CC BY-NC-ND
69
m H
o
-û
en
o
o
CN
CN
U >
LO LU
cc
LU
U <
O
u
THE LAGGED STRUCTURE OF DYNAMIC DEMAND FUNCTION
FOR MORTGAGE LOANS IN RUSSIA1
Evgeniy. Ozhegov,
This study was carried out with support from The National Research University Higher School of Economics Academic Fund Program in 2013-2014, Research Grant No. 12-01-0130
Agatha Poroshina,
National Research University Higher School of Economics, Department of Applied Mathematics and Modeling in Social Systems, Group for Applied Markets and Enterprises Studies. Lecturer, Young Research Fellow.
<i
CO i Abstract
O 1
Key issues of government policy include providing of affordable housing, identifying the main drivers of mortgage borrowing and performance of mortgage loans. Therefore the understanding of the mortgage borrowing problems, developing optimal credit contracts and effective risk management systems, especially on the residential mortgage market, are becoming crucial.
u *
The paper presents the demand-supply model for the Russian residential mortgage market. It was estimated on the aggregated data for the period 2008-2012 and corrects for the autocorrelation and the endogeneity problems. The lagged decision-making process of borrower's application causes serial correlation of errors. Endogeneity is generated by simultaneity in borrower's and credit organization's decisions and its affection on demand and supply factors.
The study overviews current approaches in the mortgage literature to model credit demand and supply. Many papers are aimed to estimate the demand and supply functions for housing service, n ^ but not on the mortgages. Then Vector autoregression (VAR) approach was applied to estimate dy-^ ^ namic demand and supply model for the mortgage market and control for possible endogeneity in demand and supply factors. It helps to understand which supply and demand factors influence the client decision to participate in the mortgage debt market, how price on this market is determined, and to predict the demand for a mortgage loan of a particular credit organization. To relay on obtained results, the set of different tests were performed for each VAR model specification like the Lagrange multiplier test, tests for normality of residuals (the Jarque-Bera test, the Skewness and -Q ^ the Kurtosis tests) and stability of estimates.
r^ ^ Obtained results provide original insights into the determinants of borrowing process on the Rus—— ^
q sian residential mortgage market and allow predicting the probability of application for mortgage. ^ Robust estimates of demand function evidenced that decision on application for mortgage loan is
lagged. It means that the probability of application depends both on the current macroeconomic
situation and its dynamics in previous time periods. In long run mortgage demand correlates posi-
n ^ tively with supply shocks and consumers' income and negatively with both prices of mortgage o < and property.
^ JEL Classification: C30, R21, R31
o
Keywords: mortgage market, demand, supply, endogeneity, VAR
1. This study was carried out with support from "The National Research University Higher School of Economics Academic Fund Program in 2013-2014, Research Grant No. 12-01-0130 ".
O
< X
O i=
O CN
u >
m ¡
_Q LL
T L±J
CQ !<
cc ¡- 0
CC
C^ O
O u
o
Introduction
Key issues of government policy include providing of affordable housing, identifying the main drivers of mortgage borrowing and performance of mortgage loans. Therefore the understanding of the mortgage borrowing problems, developing optimal credit contracts and effective risk management systems, especially on the residential mortgage market, are becoming crucial.
The initial step of mortgage-borrowing process is an application of borrower to a particular credit organization. A potential borrower realizes the necessity of borrowing, chooses a credit organization and a credit program that reflects her/his preferences, fills an application form with demographic and financial characteristics. The probability of application measures the demand on the mortgage market. Econometric estimation of this model is faced with inconsistency driven by serial autocorrelation and endogeneity.
00
The lagged decision-making process of borrower's application causes serial correlation of errors. Endogeneity is generated by simultaneity in borrower's and credit organization's decisions and its affection on demand and supply factors. This research is aimed to avoid these challenges and to consistently estimate the demand on the Russian residential mortgage market. It helps to understand which supply and demand factors influence the client decision to participate in the mortgage debt market, how price on this market is determined, and to predict the demand for a mortgage loan of a ZL © particular credit organization.
