Assessment of the situation on the regional housing market in Russia

It is also worth mentioning that during the period in study there were a different number of regions in Russia - some of the regions were included into the others: some, vice versa, were separated. Those regions that stopped their independent existence between 1996 and 2012 were included as separate object of observation if all the variables were available for at least 5 years. Otherwise the values of each variable for the region were from the beginning added to the values of the region which turned out to be its absorber. Those regions that were separated during the period in study were included whatever the time they appeared (anyway the minimal length of time series for such regions was 5 years). As a result 85 regions were observed during 17 years, so the total number of observation is 1445.

Both demand side and supply side indicators can be collected from official free sources such as Federal and Regional Statistics Services and The Central Bank of Russian Federation.Regional-level data is availableonly in “Russian regions Handbook” which is published by Russian Federal Statistic Service on a yearly basis.

The Bank of Russia provides analysts with regional-level information on mortgage rates, but regional differences of loan rates are unobservable.Cross-regional difference of borrowing rates is negligible for a couple of reasons. First of all, large constructors and developers are borrowing money not only in one particular region - they can optimize their choice and find cheaper funds, which makes space arbitrage impossible. Besides, if somewhere loan rates were higher in comparison with other regions, banks would have been started allocating more resources and issuing more loans there. At the same time banking can be considered as competitive industry - they “sell” undifferentiated product (money), so in order to “sell” more they compete on price (loan rate), and as a result interest rates become more or less equal to each other. (Wagner 2008)

Table 2.

Sources of information about factors studied in the research

Factors	Source of information
Housing prices, residential real estate stock, total population, Gross Regional Product (GRP), Consumer Price Index (CPI), size of dwelling, uninhabitable housing, disposable income per person, total workforce, unemployment level, inflation rate, construction cost index, average salary, non-housing consumption prices, current consumption share in personal income, housing consumptions share in personal income, financial assets consumption share in personal income	Publications of Russian Federal and Regional Statistics Services
Average mortgage rate, interest rates	Official cite of The Central Bank of Russian Federation

All the indicators have numerical values; however they are measured with help of different units. The Table 3 reflects each variable name in the research and their units of measurement.

Table 3.

Variables names and units of measurement

Indicator	Variable name	Units of measurement
Dependant variables
Housing prices	Real_HPI	Rub
Stock of real estate available by the end of the year	HS	thousand square meters
Demand-side indicators
Total population of the region	Total_pop	mln. citizens
Regional consumer price index	CPI	%
Average monthly disposable income	Real_disp_income	Rub
The share of current consumption of perishable and durable goods in disposable income	CGS	%
The share of housing consumption in disposable income	HC	%
The share of financial assets consumption in disposable income	FA	%
Average mortgage rate	Mortgage_rate	%
Unemployment	Unemployment	%
Supply-side indicators
Size of dwelling	Size_of_dwelling	thousand square meters
Uninhabitable residential real estate	UnH	thousand square meters
Average salary	Wage	Rub
Construction Cost Index	CCI	Index units
Average rate at which companies borrow money in Russia	Loan_rate	%
Gross regional product	GRP	mln.rub

It should be mention that the trickiest issue for real estate researchers is measurement of real estate prices. Generally there are two methods of coping with that task: housing price index construction and prices of registered deals. Using indexes allows more or less frequent and precise estimation, however there are several problems connected with their implementation. First of all, smoothing problem that appears because of illiquidity - estimated prices of real estate objects are usually used for index calculation instead of deal prices, but revaluation of these object occurs not as often as the frequency of calculating the index. This leads to lower volatility and seasonality in time series of prices because revaluation of dwelling is usually made in the last quarter of the year. One more drawback of house price indexes is time-lag of calculation. As a rule, information that appraiser has about the object is insufficient, so the specialist has to spend quite lot of time for doing precise estimation.

On the other hand deal prices formation is really opaque, prices depend on a plenty of indicators such as center location, neighbors, specific features of the object itself, etc. (Krainer and Wilcox 2013)(Yunus and Swanson 2013) It is hard to find two identical objects that totally match according to all the parameters, so each piece of real estate is unique, that is why it is hard for external investor to evaluate it precisely. However average price of real estate of each Russian region is calculated by Federal State Statistics Service, which is considered as an official and reliable source of information, whereas there is no housing price index in Russia that would be calculated on permanent basis by some well-known agencies. Therefore within this research the choice was made in favor of deal prices.

