Chapter 10

Switching models

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 1

 Switching Models

• Motivation: Episodic nature of economic and financial variables. What

might cause these fundamental changes in behaviour?
- Wars
- Financial panics
- Significant changes in government policy
- Changes in market microstructure - e.g. big bang
- Changes in market sentiment
- Market rigidities

• Switches can be one-off single changes or occur frequently back and forth.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 2

 Switching Behaviour:
A Simple Example for One-off Changes
25

20

15

10

0
1 39 77 115 153 191 229 267 305 343 381 419 457 495 533 571 609 647 685 723 761 799 837 875 913 951 989

-5

-10

Dealing with switching variables -15

We could generalise ARMA models (again) to allow the series, yt to be drawn from
two or more different generating processes at different times. e.g.
yt = 1 + 1 yt-1 + u1t before observation 500 and
yt = 2 + 2 yt-1 + u2t after observation 500
‘Introductory Econometrics for Finance’ © Chris Brooks 2013 3
 How do we Decide where the Switch
or Switches take Place?

• It may be obvious from a plot or from knowledge of the history of the

series.

• It can be determined using a model.

• It may occur at fixed intervals as a result of seasonalities.

• A number of different approaches are available, and are described

below.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 4

 Seasonality in Financial Markets

• If we have quarterly or monthly or even daily data, these may have patterns in.

• Seasonal effects in financial markets have been widely observed and are often
termed “calendar anomalies”.

• Examples include day-of-the-week effects, open- or close-of-market effect,

January effects, or bank holiday effects.

• These result in statistically significantly different behaviour during some

seasons compared with others.

• Their existence is not necessarily inconsistent with the EMH.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 5

 Constructing Dummy Variables for Seasonality

• One way to cope with this is the inclusion of dummy variables- e.g. for
quarterly data, we could have 4 dummy variables:

D1t = 1 in Q1 and zero otherwise

D2t = 1 in Q2 and zero otherwise
D3t = 1 in Q3 and zero otherwise
D4t = 1 in Q4 and zero otherwise

• How many dummy variables do we need? We need one less than the
“seasonality” of the data. e.g. for quarterly series, consider what happens if we
use all 4 dummies

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 6

 Constructing Quarterly Dummy Variables

D1t D2t D3t D4t Sumt

1986Q1 1 0 0 0 1
Q2 0 1 0 0 1
Q3 0 0 1 0 1
Q4 0 0 0 1 1
1987Q1 1 0 0 0 1
Q2 0 1 0 0 1
Q3 0 0 1 0 1
etc.

• Problem of multicollinearity so (XX)-1 does not exist.

• Solution is to just use 3 dummy variables plus the constant or 4 dummies and no
constant.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 7

 How Does the Dummy Variable Work?

• It works by changing the intercept. yt

Consider the following regression:

yt = 1 + 1D1t + 2D2t + 3D3t + 2x2t +... + ut
So we have as the constant
ˆ1  ˆ1 in the first quarter
ˆ1  ˆ 2 3
in the second quarter 1
ˆ1  ˆ 3 in the third quarter 2

̂ 1 in the fourth quarter

0

xt

Q3
Q2
Q1
‘Introductory Econometrics for Finance’ © Chris Brooks 2013 Q0 8
 Seasonalities in South East Asian
Stock Returns

• Brooks and Persand (2001) examine the evidence for a day-of-the-

week effect in five Southeast Asian stock markets: South Korea,
Malaysia, the Philippines, Taiwan and Thailand.
• The data, are on a daily close-to-close basis for all weekdays
(Mondays to Fridays) falling in the period 31 December 1989 to 19
January 1996 (a total of 1581 observations).
• They use daily dummy variables for the day of the week effects in the
regression:
rt = 1D1t + 2D2t + 3D3t + 4D4t + 5D5t + ut
• Then the coefficients can be interpreted as the average return on each
day of the week.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 9