The rest of the paper is divided into four sections. It starts with the brief literature review. The second section describes our methodology and data, which are used to obtain consistent estimates of the demand for mortgage loans. The third and fourth sections discuss the results and robustness check. GO ^ The final section concludes.
Literature review
cn
The classical dynamic economic model that describes cyclical supply and demand in a market, where
— x the amount produced must be chosen before prices are observed, is known as the cobweb model or
i 1 r^ cobweb theory. It was firstly introduced by Tinbergen (1930) and the cobweb theory proved by Eze-
bf < kiel (1938).
^ ^ The functions of demand and supply on the mortgage market are:
X YtD = a - bPt (1)
A ^ YS =-c+dPt_1 (2)
where
t - the time period,
YtD - the function of demand at the time period t, YtS - the function of supply at the time period t, Pt - prices for the mortgage at the time period t, Pt-1 - prices for the mortgage at the time period t-1,
O
^ i a, b, c, d - constant coefficients.
The supply on the mortgage market could be measured by different variables, for example, the numbers of mortgages and mortgage volume, which are offered to clients by credit organizations. In the same way, the number of applications for mortgage, the mortgage approval or denial rate, could characterize the demand on this market. When | b |>| d |, b ^ 0, d ^ 0, it can be shown that long-run market price moves to the market equilibrium.
Recent developments in macroeconomic and microeconomic theories stimulated the interest to study the demand and supply of household credit. Crook (2006) reviewed and compared results of
o
<
00 O
u u
@
studies that had examined the credit demand and supply across the countries. Studies which were aimed to explain inter-household differences in demand and the characteristics of the households that are most likely to be credit constrained showed considerable variation in the determinants of demand and in marginal effects within the countries as well as between them. Most of them have been published for Italy and the US markets.
Some papers on aggregated supply and demand on mortgage market, for example (Goetzmann et al., 2012; Xu, Zhang, 2012), take into account not only pricing factors, but other supply factors (loan-to-income ratio and loan-to-value ratio of issued loans) and demand factors (income per capita, unemployment rate, education rate, rate of retired residents) in order to control time and space variation.
Goetzmann et al. (2012) showed that demand (the number of mortgage applications and volume) should be included in supply equation (the mortgage approval rate) and vice versa to obtain efficient estimates. The approval rate was measured by both the percentage of loans and the percentage of applications that had been approved. In addition, they empirically found that both subprime borrowers and lenders considered past housing returns as a factor in a home purchase process and underwriting - perhaps extrapolating the recent past as a forecast of future home values.
Xu and Zhang (2012) analyzed the demand for nonlocal mortgages. They had shown that such demand factors as the share of out-of-state nonlocal mortgages issued should be modeled as endogenous because of their dependence on the supply (the denial mortgage rate). This paper provides evidence that the demand for nonlocal mortgages is higher in the neighborhoods where local banks tend more likely to deny mortgage applications, suggesting that borrowers of nonlocal mortgages are less creditworthy.
Magri (2002, 2007) found the statistical significance of households' socio-demographic characteristics such as age, current and future households' income, self-employment, education, residence in the demand for a loan in Italy. For example, the uncertainty of income reduces the demand for loans, except-self-employed workers, who are nevertheless subject to very rigid evaluation by X lenders. Another evidence is that age plays a very important role, essentially in the demand side.
^ ^ Cauley and Pavlov (2002) studied the demand function in housing market. Using the Los Angeles single-family dwellings data for 1985 - 1997, they founded strong evidence of lagged decisionmaking in house-buying process and necessity of using lagged explanatory variables to predict the demand response to demand shocks. The asymmetric response of real estate markets to positive ^ and negative demand shocks could be partly explained by the option value of the owner's interest.
The causal relationships in the American housing market were studied by Clayton et al. (2010). The structure of causality in price-volume process in housing market was modeled by vector -Q lu autoregression model (VAR) and tested by Granger test. It was shown that endogenous variables X - home price and trading volume should be modeled jointly by system of dynamic simultaneous
CD 5 equations. Both variables are affected by the labor market conditions, the mortgage and stock S q markets, and the effects differ across the markets. Authors find empirical evidence that home ^ cu prices Granger cause trading volume, but decreasing of price reduces trading volume, and price n q increases have no effect. Moreover, trading volume also Granger causes home prices, but only in O ^ markets with inelastic supply.