The methodology of average real estate price calculation presented by Russian Federal Statistics Service implies collection of primary information from companies and/or sole entrepreneurs, whose operations include buying and selling real estate objects in particular territories - cities, suburbs, regions, etc. All the information is collected on the regular basis; all the figures are calculated as of the 25th day of the last reporting quarter month (or the next working day after it). On the secondary market the average price per square meter of apartment is calculated as a weighted average based on actual transaction prices per square meter of total area and on the total amount of square meters of all apartments that were sold during the period. So, the formula used for calculation is the following:

	(19)

Where - average price per square meter for the period t; - actual deal price per square meter of i-th object of real estate; - total area footage of i-th object of real estate; - total number of real estate objects sold tor t period.



Data description

The graph of real housing prices of all Russian regions during the period in study is presented onthe figure4. First of all, similar dynamic of real prices in each region can be observed - there was a slight positive overall trend between 1996 and 2012, however there were obvious boom and burst of prices after 2005. The boom had been caused by mortgage loan market expansion - mortgage mass market appeared in Russia in 2005 and the financial product became popular very soon: in 2006 there was a considerable real estate demand increase which pushed prices in average up by 48%.

Fig. 4. Real housing prices of all Russian regions in 1996-2012

This phenomenon can also be an evidence of the fact that Russians consider real estate as real asset, which can help to ensure the safety of capital. After the period of hyperinflation the majority ofRussian people lost their savings and cut their consumption, however the moment mortgage market appeared they made the great demand for such expensive and illiquid asset as real estate. So it can be suggested that they hoped to save the capital from another possible round of inflation. However after the period of boom there was a period of burst - because of financial crisis in 2008-2009 real wages of Russians dropped quite dramatically, so demand for real estate and prices plummeted as well.

Besides high volatility of prices within each region, the difference between regions was quite high: the highest line on the graph reflects housing prices in Moscow region and it is quite obvious that prices there were almost 50% higher before mortgage boom in 2006 and 100% higher after it. It is also worth mentioning that the period in study is long enough and it covers at least two economic cycles. The sample includes two crises (in 1998 and in 2008), two period of recovery after them and one period of growth between them. All the period can be characterized with different economic conjuncture, different risk aversion parameters, etc. which affect both demand and supply on real estate market.

Despite the fact that the sample is not homogenous there is no need to get rid of outliers, because as it was mentioned before the general population is studied.Unobservable individual characteristics can be taken into account in the model in both cases: if there is no reason to believe that they are correlated with independent variables and if there can be assumed such correlation, but appropriate method of endogeniety correction is used for coefficients assessment.

Descriptive statistics of all the variables are presented in the Table 4 below.There isempirical evidence that real housing prices in Russia were highly dispersed during the period in study - the standard deviation of the indicator is about 56% of overall mean value.Anyway it also should be mentioned that during the period under observation the demand for housing (measured in square meters per person) also fluctuated significantly - for instance in Moscow area it had rocketed up to 47% before crisis.



Table. 4.

Descriptive statistics of the variables

Consumer price indexes of different region did not vary a lot, because neighbor regions usually have tight economic connection and according to purchasing power parity prices in different regions were more or less the same if only regional government didn't take some restricting actions. But there was a high intertemporal variance, because during recessions - in 1998-2000 and in 2008-2009 there were inflation shocks in Russia. Because of hyperinflation in some periods financial industry in Russia could not work properly that is why loan rateduring 1996-2012 varied from 8.4% to 147%.

One may wonder why the real amount of financial assets consumption can be negative whereas the amount of current and housing consumption is non-negative. This phenomenon goes from Russian Statistical service methodology of financial asset value calculation. This indicator accounts accumulated change of financial assets on year-to-year basis. During several crises that occurred during the period of observation there were inflation shocks when consumer prices could rose up to 200% a year so the real value of the assets decreased because of inflation. Besides during crises financial assets may experience significant drawdown which also diminishes their value. As a result negative values can be observed. At the same time the methodology of housing consumption calculation does not imply accounting for depreciation or appreciation and therefore it can be at least zero.

It's also worth noticing that at the same time construction cost index which is calculated as a year-on-year change of materials, machinery, details and design costs as well as other non-salary costs of construction demonstrated very unstable dynamics during the whole period under observation though it was similar for all the regions (see figure5). This graph shows that time-series data of CCI is non-stationery, which makes impossible inclusion of time-series variation into the model because of unit root. Therefore the first difference of this indicator should be included into the final model in order to get rid of this problem.

