your way through all
questions and examples, you will then properly understand the course material. This means
also attempting all questions from the self-study sections and from past examination papers
all by yourself without ‘cheating’ by looking at the provided answers. After you have
completed the questions, you should carefully compare your answer with the provided
answer.
Once you have completed the subject guide you should read more widely on each topic.
Wider reading gives you a stronger appreciation of theory and empirical evidence, and will
enable you to take a more critical, analytical approach to qualitative examination questions.
Some key further readings are listed above.
Consult the Examiners’ commentaries.
Practise questions from past papers. As the examination draws closer, practise under time
pressure, remembering that you probably only have approximately 60 minutes per question
in which to write your answer.

Examination revision strategy

Many candidates are disappointed to find that their examination performance is poorer than they
expected. This may be due to a number of reasons, but one particular failing is ‘question
spotting’, that is, confining your examination preparation to a few questions and/or topics which
have come up in past papers for the course. This can have serious consequences.

We recognise that candidates might not cover all topics in the syllabus in the same depth, but you
need to be aware that examiners are free to set questions on any aspect of the syllabus. This
means that you need to study enough of the syllabus to enable you to answer the required number of
examination questions.

The syllabus can be found in the Course information sheet available on the VLE. You should read
the syllabus carefully and ensure that you cover sufficient material in preparation for the
examination. Examiners will vary the topics and questions from year to year and may well set
questions that have not appeared in past papers. Examination papers may legitimately include
questions on any topic in the syllabus. So, although past papers can be helpful during your revision,
you cannot assume that topics or specific questions that have come up in past examinations will
occur again.

If you rely on a question-spotting strategy, it is likely you will find yourself in difficulties
when you sit the examination. We strongly advise you not to adopt this strategy.

3
FN3142 Quantitative finance

Examiners’ commentaries 2022

FN3142 Quantitative finance

Important note

This commentary reflects the examination and assessment arrangements for this course in the
academic year 2021–22. The format and structure of the examination may change in future years,
and any such changes will be publicised on the virtual learning environment (VLE).

Information about the subject guide and the Essential reading

references

Unless otherwise stated, all cross-references will be to the latest version of the subject guide (2015).
You should always attempt to use the most recent edition of any Essential reading textbook, even if
the commentary and/or online reading list and/or subject guide refer to an earlier edition. If
different editions of Essential reading are listed, please check the VLE for reading supplements – if
none are available, please use the contents list and index of the new edition to find the relevant
section.

Comments on specific questions

Candidates should answer THREE of the following FOUR questions.

Question 1

Consider a process Xt that resembles an AR(1) process except for a small twist:

Xt = (−1)t δ0 + δ1 Xt−1 + εt ,

where εt is a zero-mean white noise process with variance σ 2 and δ1 ∈ (−1, 1).

(a) Calculate the conditional and unconditional variances of Xt , that is, Vart−1 [Xt ]
and Var[Xt ].
(20 marks)
(b) Derive the autocovariance and autocorrelation functions of this process for all
lags as functions of the parameters δ0 and δ1 .
(30 marks)
(c) Explain what covariance stationarity means. Does Xt satisfy the requirements?
(15 marks)
(d) Calculate the 1- and 2-period ahead conditional and the unconditional
expectations of Xt , i.e. Et−1 [Xt ], Et−2 [Xt ], and E[Xt ]. Comment on your
findings.
(20 marks)

4
 Examiners’ commentaries 2022

(e) Consider now another process given by

Wt = δ0 + δ1 (−1)t Wt−1 + εt .

Can this process be covariance stationary? Explain.

(15 marks)

Reading for this question

Subject guide, Chapters 1–6.

Approaching the question

This is a standard exercise to test candidates’ understanding of basic concepts such as

expectations, covariances, variances, and basic properties like (covariance) stationary, and apply
them to time series. Candidates must demonstrate their ability to work with stochastic
processes, to identify stochastic and deterministic components and apply their properties, to
comment on autocovariance and autocorrelation, and how these refer to forecasting future
realisations of random variables and processes. Candidates should study the examples provided
in those chapters, especially in the end of chapters, in details, to fully master the techniques.