 Values and Significances of Day of the Week Effects in
South East Asian Stock Markets

South Korea Thailand Malaysia Taiwan Philippines

Monday 0.49E-3 0.00322 0.00185 0.56E-3 0.00119
(0.6740) (3.9804)** (2.9304)** (0.4321) (1.4369)
Tuesday -0.45E-3 -0.00179 -0.00175 0.00104 -0.97E-4
(-0.3692) (-1.6834) (-2.1258)** (0.5955) (-0.0916)
Wednesday -0.37E-3 -0.00160 0.31E-3 -0.00264 -0.49E-3
-0.5005) (-1.5912) (0.4786) (-2.107)** (-0.5637)
Thursday 0.40E-3 0.00100 0.00159 -0.00159 0.92E-3
(0.5468) (1.0379) (2.2886)** (-1.2724) (0.8908)
Friday -0.31E-3 0.52E-3 0.40E-4 0.43E-3 0.00151
(-0.3998) (0.5036) (0.0536) (0.3123) (1.7123)
Notes: Coefficients are given in each cell followed by t-ratios in parentheses; * and ** denote significance at the
5% and 1% levels respectively. Source: Brooks and Persand (2001).

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 10

 Slope Dummy Variables

• As well as or instead of intercept dummies, we could also use slope dummies:

• For example, this diagram depicts the use of one dummy – e.g., for bi-annual
(twice yearly) or open and close data. y t

• In the latter case, we could

y     x   D x   u
define Dt = 1 for open observations t t t t

and Dt=0 for close. y t     xt  u t

• Such dummies change the slope

but leave the intercept unchanged.
• We could use more slope dummies
or both intercept and slope dummies. x t

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 11

 Seasonalities in South East Asian
Stock Returns Revisited

• It is possible that the different returns on different days of the week

could be a result of different levels of risk on different days.
• To allow for this, Brooks and Persand re-estimate the model allowing
for different
5
betas on different days of the week using slope dummies:
rt = (  iDit + i DitRWMt) + ut
i 1

• where Dit is the ith dummy variable taking the value 1 for day t=i and
zero otherwise, and RWMt is the return on the world market index
• Now both risk and return are allowed to vary across the days of the
week.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 12

 Values and Significances of Day of the Week Effects in
South East Asian Stock Markets
allowing for Time-Varying risks
Thailand Malaysia Taiwan
Monday 0.00322 0.00185 0.544E-3
(3.3571)** (2.8025)** (0.3945)
Tuesday -0.00114 -0.00122 0.00140
(-1.1545) (-1.8172) (1.0163)
Wednesday -0.00164 0.25E-3 -0.00263
(-1.6926) (0.3711) (-1.9188)
Thursday 0.00104 0.00157 -0.00166
(1.0913) (2.3515)* (-1.2116)
Friday 0.31E-4 -0.3752 -0.13E-3
(0.03214) (-0.5680) (-0.0976)
Beta-Monday 0.3573 0.5494 0.6330
(2.1987)* (4.9284)** (2.7464)**
Beta-Tuesday 1.0254 0.9822 0.6572
(8.0035)** (11.2708)** (3.7078)**
Beta-Wednesday 0.6040 0.5753 0.3444
(3.7147)** (5.1870)** (1.4856)
Beta-Thursday 0.6662 0.8163 0.6055
(3.9313)** (6.9846)** (2.5146)*
Beta-Friday 0.9124 0.8059 1.0906
(5.8301)** (7.4493)** (4.9294)**
Notes: Coefficients are given in each cell followed by t-ratios in parentheses; * and ** denote significance at the
5% and 1% levels respectively. Source: Brooks and Persand (2001).