- Ll_
O Empirical studies of Russian mortgage market are limited mostly due to the lack of statistics. Polterovich and Starkov (2007) discussed strategies of developing the mortgage market in Russia
-Q
00
o
CN
CN
U >
LO LLl
cc
CL
O < and presented the model calculations for the strategy of creating mass mortgage in Russia, based
cc on the transplantation of modified construction and bank savings.
O There is extensive literature on the demand and supply for housing service, but not on the mortgages. In this paper we follow VAR approach to estimate dynamic demand and supply model for the mortgage market and control for possible endogeneity in demand and supply factors.
o
CN
< CQ
O
Ci
c; u u
.0
00
O
@ m
H
O CN
CN
-Û Ô
U
X<
LÜ
<f LH
X cd
"V LU
_û LL
CQ
2 O
< Cd
O ix
Cd
O
U
LL
O
_I
<
o
Methodology and Data
To estimate demand on the mortgage market, the following structural model was used:
| D = f (S, Df ), (3)
[ S = f2 (D, Sf).
where
D - demand function, S - supply function,
Df - demand factors,
Sf - supply factors.
The data set contains monthly aggregated regional data of the Agency of Home Mortgage Lending (AHML) branch performance, regional mortgage market characteristics and regional macroeco-nomic variables for 01/08/2008 - 31/08/2012. The data on mortgage market and macroeconomic variables is publicly available. The data on the AHML regional branch performance provided by itself and cannot be disclosed.
The AHML is state-owned provider of government-insured loans, which helps to implement government housing policy and to support mortgage lending in Russia. It uses two-level system of lending. On the first stage banks and non-credit organizations provide mortgage loans to households according to the common AHML standards. The second step is refinancing (redemption) of mortgage receivables by the AHML. The AHML develops special mortgage programs and refinances risks from its regional branches and commercial banks, which operate such programs. The list of programs contains "Young researchers", "Young teachers", "Mortgage for Soldiers", "Mothers' capital" and the other social and subprime programs. All of them have relatively high risk that is secured by the government.
The description of variables and summary statistics are presented in Table 1.
Table 1
Summary statistics
Variables Description Mean Std. Dev. Min Max
prob The amount of applications to AHML divided by the amount of households in the region - - - -
mean_LTV Average Loan-to-value ratio (LTV) in the region, % 58.24 3.40 47.9 64.9
median_maturity Median maturity for mortgage in the region, in months 201.64 12.7 173 222.2
median_rate Median contract rate for mortgage in the region, % 13.1 0.82 12 14,3
mean_DTI Average Debt-to-income ratio (DTI) in the region, % 34.81 0.7 33.44 36.68
lodging_coef_in_years Housing price to income ratio, years 3.48 0.68 2.57 4.65
mortgage_amount Total amount of issued mortgages in the region 896.57 528.89 134 2112
Demand on the mortgage market D is represented by the function for the probability of application, which is calculated as number of applications in month t divided by the amount of households. Amount of issued mortgages characterizes supply S on this market. The other variables can be divided into two groups: demand-side factors Df that characterize consumer's willingness to borrow, and supply-side factors S f that characterize bank's willingness to lend.
Df = (mean _ DTI, lodging _ coef _n _ years) (4)
Sf = (mean _LTV, median _ rate, median _ maturity) (5)
It is essential to take into account that variables have time dimension t, in this paper we adopt a vector autoregressive model (VAR) (4) for the system of dynamic simultaneous equations for demand and supply (3).
-û
en
o
CN
CN
U >
_û
u * 1988).