(a) We can write:

Vart−1 [Xt ] = Vart−1 [δ0 (−1)t + δ1 Xt−1 + εt ] = Vart−1 [εt ] = σ 2 .

Moreover:

Var[Xt+1 ] = Var[δ0 (−1)t+1 + δ1 Xt + εt+1 ]

= δ12 Var[Xt ] + Var[εt+1 ]

= δ12 Var[Xt ] + σ 2

so as long as we want to have a (co)variance stationary process, that is doable for any
δ1 ∈ (0, 1):
Var[Xt+1 ] = Var[Xt ] = δ12 Var[Xt ] + σ 2
implies:
1
Var[Xt ] = σ2 .
1 − δ12

(b) The autocovariance at lag 1 is given by::

γ1,t = Cov[Xt , Xt−1 ] = Cov[δ0 (−1)t + δ1 Xt−1 + εt , Xt−1 ]

= δ1 Var[Xt−1 ]
δ1
= σ2
1 − δ12

where we dropped the additive constants and used that εt is independent of everything that
happened before. For a lag of 2, we write:

γ2,t = Cov[Xt , Xt−2 ] = Cov[δ0 (−1)t + δ1 Xt−1 + εt , Xt−2 ]

= δ1 Cov[Xt−1 , Xt−2 ]
= δ1 γ1,t

δ12
= σ2 .
1 − δ12

5
FN3142 Quantitative finance

Similarly, we would have:

γk,t = Cov[Xt , Xt−k ] = Cov[δ0 (−1)t + δ1 Xt−1 + εt , Xt−k ]
= δ1 Cov[Xt−1 , Xt−k ]
= δ1 γk−1,t

δ1k
= σ2 .
1 − δ12

(c) A process is covariance stationary if (i) E[Xt ] does not depend on t; (ii) Var[Xt ] does not
depend on t; (iii) for all k and t Cov[Xt , Xt−k ] is finite and depends on k only. So far it
looks like this process is covariance stationary since Cov[Xt , Xt−k ] does not depend on t,
just on k, but we need to calculate the unconditional expectation, too, in part (d).
(d) For the expectation part, using E[εt ] = Et−1 [εt ] = 0, we can write:
(
t t −δ0 + δ1 Xt−1 if t is odd
Et−1 [Xt ] = Et−1 [(−1) δ0 + δ1 Xt−1 + εt ] = (−1) δ0 + δ1 Xt−1 =
δ0 + δ1 Xt−1 if t is even
while:
Et−2 [Xt ] = Et−2 [(−1)t δ0 + δ1 Xt−1 + εt ]

= (−1)t δ0 + δ1 Et−2 [Xt−1 ]

= (−1)t δ0 + δ1 ((−1)t−1 δ0 + δ1 Xt−2 )

(
−(1 − δ1 )δ0 + δ12 Xt−2 if t is odd
=
(1 − δ1 )δ0 + δ12 Xt−2 if t is even.
Finally:
E[Xt ] = E[(−1)t δ0 + δ1 Xt−1 + εt ]

= (−1)t δ0 + δ1 E[Xt−1 ]
(
−δ0 + δ1 E[Xt−1 ] if t is odd
=
δ0 + δ1 E[Xt−1 ] if t is even
which means that we cannot have E[Xt ] independent of t, unless δ0 = 0, and then E[Xt ] = 0.
Not required, but can be worth some extra marks if a candidate would not have maximum
at this point: Note that we could have a process in which E[Xt ] is the same for all odd t,
while taking a different but constant value for all even t. In this case, we could write:
E[X2t ] = δ0 + δ1 E[X2t−1 ] = δ0 − δ0 δ1 + δ12 E[X2t−2 ] = δ0 (1 − δ1 ) + δ12 E[X2t ]
i.e. we have:
δ0 (1 − δ1 ) 1
E[X2t ] = = δ0
1 − δ12 1 + δ1
while:
δ1 1
E[X2t−1 ] = −δ0 + δ0 = −δ0 .
1 + δ1 1 + δ1
(e) This process would actually violate the requirements on expectations and covariances, too,
as long as δ1 6= 0; for example:
E[Xt ] = E[δ0 + δ1 (−1)t Xt−1 + εt ]

= δ0 + δ1 (−1)t E[Xt−1 ]
(
δ0 − δ1 E[Xt−1 ] if t is odd
=
δ0 + δ1 E[Xt−1 ] if t is even.