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 13

 Markov Switching Models

• Markov switching models are a generalisation of the simple dummy variables

approach described above.
• The universe of possible occurrences is split into m states of the world, called st,
i=1,...,m.
• Movements of the state variable between regimes are governed by a Markov
process.
• This Markov property can be expressed as
P[a<ytb  y1, y2, ..., yt-1] = P[a<ytb  yt-1]
• If a variable follows a Markov process, all we need to forecast the probability that it
will be in a given regime during the next period is the current period’s probability
and a transition probability matrix:
 P11 P12 ... P1m 
P P ... P 
P 21 22 2 m 
 ... ... ... ... 
where Pij is the probability of moving from regime i to regime j.  
P
 m1 P m2 ... Pmm 
‘Introductory Econometrics for Finance’ © Chris Brooks 2013 14
 Markov Switching Models – The Transition
Probabilities
• Markov switching models can be rather complex, but the simplest form is
known as “Hamilton’s Filter”.

• For example, suppose that m=2. The unobserved state variable, denoted zt,
evolves according to a Markov process with the following probabilities

Prob[zt = 1  zt-1 = 1] = p11

Prob[zt = 2  zt-1 = 1] = 1 - p11 Prob[zt = 2  zt-1 = 2] = p22 Prob[zt = 1  zt-1
= 2] = 1 – p22
where p11 and p22 denote the probability of being in regime one, given that
the system was in regime one during the previous period, and the
probability of being in regime two, given that the system was in regime
two during the previous period respectively.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 15

 Markov Switching Models (cont’d)

m
• It must be true that P
j 1
ij 1 i

• We then have a vector of current state probabilities, defined as  t   1  2 ...  m 

where i is the probability that we are currently in state i.
• Given t and P, we can forecast the probability that we will be in a given regime
next period: t+1 = tP
• The probabilities for S steps into the future will be given by: t+1 = tPs
• The Markov switching approach is useful when a series is thought to undergo
shifts from one type of behaviour to another and back again, but where the
“forcing variable” that causes the regime shifts is unobservable.
• The model’s parameters can be estimated by maximum likelihood (see Engel and
Hamilton, 1990).

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 16

 An Application of Markov Switching Models
to the Real Exchange Rate

• Purchasing power parity (PPP) theory suggests that the law of one price
should always apply in the long run such that, after converting it into a
common currency, the cost of a representative basket of goods and services
is the same wherever it is purchased.
• Under some assumptions, one implication of PPP is that the real exchange
rate (that is, the exchange rate divided by a general price index) should be
stationary.
• However, a number of studies have failed to reject the unit root null
hypothesis in real exchange rates, indicating evidence against PPP theory.
• It is widely known that the power of unit root tests is low in the presence
of structural breaks as the ADF test finds it difficult to distinguish between
a stationary process subject to structural breaks and a unit root process.
• In order to investigate this possibility, Bergman and Hansson (2005)
estimate a Markov switching model with an AR(1) structure for the real
exchange rate, which allows for multiple switches between two regimes.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 17

 An Application of Markov Switching Models
to the Real Exchange Rate

• The specification they use is y t   s  y t 1   t

t

where yt is the real exchange rate, st (t = 1,2) are the two states and t ~
N(0, 2).
• The state variable, st, is assumed to follow a standard 2-regime Markov
process.
• Quarterly observations from 1973Q2 to 1997Q4 (99 data points) are used
on the real exchange rate (in units of foreign currency per US dollar) for
the UK, France, Germany, Switzerland, Canada and Japan.
• The model is estimated using the first 72 observations (1973Q2 - 1990Q4)
with the remainder retained for out of sample forecast evaluation.
• The authors use 100 times the log of the real exchange rate, and this is
normalised to take a value of one for 1973Q2 for all countries.
• The Markov switching model estimates are obtained using maximum
likelihood estimation.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 18

 Results

Source: Bergman and Hansson (2005)