Fig. 5. Construction cost index dynamics

Dynamics of the relative wage to capital costs (CCI) also deserves attention. The graph of this indicator demonstrates that inflation of labor cost for construction companies during past 16 years was much more severe compared to changes in cost of materials. And it is quite straightforward because the technology usually becomes cheaper with time - new technologies of construction are developed, new cheaper materials are used, and mechanization of construction becomes more widespread. At the same time labor union pressure, development of workers' rights protection laws, overall increase of life quality and employees compensation after the collapse of the Soviet Unionlead to such a rapid growth of labor cost/material cost ratio.

It should be mentioned that the calculated ratio reflects only dynamic of the labor cost/material cost indicator but not the actual value because labor cost is approximated with real salary whereas material costs were calculated as index. These regional time-series are also non-stationary; therefore first differences (which have no unit roots) will be used for model estimation. The rest of the variables are more or less stationary in time so the level of variables will be used for estimation of theoretical model parameters

Fig. 6. Labor cost/material cost ratio dynamics

In order to identify the connection existence between endogenous and exogenous variables the correlation coefficients were studied. The majority of independent variables have strong linear impact on housing prices. The signs of correlation coefficients are mostlyin line with expected signs which comply with economic theory.

Table. 5.

Correlation coefficients.



As it was assumed stricter loan conditions negatively affect housing prices, whereasthe amount current consumption is positively correlated to housing prices. It also can be noticed that the state of the regional economy reflected through real GRP comparatively strictly linked to consumption variables and to the mortgage conditions and at the same time it is rather highly correlated with the housing prices. Despite the fact that there is no multicollinearity in the strict sense it could be assumed that one of these variables could be insignificant because some of them can drag influence one from another.

Among the components of the supply equation a positive linear correlation between the size of dwelling and loan ratesseems to be quite unexpected. Anyway the coefficient itself is quite small and probably correlations can vary from region to region therefore the final conclusions could be made only after the whole model assessment.

When the multicolinearity is tested the empirical model can be estimated. The estimation of the model parameters is implemented with help of the package of statistical analysis Stata 12th Version. The empirical analysis of the theoretical model on the Russian regional-level data will allow testing the significance of the whole suggested framework of the interaction between supply and demand on the housing market. And if the model is proved to be valid for forecasting equilibrium states of the market, the analysis of model residuals will allow drawing some conclusions about the possibility of non-equilibrium states. Theseresults in turn may give a basis for further research devoted to Russian real estate market efficiency, estimation of housing demand and supply elasticity, modeling of the price adjustment mechanisms, etc.

Empirical estimates

The model was estimated with help of maximum likelihood method which allows to obtain consistent, asymptotically normal and efficient (there is no alternative estimator which allows achieving lower asymptotic variance) estimates if the sample is big enough and Gauss-Markov conditions are fulfilled.

Table 6.

The results of estimation of the initial structural model.

Dataset was already checked for multicolinearity, however heteroscedasticity was detected with help of Breusch-Pagan test. The influence of heteroskedasticity can be taken into account when assessing the significance of a particular variable by implementing robust estimation. Anyway none of the variables changed its significance level noticeably, so the influence of heteroscedasticity can be considered as negligible.

Empirical model as a whole is significant which means that the joint test of all the coefficients being equal to zero allowed reject the null hypothesis and therefore the specification of the model describes the reality quite well. This statement can also be supported with the high R-squared of the each equation included into the model and the overall goodness-of-fit (see table 7 below)

Table.7.

Goodness-of-fit of the initial model.

Endogenous variables	R^2
Ln_HP	39.89%
Ln_deltaH	11.97%
Overall	40.32%

However the results that were obtained after model estimation appeared to be quite unexpected. First of all due to the fact that such variable as construction cost index and wages were taken as a first differences in order to eliminate unit root in the initial data inclusion of trend variable (FEyear) was a bad idea, because it became insignificant at any level.

The influence of regional-specific growth factor which was approximated with help of GRP growth rate appeared to be insignificant as well. The reason of such result may lie in the fact that there is no significant cross-regional variation of GRP growth rate and all the regions during the period under observation were developing along with the national economy. And the influence of the national business cycle had already been accounted into other variables in particular in the credit market conditions (loan rate and mortgage rate) reflected in the variable ln_HC = . So, both time trend and regional-specific growth rate were removed from the initial model which allowed increasingthe goodness-of fit of the supply equation and the whole structural model as well.