X w Stability of VAR estimates requires that the moduli of the eigenvalues of the dynamic matrix to lie
^ within the unit circle. As there is more than one lag in the VAR, it is likely that complex eigenvalues,
X c£ leading to cycles, will be encountered.
e
... <c
O
u
Y = < + aYt-l + a2Yt-2 + ... + aJt-m + bl Xt-1 + b2X t-2 + ... + bkXt-k + e, s = l,..., n
E (et\Y-i, Yt-2,...) = 0, (6)
E[eses' ] = Z, E[eseT' ] = 0 if s where
t = 1,...,T,T - the set of time moments,
rn
H 1 n
Yt = (Yt ,...,Ytn)- the vector of dependent variables at time t,
Yt-1,..., Yt_m - lagged values of dependent variables,
Om - a maximum lag of dependent variables,
-LU
!û
a0, a1,..., am - constant coefficients,
i-
2 1 n
et = (en...,ent )- vector of error terms.
u
Xt - the vector of independent variables at time t,
Xt-1,..., Xt-k - lagged values of independent variables, k - a maximum lag of independent variables,
b1,..., bk - constant coefficients.
If the variables are non-stationary and are cointegrated in the same order, the correct method to estimate the VAR model is the VAR in first-differences with the addition of a vector of cointegrating residuals, which is called Vector Error Correction Model (VECM). The proper specification, considering possible non-stationarity and cointegration, will be tested by Johansen approach (Johansen,
O The supply-demand relationship on the mortgage market and its determinants sometimes are ambiguous and there is ongoing debate in this area. To test causal relationship the Granger causality test was conducted. It allows determination whether one time series is useful in forecasting another _Q lZ (Granger, 1969).
~ Suppose we have the vector autoregressive model (VAR):
^ c^
^ O Yts = al + aYt-i + a2Yt-2 +... + amYt-m + bXM + b2X^ +... + bkX-k + est,s = 1,...,n £ g E(et\Yt-i,Yt-2,...) = 0 (7)
E[e4e4 ' ] = Z, E[e4eT' ] = 0 if s ^r
O
^ _i where
o <
-V ^ t = 1,..., T, T - the set of time moments, cc
O Yt - the vector of dependent variable at time t,
Yt-1,..., Yt_m - lagged values of dependent variable, m - a maximum lag of Y,
a0, a1,..., am - constant coefficients,
o
CN
< CÛ
O
ci
lu ^
u u
-0
CO
O
@ m
H
O CN
CN
et - the vector of error terms.
To test the null hypothesis that Xdoes not Granger cause Y, we add in (7) lagged values of X.
Yt = a0 + aYt-l + a2Yt-2 + ... + aJt-m + b1 Xt-1 + b2 X t-2 + ... + bkXt-k + et (8)
where
Xt - the vector of independent variable at time t,
Xt _,..., Xt _k - lagged values of X, k - lag of X,
b,..., bk - constant coefficients.
In regression (8) we remain all the lagged values of X that are individually significant. If the F-test shows that lagged X collectively add explanatory power to the regression, then the null hypothesis that X does not Granger cause Y is not rejected.
In addition, the Granger causality test allows determining endogenous variables, which leads to inconsistent estimates of parameters in (3). In case of a functional relationship:
Y = f (X) (9)
if Y Granger cause X is accepted, it means that X is endogenous variable in the equation (9) and is needed to be instrumented.
_û
u
_û
cû
2 O
u
CC <
LU LO LU
cc
LU
U <
<
cc
O ix cc O
U
LL
O
_I
<
cc ZD O
Results
On the first step endogenous and exogenous variables from supply Sf and demand Df factors were determined. The results of the Granger causality test are presented in Tables 2 and 3.
Table 2
p-value for Tests of Granger Causality Based on VAR (supply factors S f)
Dependent Variable Probability of application Mean LTV Median rate Median maturity all
Probability of application - 0.039 0.495 0.071 0.010
Mean LTV 0.849 - 0.004 0.138 0.017
Median rate 0.849 0.675 - 0.094 0.186
Median maturity 0.146 0.450 0.421 - 0.352
All supply factors do Granger cause the probability of application (p-value=0.010). The probability of application is statistically insignificant in specifications for all supply variables such as mean LTV, median rate and median maturity. The last result indicates that supply factors have to be exogenous in the model (3) for demand equation. In addition, this finding supports that supply factors can be used as instruments for demand estimation. In other words, covariance between supply factors and error term in the demand equation equals to zero.