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 19

 Analysis of Results

• As the table shows, the model is able to separate the real exchange rates
into two distinct regimes for each series, with the intercept in regime one
(1) being positive for all countries except Japan (resulting from the
phenomenal strength of the yen over the sample period), corresponding to
a rise in the log of the number of units of the foreign currency per US
dollar, i.e. a depreciation of the domestic currency against the dollar.
 2, the intercept in regime 2, is negative for all countries, corresponding
to a domestic currency appreciation against the dollar.
• The probabilities of remaining within the same regime during the
following period (p11 and p22) are fairly low for the UK, France, Germany
and Switzerland, indicating fairly frequent switches from one regime to
another for those countries' currencies.
• Interestingly, after allowing for the switching intercepts across the
regimes, the AR(1) coefficient, , is a considerable distance below unity,
indicating that these real exchange rates are stationary.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 20

 So PPP Holds After All?

• Bergman and Hansson simulate data from the stationary Markov

switching AR(1) model with the estimated parameters but they assume
that the researcher conducts a standard ADF test on the artificial data.

• They find that for none of the cases can the unit root null hypothesis be
rejected, even though clearly this null is wrong as the simulated data are
stationary.

• It is concluded that a failure to account for time-varying intercepts (i.e.

structural breaks) in previous empirical studies on real exchange rates
could have been the reason for the finding that the series are unit root
processes when the financial theory had suggested that they should be
stationary.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 21

 Use of the Markov-Switching Model for Forecasting

• Finally, the authors employ their Markov switching AR(1) model for
forecasting the remainder of the exchange rates in the sample in
comparison with the predictions produced by a random walk and by a
Markov switching model with a random walk.

• They find that for all six series, and for forecast horizons up to 4 steps
(quarters) ahead, their Markov switching AR model produces predictions
with the lowest mean squared errors; these improvements over the pure
random walk are statistically significant.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 22

 An Application of Markov Switching Models
to the Gilt-Equity Yield Ratio

• The gilt-equity yield ratio (GEYR) is defined as the ratio of the income
yield on long-term government bonds to the dividend yield on equities.
• It has been suggested that the current value of the GEYR might be a
useful tool for investment managers or market analysts
• The GEYR is assumed to have a long-run equilibrium level, deviations
from which are taken to signal that equity prices are at an unsustainable
level.
• Thus, in its crudest form, an equity trading rule based on the GEYR
would say, “if the GEYR is low, buy equities; if the GEYR is high, sell
equities.”
• Brooks and Persand (2001) employ monthly stock index dividend yields
and income yields on government bonds covering the period January
1975 until August 1997 (272 observations) for three countries - the UK,
the US, and Germany.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 23

 GEYR

0.5
1.5
2.5
3.5

0
1
2
3
No
v-
No 75
v-
No 76
v-
No 77
v-
No 78
v-
No 79
v-
No 80
v-
No 81
v-
No 82
v-
No 83
v-
No 84
v-
No 85
v-
US GEYR No 86

Date
v-

0.5
1.5
2.5
3.5
4.5

0
1
2
3
4
No 87
v-
No 88
v-
Nov-75 No 89
v-
Nov-76
The Value of UK GEYR 1975-97

No 90
v-
Nov-77 No 91
v-
Nov-78 No 92
v-
Nov-79 No 93
v-
Nov-80 No 94
v-
Nov-81 No 95
v-
Nov-82 96

Nov-83
Nov-84
Nov-85 GEYR

Date
0
1
2
3
4
5
6

Nov-86 No
v-
Nov-87 No 75
v-
Nov-88 No 76

‘Introductory Econometrics for Finance’ © Chris Brooks 2013

v-
Nov-89 No 77
The Value of US GEYR 1975-97

v-
Nov-90 No 78
v-
Nov-91 No 79
v-
Nov-92 No 80
v-
Nov-93 No 81
v-
No 82
Nov-94 v-
No 83
Nov-95 v-
No 84
Nov-96 v-
Time Series Plots of the GEYR

