Òàäæèêóëîâà
Æ.
Ëåêòîð ÊÁÒÓ
Recent House Price Dynamics in
Kazakhstan
Over the past decades
the house price movements were among the most popular issues raised by
economists (the most humble list includes: Cutler, 1991, Diba and Grossman,
1988, Shiller 1978, Blanchard and Watson, 1982, Summers, 1986, Poterba, 1984,
Schaller and van Norden, 1997, Brock 1974, Samuleson, 1958, etc). It was majorly due to the long period of
increasing house prices (which were in their turn caused by low interest rates)
which made economists alarmed due to possibility that high house prices might
be overvalued and instead of reflecting market fundamentals are driven by
speculative players of the market. Households, who are the main players in
house market started building the belief that the prices will grow in the
future. This, as we now know, lead to the mortgage crisis of 2007. Since then,
the house prices issue became almost national one for the US economists.
Scientists from all over the world also started analysing their domestic house
price behaviours in an attempt to identify any negative and unexplainable
shifts in the prices for houses (Yu, 2010, Otto, 2007, Murphy, 2007, Stepanyan,
et al, 2010, etc). Among those
countries Kazakhstan was not an exception. Stepanyan et al in their study tried to identify house price determinants for
selected countries of the former Soviet Union.
They posit that along with GDP foreign remittances for domestic economy
play significant role in explaining house price movements. The study is based
on macroeconomic data. As long as I know this is the only study which is
related to house price movements in Kazakhstan and no study has been commenced
to understand determinants of house price movements using the two methods of
asset price evaluation that will be covered further in this work.
The main objective of
this research is to determine whether house prices in Kazakhstan during the
period of Jan 2007 and May 2010 deviated away substantially from their so
called “fundamentals”. My major aim is to ascertain that house price movements
are explained by market fundamentals and to find out whether non-fundamental
part of house prices is statistically significant in explaining house price
variations. Two different models of house price evaluation that are used in
this research will be compared for their goodness of fit and other statistical
parameters. Also variables of both models will be interchanged among each other
and different combinations of them will be used. All of the mentioned above
will be done to determine best model for house price evaluation in Kazakhstan.
I will also apply cointegration test in order to determine whether house prices
of the country move in accordance with fundamentals.
In order to achieve
these goals two different methods of calculating fundamental prices is used.
First one is based on inverted demand model. This method tries to identify
statistically significant variables that make up the house price and in this
way it estimates the proxy for house price, i.e. fundamental prices.
Fundamentals in our case are represented by real mortgage rate, disposable
income and net migrants as a percentage of total population. Second method is based on the conventional
asset pricing model, which equates the value of the asset to present discounted
values of future cash inflows. It is one of the most common ways of assessing
the fundamental value of houses. It looks at the imputed rent of a house. The
imputed rent of a house reflects the costs that arise from owning a house for
one period. In equilibrium these costs should be equal to the costs of renting
a house for one period (actual rent). In this paper I assume that fundamental
and actual rents are equal.
The choice of
econometric methodology is based on application of conventional techniques for
evaluation of asset prices. Inverted demand model and imputed rent model are
used for estimating non-fundamental prices. Johansen’s (1995) multivariate
approach for estimation of cointegration among variables is applied for
cointegration test between real house prices and its proxy (i.e. fundamental
prices).
According to AOS the
major condition for territorial sample is its representativeness in terms of
the completeness of data including all the regions and the account is taken of
the special features of economy in each of the regions. This is reached by
investigation of both administrative centres and rural population aggregates
that are considered to have high level of congestion of consumers market and
the ones that reflect specialization of industrial or agricultural production
of the region.
As it was mentioned
above fundamental prices are calculated according to two commonly used methods
in economics. First method is based on the work of Poterba (1991), who offered
three alternatives for explanation of house price movements. They are- shock in
construction costs, unanticipated inflation and demographics. Since then
economists added different variables that can explain house price variations.