6
 Examiners’ commentaries 2022

For these to be identical, we need δ1 = 0, which then implies E[Xt ] = δ0 .

For variance and covariance, we can write:

Var[Xt+1 ] = Var[δ0 + δ1 (−1)t+1 Xt + εt+1 ]

= δ12 Var[Xt ] + Var[εt+1 ]

= δ12 Var[Xt ] + σ 2

and:

γk,t = Cov[Xt , Xt−k ] = Cov[δ0 + δ1 (−1)t Xt−1 + εt , Xt−k ]

= δ1 (−1)t Cov[Xt−1 , Xt−k ]

= δ1 (−1)t γk−1,t .

While Var[Xt+1 ] is independent of time, for Cov[Xt , Xt−k ] to be independent of time we

would need δ1 = 0.

Question 2

(a) Define Value-at-Risk. What are its pros and cons relative to variance as a
measure of risk? Explain in detail.
(15 marks)
(b) Consider a portfolio consisting of a $50,000 position in asset K and a $150,000
position in asset L. Assume that returns on these two assets are i.i.d. Gaussian
with mean zero, that the daily volatilities of these two assets are 1% for asset K
and 2% for asset L, and that the coefficient of correlation between their returns
is 0.2.
i. What is the 10-day VaR at the 1% critical level for the portfolio?
ii. Compare your answer above to the 1% critical level VaRs that we would
have on investing in K and L assets separately. By how much does
diversification reduce the VaR?
(15 marks)
For parts (c) to (f ), consider a dummy variable vt that takes value 1 when a
daily loss is exceeding the VaR threshold on date t and 0 otherwise. Suppose
that you build such a variable for three VaR models that are constructed using
(i) a simple MA (moving average), (ii) an EWMA (a model called
‘exponentially weighted moving average’) and (iii) GARCH volatility estimates
and use the critical value 1%. Suppose further that you use this dummy
variable to run the following regressions:

vt = γ0 + εt

and obtain the (volatility-model dependent) estimates in the table below, with
standard errors in parentheses:
γ0
MA: 0.028514
(0.003336)
EWMA: 0.024899
(0.003123)
GARCH: 0.022088
(0.002946)
(c) Explain how the above regression outputs can be used to test the accuracy of
the VaR forecasts from these models (on their own).
(20 marks)

7
FN3142 Quantitative finance

(d) How do the empirical performances of the three methods for constructing the
VaRs compare? Explain.
(15 marks)
(e) Explain how you could compare the relative performances of the VaR forecasts.
(20 marks)

Suppose now that you use the dummy variable to run the following regressions:

vt = γ1 + γ2 vt−1 + εt

and obtain the (volatility-model dependent) estimates in the table below, with
standard errors in parentheses:

γ1 γ2
MA: 0.0272 0.0431
(0.0033) (0.0200)
EWMA: 0.0238 0.0406
(0.0031) (0.0201)
GARCH: 0.0217 0.0146
(0.0029) (0.0201)

(f ) Briefly explain how the above regression outputs can be used to evaluate the
accuracy of the VaR forecasts from these 3 models (on their own).
(15 marks)

Reading for this question

Subject guide, Chapters 13–14.