From the demand side, mean DTI has to be exogenous in the model (3) for supply equation, while lodging coefficient in years - endogenous. It is confirmed by results of the Granger-Causality tests in Table 3.
Table 3
p-value for Tests of Granger-Causality Based on VAR (demand factors D f )
Dependent Variable Probability of application Mean DTI Lodging coefficient in years all
Probability of application - 0.004 0.072 0.001
DMean DTI 0.362 - 0.489 0.257
DLodging coefficient in years 0.000 0.001 - 0.000
o
CN
< CÛ
O
ci
lu ^
u u
-û
CÛ
O
@ m
H
O CN
CN
_û
U <
e
LU
Z
CÛ
2 O
u
CC <
LÜ 00 LÜ
cc
LU
U <
<
cc
O eu cc O u
LU
o
_I
<
cc ZD O
Note: D - the first-difference.
Based on the above-mentioned results, the econometric model for demand is the following:
f Dt = fi(Dt-i,...,Dt_m,D}"dt_i,...,De;dt-m,Df ,Sf )
(10)
{Def"d = f2(Dt-i,...,D-m,D^.i,...,D^,Df ,Sf ) where
D - demand on the mortgage market (the probability of application),
De"d - endogenous demand factors (first difference of lodging coefficient in years),
Df - exogenous demand factors (first difference of mean DTI),
Sf - exogenous supply factors (mean LTV, median rate and first difference of median maturity), m - a maximum lag of dependent variables.
Mean DTI, mean LTV, median rate and median maturity are the instruments for demand function because they are exogenous, according to the results of the Granger causality tests. However, demand is the function not only from demand factors, but from the supply ones too, which can be endogenous in the model (3). It is confirmed by results of the Granger causality tests in the Table 4.
Table 4
p-value for Tests of Granger Causality Based on VAR (supply S )
Dependent Variable Probability of application DLn mortgage amount DLodging coefficient in years all
Probability of application - 0.276 0.037 0.132
DLn mortgage amount 0.164 - 0.000 0.000
DLodging coefficient in years 0.001 0.441 - 0.006
Note: D - the first-difference.
The first difference of lodging coefficient in years does Granger cause the first difference of the natural logarithm of mortgage amount. It means that supply influences the endogenous demand factor. This finding proves that supply is endogenous. The extended econometric model for demand should be the following:
Dt = fi ( Dt_i,..., Dt D fi = f2 ( D_i,..., D St = f3( D_i,..., D^
r\ end end o o 2 o« \
, Dft ,,..., D,t St_l v-:, St_m , D ft, S f,t)
V ,t-1 , D end
f ,t_m '
r\end o o ng ni; \
f ,t-1,..., D f ,t_m, St-1,..., St_m5 D f ,t, S f ,t y
(11)
end end
, Df ,t_1,..., D f ,t_m , St_1,..., St-
, Df, Sf )
To check the stability of coefficients we follow Cauley and Pavlov (2002) and include lagged exogenous variables considering decision-making process on application as lagged. Thus, the model is the following:
Dt = f (Dt_1,..., Dt_m , Dfnd_1,..., Dfnd_m , St _1,..., S Dfnd = A (Dt_1,..., Dt_m, Dfnd_1,..., Dfnd_m, St_1,
ps ps ns ni \
t_m> D f ,t,..., D f ,t_k , S f ,t,..., S f ,t_k )
S< St_m> Df,t,..., Df,t_k , Sf,t,..., Sf,t_k) (12)
o _ -f i D D r\end r\end o o n« ng
S ^ f 3 (Dt_1,..., Dt_m , Df ,t_1,..., Df ,t_m , S t_1,..., St_m , Df ,t,..., Df ,t_k , S f ,t,..., S f ,t_k
Estimated parameters of demand in models (10)-(12) are presented in columns (1)-(3) in Table 5, correspondingly. Standard errors of parameters are in parenthesis.