No 85
v-
No 86
Date

v-
No 87
v-
No 88
v-
No 89
v-
No 90
v-
The Value of German GEYR 1975-97

No 91
v-
No 92
v-
No 93
v-
No 94
v-
No 95
24

v-
96
 The Distribution of US GEYR

0.45

0.40

0.35

0.30

0.25

0.20

0.15

0.10

0.05

0.00
-4 -2 0 2 4

• The distribution looks as if it could usefully be spilt into two parts

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 25

 Estimated Parameters for Markov Switching Models

Statistic 1 2 12 22 p11 p22 N1 N2

UK 2.4293 2.0749 0.0624 0.0142 0.9547 0.9719 102 170
(0.0301) (0.0367) (0.0092) (0.0018) (0.0726) (0.0134)
US 2.4554 2.1218 0.0294 0.0395 0.9717 0.9823 100 172
(0.181) (0.0623) (0.0604) (0.0044) (0.0171) (0.0106)
Germany 3.0250 2.1563 0.5510 0.0125 0.9816 0.9328 200 72
(0.0544) (0.0154) (0.0569) (0.0020) (0.0107) (0.0323)

Notes: Standard errors in parentheses; N1 and N2 denote the number of observations

deemed to be in regimes 1 and 2 respectively.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 26

 Analysis of Results

• It is clear that the regime switching model has split the data into two distinct
samples - one with a high mean (of 2.43, 2.46 and 3.03 for the UK, US and
Germany, respectively) and one with a lower mean (of 2.07, 2.12, and 2.16).
• Also apparent is the fact that the UK and German GEYR are more variable at
times when it is in the high mean regime, evidenced by their higher variance
(in fact, it is around four and 20 times higher than for the low GEYR state,
respectively).
• The number of observations for which the probability that the GEYR is in the
high mean state exceeds 0.5 (and thus when the GEYR is actually deemed to
be in this state) is 102 for the UK (37.5% of the total), while the figures for the
US are 100 (36.8%) and for Germany 200 (73.5%).
• Thus, overall, the GEYR is more likely to be in the low mean regime for the
UK and US, while it is likely to be high in Germany.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 27

 Analysis of Results

• The table also shows the probability of staying in state 1 given that the GEYR
was in state 1 in the immediately preceding month, and the probability of
staying in state 2 given that the GEYR was in state 2 previously.

• The high values of these parameters indicates that the regimes are highly stable
with less than a 10% chance of moving from a low GEYR to a high GEYR
regime and vice versa for all three series.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 28

 Conclusions from the GEYR Application

 The Markov switching approach can be used to model the gilt-equity

yield ratio.

 The resulting model can be used to produce forecasts of the probability

that the GEYR will be in a particular regime.

 Before transactions costs, a trading rule derived from the model

produces a better performance than a buy-and-hold equities strategy, in
spite of inferior predictive accuracy as measured statistically.

 Net of transactions costs, rules based on the Markov switching model

are not able to beat a passive investment in the index for any of the
three countries studied.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 29

 Threshold Autoregressive (TAR) Models

• Intuition: a variable is specified to follow different autoregressive processes in

different regimes, with movements between regimes governed by an observed
variable.
1  1 y t 1  u1t if st  k  r
• The model is yt  
 2state
• But what is st-k? It is the
  2 y t 1  u 2t if s t  k  r
determining variable and it can be any variable which
is thought to make yt shift from one regime to another.
• If k = 0, it is the current value of the state-determining variable that influences the
regime that y is in at time t.
• The simplest case is where st-k = yt-k we then have a self-exciting TAR, or a
SETAR. The model is
1  1 y t 1  u1t if y t  k  r
yt  
• We could of course have more than one
2  lag  u 2t if y t  k  r
2 y t 1in each regime (and the number of lags
in each need not be the same).
• Under the TAR model approach, unlike the Markov switching model, the transitions
between regimes are discrete.

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 30

 Threshold Models: Estimation Issues

• Estimation of parameters in the context of threshold models is complex.