Approaching the question

A comprehensive question on Value-at-Risk: definition, both formal and informal; calculations

for individual assets vs. portfolios and different horizons; its relation to and combination with
volatility models, backtesting, and performance evaluation and portfolios, and the understanding
of risk measures. The calculations tended to be correct for the individual assets, but combining
them to portfolios turned out to be more challenging, suggesting that candidates are using ad
hoc techniques as opposed to doing it in a more structural way. Candidates tended to highlight
basic properties but there seemed to be a slight deficiency in understanding the limitations of the
measures and to critically think about the measures. The parts starting with (c) seemed to
surprise candidates by presenting a backtesting technique in a slightly different way compared to
what they saw in class, and in general they ended up not being able to show a good
understanding of what was going on. This might have been the reason why very few candidates
attempted this question and why those who did performed poorly. Candidates are encouraged to
spend more time on this material to understand the basic idea and techniques for evaluating and
comparing VaR forecasts, which constitute a very important aspect of risk management models.

(a) The ideal answer gives both the mathematical definition of VaR:

P r[rt+1 ≤ −VaRα
t+1 | Ft ] = α

and a verbal definition: The α% VaR is the cut-off such that there is an α% probability that
we will see a return as low or lower, where common choices for α are 1%, 5%, and 10%.
Pros: Variance as a measure of risk has the drawback that it ‘penalises’ (by taking a higher
value) large positive returns in the same way as large negative returns. As investors,
however, we would generally only consider ‘risk’ to be the possibility of the value of our
portfolio falling, not rising. VaR overcomes this by focusing on the lower tail of the
distribution of returns only. It is also easier to illustrate.

8
 Examiners’ commentaries 2022

Cons: it is less familiar to market participants, needs more information about the
distribution to calculate, is not subadditive, only provides a threshold for the worst α%
cases but does not take into account the shape of the distribution in that range (unlike ES).
Additional sensible pros and cons are also acceptable.
(b) We write the portfolio variance as:
 2  2
2 1 2 3 1 3
σΠ = × (0.01) + × (0.02)2 + 2 × 0.2 × × × 0.01 × 0.02 = 0.00024625
4 4 4 4
so the standard deviation is σΠ = 0.015692. From here, the 10-day 99% VaR is:
√
VaRΠ = 2.33 × 10 × 0.015692355 × 200,000 = $23,124.59.
When investing in the two assets separately, we would have:
√
VaRK = 2.33 × 10 × 0.01 × 50,000 = $3,684.05
and: √
VaRL = 2.33 × 10 × 0.02 × 150,000 = $22,104.32
so diversification reduces VaR by $2,663.78.
(c) The variable defined here is the ‘hit rate’, denoted by Hit t (α), and is defined by:
(
1 if rt ≤ −VaRt (α)
vt =
0 otherwise.

The hit function sequence, such as (0, 0, 1, 0, 0, . . . , 1), tallies the history of whether or not a
loss in excess of the reported VaR has been realised. Taking unconditional expectations of
the regression we see that γ0 is the violation ratio defined as:
No. of days for which rt ≤ −VaRt
Violation ratio =
No. of days in sample
which measures the full-sample average performance for each model.
If a model is optimal, it should yield VaR forecasts that are only violated α = 1% of the
time. These regressions simply estimate the mean of the ‘hits’ (since the regression only has
a constant in it). If the estimated parameter is far from α = 1% then we know that the VaR
forecast is not optimal.
For full marks one can add more about the test statistic.
(d) The formal hypothesis testing is described by:
H0 : E[vt ] = 0.05 vs. Ha : E[vt ] 6= 0.05.
If we test the 1% VaR models, the t statistic under the three models are:
γ0 t statistic (1%) t statistic (2%)
0.0285 − 0.01 0.0285 − 0.02
MA: 0.0285 = 5.61 = 2.57
0.0033 0.0033
(0.0033)
0.0249 − 0.01 0.0249 − 0.02
EWMA: 0.0249 = 4.81 = 1.58
0.0031 0.0031
(0.0031)
0.0221 − 0.01 0.0221 − 0.02
GARCH: 0.0221 = 4.17 = 0.72
0.0029 0.0029
(0.0029)
Hence, for all three methods we have empirical evidence against the optimality of that
particular VaR model: they generate too many hits. Our conclusion would be the same if
these were coming from 5% or 10% VaR models, except in these cases there would be too
few hits. Finally, if these numbers were coming from α = 2% models, then we would reject
the MA model, but we cannot reject the optimality of the EWMA and GARCH VaR
forecasts; whereas with α = 2.5% we would not be able to reject any of them.