For example Roche (2001) uses disposable income, real mortgage rate and net
migrants as variables for fundamental price. In my study of fundamental prices
for Kazakhstan I applied same variables. General form of the model appears as
following:
pft
= α + β1yt + β2 it +
β3 nt
Where pt is
average real new house prices, yt expected real disposable income, it
expected real mortgage rates and nt is number of net migrants.
The second method is
based on efficient market theory proposed by Fama (1970). Fama defined
efficient market as one, in which prices reflect relevant information.
Therefore the current price is unbiased predictor of its future value. Or
otherwise speaking: the fundamental price of the asset is equal to its
discounted present value. Meese and Wallace (1990) offer to use rents instead
of dividends in estimating current fundamental price. The general formula is:
Ptf=
Next step in my analysis was the evaluation of
non-fundamental price. Roche proposes to use residuals from the formula (1) as
a proxy for non-fundamental price.
If we subtract the mean price-rent ratio
mentioned in formula (2) from the real house prices we will obtain second
method of evaluation of non-fundamental price.
As defined by Roche the price of the
asset, Pt consists of two
elements: fundamental and non- fundamental price. Fundamental price, Ptf , is the price
which is driven by economic forces and is represented by healthy indicators,
whereas non- fundamental price, Ptnf
is the one that deviates away from market fundamentals and which, for example,
can be caused by excessive demand on houses in expectation of further price
increases. Thus:
Pt = Ptf + Ptnf
Where
fundamental price is assumed to be non-stationary:
Ptf = Pft-1 + et, et ~ iid (0, σe2)
And non-
fundamental price is persistent but does not grow forever:
Ptnf = ρ Pt-1nf
+ vt, 1 > ρ > 0, vt ~ iid (0, σv2)
In the
absence of unique model for fundamental price evaluation different proxy price
measurements are used, which Summers (1986) and Cutler et al. (1991) estimated with error-in-variables approach, such
that:
Ptp = Ptf + ut, ut
~ (0, σu2)
Where Ptp is the proxy
price and ut is the
measurement error. By manipulating
equations stated above it is shown that:
Pt+1 – Pt = β1(Pt
– Ptp) + εt, εt ~ iid (0, σε2)
Finally, by
rewriting change in price as an excess return from investing in housing and
assuming that the difference between actual and predicted fundamental prices as
non-fundamental price and adding measurement error we get fad’s model proposed
by Summers of the following form:
Rt+1 = β0 + β1
Ptnf+ ηt
Further I use the fad’s model in order to estimate
statistical significance of non-fundamental prices. Referring to the formula
(1) to find the residuals I ran the ordinary least squares regression in
Eviews, the program that is used during the process of current research.
The two methods of
estimating non- fundamental prices that were mentioned above are compared and
result on the presence of the non-fundamental prices in house market of
Kazakhstan is drawn.
Further on I will apply cointegration test proposed by
Johansen. My major aim is to find whether there exist stationary relationship
between real house prices and any of the explanatory variables mentioned in
formula (1). Johansen’s (1995) multivariate approach to cointegration is the
one that is provided by Eviews and which appears to be suitable for my
research, because it allows to uncover stationary relationships among a set of
non-stationary data. These relationships are interpreted as a long-run
equilibrium.
RESULTS
Estimation of the Fad’s Model
Running the
log-linear regression for estimation of non-fundamental price (Method
1)resulted in the following OLS estimation:
pt =
9.14 + 0.13yt + 0.53it + 0.01nt
with the t-statistics shown below the corresponding
regressors. According to the figures net migrants as a percentage of population
have insignificant explanatory power for the average real new house prices.
Kazakhstan is among the least populated countries in the world and number of
net migrants as a percentage of population in fact does not represent the
majority of population that can affect house prices. In this case the
proportion of population aged between 25 and 45 could give more explanatory
power; such information is only available on annual basis.