• Quantities to be determined include the number of regimes, the threshold variable, the
threshold variable lag, the model order in each regime, the value of the threshold, and
the coefficients for each regime.
• We cannot estimate all of these at the same time, so some are usually specified a priori
based on theory or intuition and the others estimated conditional upon them. E.g., set k
= 1, J = 2, r may not require estimation, etc.
• The lag length for each regime can be determined using an information criterion
conditional upon a specified threshold variable and fixed threshold value. For example
Tong (1990) proposes a modified version of AIC:

AIC ( p1 , p 2 )  T1 ln ˆ 12  T2 ln ˆ 22  2( p1  1)  2( p 2  1)
where T1 and T2 are the number of observations in regimes 1 and 2 respectively, p1 and
p2 are the lag lengths, and and are2 the residual variances.
• Estimation of the autoregressive coefficientsˆ 1 ˆ 2can then be achieved using nonlinear least
2

squares (NLS).

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 31

 An Example of a SETAR Model for the
French franc / German mark Exchange Rate
• From Chappell et al., 1996, Journal of Forecasting
• The study used daily data from 1/5/90 - 30 / 3/ 92.
• Both the FRF & DEM were then in the ERM which allowed for “managed floating”.
• Can use a SETAR to allow for different types of behaviour according to whether the
exchange rate is close to the ERM boundary. Currencies are allowed to move up to
2.25% either side of their central parity in the ERM.
• This would suggest the use of the 2-threshold (3-state) SETAR. This did not work as
the DEM was never a weak currency then.
• The model orders for each regime are determined using AIC
• Ceiling in the ERM corresponded to 5.8376 (log of FRF per 100 DEM).
• The first 450 observations are used for model estimation, with the remaining 50
being retained for out of sample forecasting.
• Forecasts are then produced using the threshold model, the SETAR model with 2
thresholds, a random walk and an AR(2)

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 32

 Estimated FRF-DEM Regime Switching Model
and Out of Sample Forecast Accuracies

M odel F o r R e g im e N um ber of
o b s e r v a t io n s
Eˆ t  0 . 0222  0 . 9962 E t1
E t-1 < 5 .8 3 0 6 344
( 0 . 0458 ) ( 0 . 0079 )
Eˆ t  0 . 3486  0 . 4394 E t  1  0 . 3057 E t  2  0 . 1951 E t  3 E t-1  5 . 8 3 0 6 103
( 0 . 2391 ) ( 0 . 0889 ) ( 0 . 1098 ) ( 0 . 0866 )
S o u r c e : C h a p p e ll, P a d m o r e ,a n d M is tr y ( 1 9 9 6 ) . R e p r in te d w ith p e r m is s io n o f J o h n W ile y a n d S o n s .
Panel A: Mean Squared Steps Ahead
Forecast Error
1 2 3 5 10
Random Walk 1.84E-07 3.49E-07 4.33E-07 8.03E-07 1.83E-06
AR(2) 3.96E-07 1.19E-06 2.33E-06 6.15E-06 2.19E-05
One threshold SETAR 1.80E-07 2.96E-07 3.63E-07 5.41E-07 5.34E-07
Two threshold SETAR 1.80E-07 2.96E-07 3.63E-07 5.74E-07 5.61E-07

Panel B: Median Squared

Forecast Error
Random Walk 7.80E-08 1.04E-07 2.21E-07 2.49E-07 1.00E-06
AR(2) 2.29E-07 9.00E-07 1.77E-06 5.34E-06 1.37E-05
One threshold SETAR 9.33E-08 1.22E-07 1.57E-07 2.42E-07 2.34E-07
Two threshold SETAR 1.02E-07 1.22E-07 1.87E-07 2.57E-07 2.45E-07
Source: Chappell, Padmore,and Mistry (1996). Reprinted with permission of John Wiley and Sons.
‘Introductory Econometrics for Finance’ © Chris Brooks 2013 33