9
FN3142 Quantitative finance

(e) The above tests of VaR forecast optimality may help eliminate some VaR forecasts as being
clearly sub-optimal. For regarding relative performances, just like with other types of
forecasts, we can employ a Diebold–Mariano (DM) test to compare VaR forecasts. One
needs to be careful with the loss function used – a good one is the ‘tick’ loss function,
defined as:     
d i,t = 2 α − 1
L rt , VaR rt + VaR
d i,t .
rt ≤−VaR
d i,t

(This formula is not needed for a complete answer.) To compare the forecast we can then
construct:    
d 1,t − L rt , VaR
dt = L rt , VaR d 2,t

and test:
H0 : E[dt ] = 0 vs. Ha : E[dt ] 6= 0
using a standard DM test. Clearly here we do not have enough information to do this test.

(f) Taking conditional expectations of the second regression model we have:

Et [vt+1 ] = γ1 + γ2 vt .

Comparing the point estimates we find a statistically positive γ2 = 0.04316 for the MA5 and
0.04061 for the EWMA model, while γ2 is not statistically different from zero for the
GARCH model. The interpretation for this finding is that given a violation today the
conditional expected violation rate tomorrow is:

Et [vt+1 ] = γ1 + γ2 × 1.

Notice that these results would suggest that if we conducted a conditional coverage backtest
proposed by Christoffersen (1998), we would reject that the hit variable is i.i.d. Bernoulli(α)
under the null hypothesis of no serial correlation, because a perfect VaR forecast under this
assumption generates a series of hits that has the following transition matrix:
 
1−α α
Π? = .
1−α α

Here, if our regression coefficient γ2 is significant, that means the probability of vt+1 = 1
does depend on the previous hit, and thus we would reject the null of hits being
conditionally independent. Pointing this out is definitely needed for full marks.

Question 3

(a) What is volatility clustering? Suggest tests for it. Explain your answers in
detail.
(20 marks)
(b) Explain Black’s observation about the ‘leverage effect’, i.e. the link between
stock returns and changes in volatility, and provide an explanation for this
effect.
(20 marks)
(c) Does a simple GARCH(1, 1) model capture the leverage effect? Explain.
(20 marks)
(d) Describe two GARCH-type models that account for the leverage effect in your
own words.
Note: For full marks, write down the processes with equations and explain
analytically how they work.
(20 marks)

10
 Examiners’ commentaries 2022

(e) Consider the volatility model called EWMA (‘exponentially weighted moving
average’) model of volatility:
2
σt2 = λσt−1 2
+ (1 − λ)rt−1 ,

which formally looks like a special case of the GARCH(1, 1) with setting ω = 0,
α = λ, and β = 1 − λ.
i. Show why this formula corresponds to weights assigned to the rt2 that
decrease exponentially as we move back through time (rt is the percentage
change in the market variable between day t − 1 and day t).
ii. What undesirable property does this model have compared to GARCH(1, 1)
based on the parameter values? Explain.
(20 marks)

Reading for this question

Subject guide, Chapters 8–9.

Approaching the question

A question on the most important properties of price/return volatility, namely volatility

clustering, asymmetry, stationarity, and how these properties can be modelled. There are
subquestions about technicalities and the understanding of how these models work, too. The
specific calculations are rather simple, following those in the subject guide, but candidates
tended not to pay attention to mathematical details, and thus ended up with lower scores than
they might have expected. The question also included short essay-like subparts, which asked
candidates to explain things using their own words – the average performance on these
subquestions was rather poor, suggesting that candidates did not have a good understanding of
the material. Finally, part (e) asked candidates to apply their knowledge in a specific volatility
model that the subject guide does not emphasise much – this again showed lack of deep
understanding and being surprised by unseen cases where one needs to apply their knowledge.