Using the series of values of the residuals supplied by the
program we now can use them as a series of non-fundamental prices in our fad’s
model. The missing bit in the equation
is the excess return on investing in housing, which is calculated as the
percentage change in house price less the monthly yield from holding 20-year
government bonds. Adding this series of data into fad’s model and running the
linear regression we will obtain the corresponding results.
Results show that t-statistics for both intercept and slope
coefficients are low and together with p-value denotes statistical
insignificance of the regressor. R-squared shows that in less than 8 cases out
of 100 the variability in excess returns from investing in housing is explained
by the proxy for non-fundamental price estimated by the first method. F-statistic fails to reject the null, which
states that the slope coefficient is equal to zero under 5% significance level.
The value of Durbin-Watson statistics indicates that there appears to be
positive correlation in successive error terms. So, overall we can conclude
that: a) the fitness of model is low, so our model requires more explanatory
variables in calculating the proxy for non-fundamental price, or; b)
non-fundamental price has very low explanatory power for movements in excess
return, so that larger time span should be taken, or; c) house price movements
in Kazakhstan during the period of January 2007 and May 2010 are driven by
market fundamentals.
There is
sufficient evidence to conclude that non-fundamental price is non-stationary
meaning that the residuals of the inverted demand model have linear
association. To get rid of non-stationarity I will use common procedure of
differencing the series. After the implementing first order differencing, the
Kwiatkowski-Phillips-Schmidt-Shin test failed to reject the null under
conventional significance levels of 1%, 5% and 10%. Augmented Dickey-Fuller
test also rejected the null hypothesis that states that non-fundamental price
has unit root at the significance levels mentioned above.
As the indicators state the model’s goodness of fit
represent a stronger model. R-squared increased and adjusted R-squared as well,
which indicates that model is improved by eliminating the unit root more than
it would be expected to improve by chance. But Durbin-Watson statistics’ value
of 0.82 constitutes for the evidence of higher first-order correlation of the
residuals.
Next, I will add the fundamental price component of the
non-fundamental price estimate, which is calculated by the second method. I am
implementing this in order to determine if this combination of explanatory
variables will lead to a better specified model. Namely I will run the
regression with two explanatory variables, one is non-fundamental price, and
the second is the data obtained from calculating
.
Now the model possesses better results in terms of the
goodness of fit. Durbin-Watson statistic’s value implies almost no linear
association among adjacent residuals. Akaike information criterion is
substantially less than in the previous case. Adjusted R-squared is positive
and R-squared itself shows that more than 60% of movement in excess return is
explained by the regressors. Non-fundamental prices persistently show
statistical significance.
Second method is based on standard asset pricing model or in
relation to house market the model is called imputed rent model, which states
that the fundamental price of an asset is equal to the present discounted value
of future dividends. Meese and Wallace (1990) use rent on housing as an
alternative for dividends. And later, Schaller and van Norden (1997) show that
the fundamental price is equal to the multiple of rents:
Ptf=
So that the second measure of non-fundamental price is equal
to real house price minus mean price-rent ratio multiplied by real rent.
As one can see both coefficients appear to be statistically
significant. It is proved by t-statistics with corresponding p-values. As well
as F-statistic’s p-value, which indicates that slope coefficient is not equal
to zero under 1% confidence interval. Akaike information criterion also shows
higher value, i.e the weaker goodness of fit as compared to the fad’s model
using non-fundamental price calculated by the first method. But R-squared value represents that only 24%
of movements in excess returns can be explained by non-fundamental price
calculated by the current method.
We now will test non-fundamental price calculated by method
2 for the presence of unit root. Kwiatkowski-Phillips-Schmidt-Shin (1992)
(further KPSS) test is used to verify that non-fundamental price are
stationarity. And the second test is Augmented Dickey-Fuller test (further ADF)
with the null hypothesis that non-fundamental price has unit root.