In all model specifications for the demand equation estimated coefficients remain the signs and statistical significance, but the third model has also significant coefficients behind lagged exogenous
o
CN
< CO
O
cl
LU ^
u u
-Q
CO
o
@
o
CN
CN
-O £
U
X<
LU
<f CO
X cc
"v LU
S- 1
_Q ll
CQ
2 o
<
cc O
(X
cc O u
LL
o
_I
<
o
variables. However, in terms of adjusted R2 and log likelihood, the model specification, which includes endogenous supply and lagged independent variables, fits data better. The demand on the residential Russian mortgage market is mostly determined by the demand in the previous period of time, the demand factors and the supply. From the demand side, the lagged probability of application (with lags equals to one and two) and Dlodging_coef_in_yearst_2 are highly statistically significant. Significant coefficients are also behind such exogenous variables as mean maturity (and its first lag) and mean LTV These findings provide evidence for a lagged borrowing decision process and follow the results of Cauley and Pavlov (2002). Thus, a potential borrower makes a time lagged decision, which is based not on the macroeconomic situation in the current time period, but on its dynamics in previous periods.
Interpretation of signs of significant coefficients is strictly intuitive. Probability of application positively correlated with income (negative sign behind lodging coefficient and DTI ratio), negatively correlated with flat value (negative signs behind lodging coefficient and positive ones behind LTV), positively linked with extension of maturity and not affected by interest rate (the last two points jointly were revealed in numerous papers dedicated to analysis of credit constrained borrowers' demand for mortgage).
More clear interpretation of estimated VAR parameters can be shown by impulse response (IRF) and dynamic multiplier (DMF) functions for endogenous and exogenous variables respectively. With IRF and DMF it is possible to measure dynamic effects of change in endogenous variables, as well as the effects of change in endogenous variables by shocks in exogenous ones through the other endogenous variables. Thus, Fig.1 shows dynamic responses of mortgage demand to shocks in exogenous variables through DMF. Fig.2 demonstrates responses of demand to shocks in endogenous variables using orthogonal IRF.
Interpreting all the responses is similar to interpretation of coefficient signs, except for changes in rate. Thus, increase of mean DTI by one standard error gains the highest response of demand (up
to a half of standard error of demand in each period, oproh = 3 .3). Along with negative response of demand to shock of lodging coefficient it gives strong evidence of positive correlation between demand and consumers' income. Supply shocks positively impact the demand which corresponds with the equilibrium theory.
Median rate variable has no significant influence on demand change directly. However, it influences the supply and as a consequence, the demand for mortgages. Long-run dynamics of demand change by shock of rate shows negative impact of rate increase on demand for mortgage. Along with negative response to shock of DTI and positive response to shock of maturity it gives evidence of negative correlation between demand and price-of-mortgage indicators. Positive responses of demand for
mortgage to a shock of LTV reflect negative correlation between demand and price of property.
Table 5
Estimated parameters for the probability of application equation
Model (1) Model (2) Model (3)
probt-i 0 754*** 0.754*** 0.488***
(0.153) (0.162) (0.144)
probt-2 0.104 0.034 0.336**
(0.166) (0.177) (0.153)
pr°bt-3 - - 0.085
(0.144)
probt-4 0.013 0.018 0.048
(0.132) (0.127) (0.115)
Dlodging_coef_in_yearst-1 31.223 45.984 -27.681
(38.104) (39.026) (33.734)
Dlodging_coef_in_yearst 2 -57.841* -103.802*** -74.464*
o
CN
< CO
O
cl
LU ^
u u
-Q
00 o
<3 m
H
o
CN
CN
_Q X o
U Cd
< UJ
< co 1 1 1
X Cd
LU O
111 <
Z Ll_
x LU 1_
CD
1_ cc O
£ CL CC
CL o
O o
c U_ o
CL _1
O <
cc
ZD
o
Model (1) Model (2) Model (3)