(a) Volatility clustering refers to the fact that while in general serial autocorrelation in returns
is not observed in financial data, a typical feature is serial autocorrelation in squared returns
– i.e. there is a substantial amount of predictability in return volatility: high squared
returns are generally suggesting high future squared returns (but not necessarily anything
about the level or returns), whereas low squared returns predict low future squared returns.
Plotting realised returns, volatility clustering refers to the fact that one can detect relatively
calmer periods and more volatile periods.
There are two simple tests to test for volatility clustering. McLeod and Li (1983) suggest
using the Ljung–Box test on the squared residuals (or squared returns, if they have
conditional mean zero) to test jointly for evidence of serial correlation.
An alternative test is the Lagrange multiplier test of Engle (1982), which involves running
the regression:
e2t = α0 + α1 e2t−1 + · · · + αL e2t−L + ut
and then performing a χ2L test of the hypothesis:

H0 : α1 = α2 = · · · = αL = 0

against the alternative that at least one α is different from zero.

One can also mention plotting ACF, squared returns to inspect volatility clustering.
(b) Black (1976) was perhaps the first to observe that stock returns are negatively correlated
with changes in volatility: that is, volatility tends to rise (or rise more) following bad news
(a negative return) and fall (or rise less) following good news (a positive return). This is
called the leverage effect, as firms’ use of leverage can provide an explanation for this
correlation: if a firm uses both debt and equity then as the stock price of the firm falls, the

11
FN3142 Quantitative finance

overall value of the firm also falls, but the debt value does not, so its debt-to-equity ratio
rises. This will raise equity return volatility if the firm’s cash flows are constant. The
leverage effect has since been shown to provide only a partial explanation to observed
correlation, but the name persists.
(c) The GARCH(1, 1) model is given by:
2
σt+1 = ω + βσt2 + αε2t .

This shows that tomorrow’s volatility is quadratic in today’s return, so the sign of today’s
return does not matter. However, the leverage effect implies that one might expect that
negative returns today lead to higher volatility tomorrow than do positive returns. Hence
the GARCH(1, 1) model cannot account for the empirical observation of leverage.
(d) One simple extension to accommodate this relation is the model of Glosten, Jagannathan
and Runkle (1993) (so-called GJR–GARCH, sometimes known as Threshold–GARCH):
2
σt+1 = ω + βσt2 + αε2t + δε2t 1{εt < 0}.

If δ > 0 then the impact of today’s return on tomorrow’s volatility is greater if today’s
return is negative compared to when today’s return is positive.
The other widely-used GARCH model that allows for leverage effects is the exponential
GARCH, or EGARCH model of Nelson (1991). Nelson proposed this model to remedy two
shortcomings of the standard GARCH model. The first is its inability to capture the
leverage effect, and the second is that conditions have to be imposed on the parameters of
the GARCH model to ensure a positive volatility estimate. The EGARCH model deals with
both:
2 εt εt
ln σt+1 = ω + β ln σt2 + α +γ .
σt σt
If γ < 0, negative shocks will increase the volatility more than positive shocks.
(e) The EWMA (‘exponentially weighted moving average’) model of volatility is defined as:

σt2 = λσt−1
2 2
+ (1 − λ)rt−1 .

i. One can simply iterate the equation as:

σt2 = λσt−1
2 2
+ (1 − λ)rt−1
2 2 2
= λ[λσt−2 + (1 − λ)rt−2 ] + (1 − λ)rt−1

= λ2 σt−2
2 2
+ λ(1 − λ)rt−2 2
+ (1 − λ)rt−1

= λ2 [λσt−3
2 2
+ (1 − λ)rt−3 2
] + λ(1 − λ)rt−2 2
+ (1 − λ)rt−1

= λ3 σt−3
2
+ λ2 (1 − λ)rt−3
2 2
+ λ(1 − λ)rt−2 2
+ (1 − λ)rt−1
..
.
k
X
= λk σt−k
2
+ (1 − λ) λj−1 rt−j
2

j=1

after k steps. Since 0 < λ < 1, the first term converges to zero when k → ∞ and thus:
∞
X
σt2 = (1 − λ) λj−1 rt−j
2

j=1

where the weights indeed decrease exponentially.

ii. The main issue is that due to λ + (1 − λ) = α + β = 1, the EWMA process does not have
a finite long-term mean to which the level of volatility mean-reverts, i.e. it is persistent,
unlike, for example, the GARCH model. Hence, it is also not covariance stationary.
Further, it does not allow for the flexibility regarding the choice of α and β – one
essentially pins down the other, too.