The null hypothesis of KPSS test states that non-fundamental
prices are stationary and it is rejected at 5% and 10% significance levels, but
it is not rejected at 1% significance level. ADF test shows 93.63% probability
that non-fundamental prices have unit root.
So overall one can state that we have sufficient evidence to conclude
that non-fundamental prices have unit root. Non-stationarity can lead to
elimination of standard assumptions for asymptotic analysis, so that usual “t-ratios”
will not follow a t-distribution, and one cannot validly undertake
hypothesis tests about the regression parameters[1]. To eliminate
non-stationarity I will use the differencing and generate new series in Eviews
by inputting equation of: dpnf02=pnf02-pnf02(-1).
We can observe that our model has now better fit: 66 percent
of movements in excess return can be explained by movements in non-fundamental
price. And Akaike information criterion supports this as well, because it
decreased from 3.15 to 2.34. Adjusted R-squared increased from 22.39% to
65.39%, which indicates that the new price estimation improves the model more
than would be expected by chance. Durbin-Watson statistics indicates less
evidence for linear associations in residuals.
Second method of calculating non-fundamental price includes
only values for fundamental price and real rent. Further on, real mortgage
interest rate data is added to the regression to understand whether this will
result in a better fitted model.
As we can notice non-fundamental price has persistent
indicators for explanatory power. R-squared and Akaike information criterion
show higher goodness of fit of the model. Adjusted R-squared increased. All the
explanatory variables appear to be statistically significant and non-zero
according to F-statistic. Durbin-Watson indicated almost no first-order
correlation of residuals.
Next I will add more independent variables, such as the
number of net migrants as a percentage of total population and real disposable
income (in this study real disposable income is approximated by subtracting
average consumption expenses from average monthly wages) to the regression one
by one, and compare their statistical properties. I will also compare different
combinations of the explanatory variables to find the one that fits the real
data best.
SUMMARY AND
CONCLUDING REMARKS
House price dynamics is among most prevalent issues
considered today by the economists. Vast literature is available for analyzing
the housing market, house price variations and estimation of fundamental
prices. Among the various models covered by existing literature, which estimate
asset prices the two most conventional models were chosen to analyze house
price movements in Kazakhstan during January, 2007 and May, 2010: the
structural model of evaluation long-run equilibrium based on demand and supply
side factors and imputed rent model, which equates fundamental house prices to
the present discounted value of future cash flows. Moreover, Johansen’s
cointegration technique was utilised to test whether real house prices have
long-run relationships with its fundamentals.
Estimation results of inverted demand model and imputed rent
model were nested into fad’s model and non-fundamental prices were subsequently
proven t be statistically significant in explaining house price movements in
Kazakhstan during the period mentioned earlier. That can be an alarming sign
for the government officials, because the history of asset price bubbles
starting from Holland’s ‘tulip mania’ in 1600’s and ending with recent mortgage
crisis demonstrates how entire economies were severely hurt by the collapses of
price bubbles.
The best fitted model in explaining house price dynamics
appears to be the one that incorporates non-fundamental price calculated by the
Method 2, logarithm of real mortgage rates and logarithm of the net migrants as
a percentage of total population.
Criticism of the current work consists of the following:
1) Very short
period of time was used in the research, whereas conventional researches
utilize the time span of more than 30 years period. Unfortunately, such data is
absent for Kazakhstan
2) As compared
to the existing statistical techniques and models this research utilised more
humble and unsubdivided techniques.
Further research areas are as vast as the country in
consideration. In the meantime Kazakhstan lacks researches related to its
housing market. The analysis of house price dynamics that was carried out in
current work can be further extended to test whether house prices reflect
market fundamentals, seasonal fads or bubbles. Or decision tree approach
proposed by Fan et al. (2006) can be
utilized in identifying the determinants of public housing resale prices. The
model is more effective in terms that it eases the conventional assumptions
about data distribution.