(31.660) (38.703) (39.282)
Dlodging_coef_in_yearst 3 - - 58.698
(42.227)
Dlodging_coef_in_yearst_4 -36.060 -11.087 -58.153*
(35.949) (37.202) (33.305)
Dmean_dti -9.183 -5.966 -7.984
(7.696) (7.859) (6.449)
Dmean_dti. 1 - - -19.352***
(6.341)
mean_ltv 1.826 1.321 4.056***
(1.530) (1.543) (1.439)
mean_ltvt-1 - - -2.097
(1.891)
mean_ltvt 2 - - -0.452
(1.517)
median_rate -8.181 -8.274 8.529
(5.917) (5.780) (6.893)
Dmedian_maturity 1.195 0.773 1.388*
(0.944) (0.975) (0.720)
Dmedian_maturityt 1 - - 1.479**
(0.698)
Dln_mat-1 - -13.390* 0.373
(7.376) (7.632)
Dln_mat 2 - -10.741 23.400***
(6.933) (7.265)
Dln_mat-3 - - 15.524***
(5.894)
Dln_mat4 - -0.578 -8.527
(5.201) (5.451)
Constant 3.875 37.191 -398.091***
(65.650) (65.911) (132.815)
Log likelihood -194.629 -192.772 -174.245
Adjusted R2 0.615 0.646 0.845
Note:*p < 0.1, **p < 0.05, ***p < 0.01
Responses to a Shock in Mean DTI
Responses to a Shock in Median Maturity
Responses to a Shock in Mean LTV
Responses to a Shock in Median Rate
Fig.1. Responses of demand to shocks in endogenous variables
-10
-20-
3-
-10-
-20-
12 0
4
o
CN
< CÛ
O
ci
lu ^
u u
-û
CÛ
O
@ m
H
O CN
CN
_û
U
_û
cû
2 O
u
CC <
LÜ 00 LÜ
cc
LU
U <
<
cc
O eu cc O u
LU
o
_I
<
cc ZD O
Responses to a Shock in Lodging Coefficient
a Shock in a Supply
Fig.2. Responses of demand to shocks in exogenous variables
Robustness Checks
To relay on obtained results, the set of different tests were performed for each VAR model specification. Based on the statistical significance of lagged variables in the Wald test, variables with corresponding lags were used. Observation of all the VAR models revealed the absence of autocorrelation in error terms, the normal distribution of error terms, and the absence of unit roots. To this purpose the Lagrange multiplier test, the Jarque-Bera test, and the test of VAR estimates for stability condition were performed correspondingly. The results of these tests applied to the three specifications of application probability model are presented in Table 6-7.
Table 6
p-value for Tests of the Lagrange Multiplier Based on VAR models
lag Model (1) Model (2) Model (3)
1 0.45668 0.22008 0.33127
2 0.38255 0.51196 0.87268
Table 7
p-value for Tests of the Jarque-Bera, the Skewness test, and the Kurtosis test Based on VAR models
Test Model (1) Model (2) Model (3)
the Jarque-Bera test 0.29198 0.47653 0.46312
the Skewness test 0.27432 0.28061 0.27436
the Kurtosis test 0.32865 0.63413 0.62144
In addition, for all specifications of application probability model all the eigenvalues lie inside the unit circle, which is reported in Table 8. It means the absence of unit roots and that VAR models satisfy stability condition.
Table 8
Test for the unit roots on VAR for the Probability of Application
Model (1) Model (2) Model (3)
Eigenvalue Modulus Eigenvalue Modulus Eigenvalue Modulus
0.4703276 0.7342361i 0.871958 0.8962194 0.896219 0.964999 0.964999
0.4703276 -0.7342361i 0.871958 0.3845592 0.7185519i 0.814986 -0.942299 0.942299
-0.8580232 0.858023 0.3845592 -0.7185519i 0.814986 0.4287639 0.80291i 0.910022
-0.617194 0.4774606i 0.780319 -0.5251282 0.5179861i 0.73761 0.4287639 -0.80291i 0.910022
-0.617194 -0.4774606i 0.780319 -0.5251282 -0.5179861i 0.73761 0.6912427 0.39293i 0.795124
0.4804812 0.4765577i 0.676734 -0.6721792 0.672179 0.6912427 -0.39293i 0.795124
0.4804812 -0.4765577i 0.676734 0.6158586 0.1335075i 0.630164 -0.341006 0.52124i 0.622864
0.07134631 0.5604926i 0.565015 0.6158586 -0.1335075i 0.630164 -0.341006 -0.52124i 0.622864
10
12 0
12
0.07134631 -0.5604926i 0.565015 0.09614097 0.5665891i 0.574688 -0.408289 0.408289
-0.3016162 0.4701563i 0.558587 0.09614097 -0.5665891i 0.574688 0.3808256 0.380826
-0.3016162 -0.4701563i 0.558587 -0.04656162 0.1857423i 0.191489 0.0259851 0.36049i 0.361428
0.5292138 0.529214 -0.04656162 -0.1857423i 0.191489 0.0259851 -0.36049i 0.361428
The Johansen test for cointegration gives the evidence of no cointegration among the dependent variables, so the VAR in form of (9) is consistent and there is no necessity of VECM estimation.