12
 Examiners’ commentaries 2022

Question 4

(a) What is the efficient market hypothesis statement according to Malkiel (1992)?
Explain in your own words.
(20 marks)
(b) Black (1986) gives an alternative definition of market efficiency. What is it and
why is Black’s definition difficult to test? Explain in your own words.
(20 marks)
(c) Suppose we are at time t, and we are interested in the efficiency of the market
of a given stock. Let Ωw
t denote the weak-form efficient markets information set
at time t, Ωss
t denote the semi strong-form efficient markets information set at
time t, and Ωst denote the strong-form efficient markets information set at time
t. To which information set, if any, do the following variables belong? Explain.
1. The nominal size of the short position Melvin Capital (a hedge fund)
currently has in a given stock.
2. The size of the long position in Gamestop shares purchased today by DeepF,
a user of the subreddit r/WallStreetBets.
3. The current 3-month US Treasury bill rate.
4. The value of the stock at time t + 2.
(20 marks)
(d) Explain how ‘unit root’ models and models with time trends look like. What
are the similarities and differences between these models?
(20 marks)
(e) Explain a way how one can test for unit root.
(20 marks)

Reading for this question

Subject guide, Chapters 7 and 18.

Approaching the question

A question with two parts – one on the different definitions of market efficiency and the
classification of different types of information, and another on spurious regressions and persistent
time series, both in terms of understanding how these models work and technicalities in their
testing. Candidates performed better on the first part of the exercise – not surprisingly, as
market efficiency is one of the most emphasised components of the course material. Again, the
challenge in these largely essay-like questions was to explain things using one’s own words.
Therefore, it is advised that after reading the material of the subject guide and performing
exercises, candidates should pay attention to how they would demonstrate their understanding in
a structured and concise manner.

(a) One should start with Jensen’s definition of market efficiency, from 1978, which stated that
a market is efficient with respect to information set Ωt if it is impossible to make economic
profits by trading on the basis of information set Ωt .
This definition was refined by Malkiel (1992). He stated that a capital market is efficient if
it fully and correctly reflects all relevant information in determining securities prices.
Market efficiency is defined with respect to some information set Ωt – and securities prices
would be unaffected by revealing that information to all participants.
Overall, the three important components in defining market efficiency are:
1. The information set: what information are we considering?
2. The ability to exploit the information in a trading strategy: can we use the information
to form a successful trading strategy/investment rule?

13
FN3142 Quantitative finance

3. The performance measure for the trading strategy is economic profits: that is,
risk-adjusted (because investors are risk-averse) and net of transaction costs.

(b) A reasonable summary of the EMH is that it implies the absence of arbitrage opportunities
(given some information set). Black (1986) proposed that, in addition to the no-arbitrage
condition, a market efficiency definition should include some reference to possible deviations
of the market price from the fundamental value. He proposed that the price should be
within a factor of 2 (admittedly an arbitrary figure) of the intrinsic value of the assets. This
relates to asset price ‘bubbles’, where the price of an asset far exceeds the fundamental value
of the asset. (One example is the technology bubble of the late 1990s.) According to Black’s
definition a market may be efficient if asset price bubbles do not grow ‘too big’ or last ‘too
long’ before they are corrected. The problem with empirically testing this definition is that
the fundamental value of most assets is hard to measure and so any test would be a joint
test of Black’s market efficiency definition and the model used to determine the intrinsic
value of the assets. An absence of arbitrage opportunities does not rule out all forms of
predictability in asset returns.

(c) Roberts (1967) defined three types of information set available to investors:
1. Weak form efficiency: The information set includes only the history of the prices or
returns.
2. Semi-Strong form efficiency: The information set includes all information known to all
market participants (publicly available information).
3. Strong form efficiency: The information set includes all information known to any
market participant (private information).
Then the answers:
1. The nominal size of the short position Melvin Capital (a hedge fund) currently has in a
given stock. Strong (private information).
2. The size of the long position in Gamestop shares purchased today by DeepF, a user of
the subreddit r/WallStreetBets. One can argue for semi-strong since it is non-price
public information (group members used to post them in the forum), strong otherwise.
3. The current 3–month US Treasury bill rate. Weak (= bond price).
4. The value of the stock at time t + 2. Nothing – would be weak in 2 day’s time.

(d) The similarity is that they are both models to capture non-stationarity in economic and
financial data: models with a time trend, and random walk models, respectively, can be
written as:

Yt+1 = β0 + β1 t + εt+1 , εt+1 ∼ WN (0, σ 2 )

Yt+1 = φ0 + Yt + t+1 , t+1 ∼ WN (0, σ 2 ).

Difference: the forecasts from the random walk model (also known as a ‘unit root’ model)
just extrapolate from where the series is at the point the forecast is made – for example,
going up as 0 is positive, whereas the forecasts from the trend model assume that the series
will revert back to its time trend.
Also that the forecast intervals are different: the time trend model’s intervals are constant;
not a function of the forecast horizon. The intervals from the random walk model grow with
the square-root of the forecast horizon (since the variances grow linearly with the horizon).
Random walk models are far more commonly used in finance, and they are usually used on
log prices or exchange rates. This is so that the difference in the log variable is interpretable
as the (continuously-compounded) return on the asset.
The random walk model above is also known as an AR(1) model with a unit root. This is
because the AR(1) coefficient is equal to one. A unit root process is non-stationary, and
standard econometric methods often break down when applied to such data – in turn, there
is a large, advanced, econometrics literature on unit roots.

14
 Examiners’ commentaries 2022

(e) The simplest way of dealing with a unit root is to first-difference the series: ignoring the
constant, if:
Yt+1 = Yt + t+1 , t+1 ∼ WN (0, σ 2 )
then:
∆Yt+1 = Yt+1 − Yt = t+1
and so the first-differenced series is white noise.
A number of tests are available for testing for the presence of a ‘unit root’ in a time series.
Most of these test have a unit root as the null hypothesis, against an I(0) alternative. In
particular, we test:
H0 : Yt ∼ I(1) vs. Ha : Yt ∼ I(0).
If, for example, we think about an AR(1) model:

Yt+1 = φ0 + φ1 Yt + t+1 , t+1 ∼ WN (0, σ 2 )

then:

∆Yt+1 = φ0 + (φ1 − 1)Yt + t+1

= φ0 + ψYt + t+1

and a test of φ1 = 1 is equivalent to a test of ψ = 0:

H0 : ψ = 0 vs. Ha : ψ < 0.

This is a test that the series is at least I(1) against the alternative that it is I(0). One such
test is the Dickey–Fuller test. The test statistic is the usual t statistic on ψ:
b

ψb
t statistic = q
V [ψ]
b

but under the null hypothesis this does not have the Student’s t nor standard normal
distribution, as the variable on the right-hand side of the regression is a random walk.
Dickey and Fuller used simulations to determine the critical values for this test statistic.
When we include a constant term in the regression, the 95% critical value is −2.86,
compared with −1.64 critical value for a one-sided test based on the standard normal
distribution. If our test statistic is less than the critical value then we reject the null
hypothesis of a unit root in favour of the hypothesis that the series is I(0): If we fail to
reject the null then we conclude that the series is I(1).
The Dickey–Fuller test only considers an AR(1) process. If the series has non-zero
autocorrelations beyond lag one then the residuals from the regression will be
autocorrelated, and thus t+1 will not be white noise. We may overcome this problem by
using the Augmented Dickey–Fuller (ADF) test.

15