o
CN
< CÛ
O
ci
lu ^
u u
-0
CO
O
@ m
H
O CN
CN
-û Ô
U
X<
LU
<f CO
X cc
"V LU
S- 2
_û LL
CQ
2 o
<
cc O
(X
cc O
U
LL
o
_I
<
o
Conclusion
This paper provides original insights into the determinants of borrowing process on the Russian residential mortgage market. Using monthly aggregated regional mortgage data for 2008-2012, we estimate the dynamic demand-supply model that allows predicting the probability of application. It controls possible endogeneity of demand and supply factors and autocorrelation in error disturbance process. Firstly, the results show that all supply side and demand side variables Granger cause the probability of application. Secondly, we empirically established that borrowing decision process is lagged. In other words, the probability of application depends both on the current macroeconomic situation and its dynamics in previous time periods. In long run mortgage demand correlates positively with supply shocks and consumers' income and negatively with both prices of mortgage and property. The further research may reveal how different demographic variables such as gender, age, education level, unemployment rate and etc. may affect demand and supply on the mortgage market. However, for this purpose micro-level or cross-regional aggregated data is required.
References
1. Cauley S.D., Pavlov A.D. Rational delays: The case of real estate. Journal of Real Estate Finance and Economics, 2002, 24(1), pp. 143-165.
2. Clayton J., Miller N., Peng L. Price-volume correlation in housing market: Causality and co-movement". Journal of Real Estate Finance and Economics, 2010, 40, pp. 14-40.
3. Crook J. Household debt demand and supply: A cross-country comparison. The Economics of Consumer Credit, edited by Bertola G., Disney R., Grant C. MIT, 2006, pp. 63-92.
4. Ezekiel M. The Cobweb theorem. Quarterly Journal of Economics, 1938, 52, pp. 255-280.
5. Goetzmann W.N., Peng L., Yen J. The subprime crisis and house price appreciation. Journal of Real Estate Finance and Economics, 2012, 44, pp. 36-66.
6. Granger C. W. J. Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 1969, 37 (3), pp. 424-438.
7. Johansen S. Statistical analysis of cointegrating vectors. Journal of Economic Dynamics and Control, 1988, 12, pp. 231-254.
8. Kosareva N.B. Osnovi ipotechnogo credtovanya [Basics of Mortgage Lending]. Moscow, Foundation «Institute for Urban Economics», 2006.
9. Magri S. Italian households' debt: determinants of demand and supply. Banca d'ltalia, Vol. 454, 2002.
10. Magri S. Italian households' debt: the participation to the debt market and the size of the loan. Empirical Economics, 2007, 33(3), pp. 401-426.
11. Polterovich V.M., Starkov O.U. Strategya formirovanya ipotechnogo rinka v Rossii [Strategy of development of a mortgage market in Russia]. Economics and Mathematical Methods, 2007, 43(4), pp. 3-22.
12. Tinbergen J. Bestimmung und Deutung von Angehtkuven. Eien Beispiel, Zeitschrift fur Nationalokonomic, 1930.
13. Xu Y., Zhang J. Nonlocal mortgage lending and the secondary market involvement. Journal of Real Estate Literature, 2012, 20(2), pp. 307-322.