The most unexpected result is the positive influence of the housing price on the households' demand which follows from the positive coefficient of the ln_HSP (housing stock per person) variable. This result contradicts the basic law of demand and common sense that higher prices restrict some households' real estate consumption. The problem can be caused by the wrong model specification, data features (which means that that phenomenon really existed in Russian housing market during the period in study) or endogeniety problem which could become a reason of biasedinconsistent estimates and the wrong sign in the demand function.

According to the results of Wald testwhich allowed testing the joint hypothesis of non equality to zero for all the coefficients the model as a whole is significant. Therefore the first reason of the incompliance of empirical results to economic theory can be rejected.

At the same time the existence of the positive connection between the price of the housing and the amount of housing consumption can be explained as a phenomenon of Veblen good. That means housing can be considered as a positional good for Russians, however this violates the main assumption of the model about rational households' behavior. And besides that effect usually implies the high income of the consumers however the average real income in Russia grew with much more moderate paces compared to the growth of housing prices. And finally there is no statistical evidence of such phenomenon according to the linear correlation coefficient which is negative (see Table 5) Therefore the second reason of wrong sigh in demand function with high likelihood can be also rejected.

Finally the existence of the endogeniety problem can be assumed. This problem can be caused by wrong measurement of the indicator, simultaneity problem, self-selection bias or omitted variables in the model. Due to the fact that the same indicator was proved to be a relevant and suitable in many other research papers such as (Kenny,1999), (Cheshire and Sheppard, 1998), (Fingleton,2008), etc. because the results that were obtained complied totally with economic theory. So housing stock per person can be believed as a reliable measurement of the amount of housing available on the secondary market for a particular household.

Simultaneity problem means that the amount of housing changes immediately along with housing prices fluctuations which seem to be absolutely unrealistic assumption. In average the building process of apartment house in Russia lasts more than one year and the assumption used in the model implies that the housing becomes available on the secondary marketat the period which follows its commissioning.

Self-selection is a problem which is connected to the sampling process, however due to the fact that all the regions participate in the model estimation there actually no such a process because general population is studied. Therefore this cause can be rejected as well which in turn leads to the conclusion that the most possible reason of housing quantity endigeniety is the presence of omitted variable. For instance the availability of the land lots that are suitable for residential construction is an important factor of housing prices and construction itself, it can be assumed that it vary from region to region, but this indicator cannot be directly measuredtherefore cannot be included into the model.

Anyway this means that there is a correlation between the housing stock per person and individual error term in the model. In order to eliminate the effect sucha bias this variable should be instrumented with help of other variables that cannot be connected to the error term. Instrument variables have to be valid and relevant, which means that they should be exogenous and provide high descriptive capacity (e.g. influence significantly on endogenous variables) subject to other Gauss-Markov's conditions.

The results obtained in the previous research showed that housing construction tightly connected to the business cycle. This influence of the business cycle of the construction activity was found not only for developed countries (Tsatsaronis and Zhu 2004) but also for developing ones, in particular for the Eastern Europe countries(Catte et al. 2004). And this influence seems to be sustainable in time according to the model of multi-sector equilibrium developed by (Davis and Heathcote, 2001).

Taking the look on those results through the lens of the established theoretical model it could be stated that when the economy is rising constructors form upward housing price expectations and to the certain moment construction activity increases along with economic boom. However when the amount of construction becomes unsupported by effective demand which signals slowing of the economy the price expectations become negative and construction activity moderates.

So it can be assumed that the indicators of the business cycle that were removed from the initial structural model because of their insignificance (regional specific growth approximated through GRP growth rate and overall economy trend presented by time variable) can be relevant factors determining construction activity in the region. Besides the indicators of regional and national economy are unlikely connected with suchpotential omitted variable as residential land lots availability or quality of construction.

All things considered in order to eliminate the bias of the initial model estimator the intermediate estimation of additional equation should be implemented. Based on the results of previous research it was assumed that the amount of residential real estate building is a pro-cyclical variable which means that it is positively correlated to the indicators of the regional and national business cycle. If this instrument set is proved to be valid and relevant and does not cause further endogeniety the predicted within the new equation values of the endogenous variable (ln_HSP) will be used for further structural modeling assessment.

In order to obtain consistent estimator fixed effect models need to beassessed, because this is necessary to obtain some benchmark to compare other estimators to. Fixed effect estimators are consistent because under assumption that all the Gauss-Markov's conditions except possible correlation of independent variables with idiosyncratic error term are provided within transformation allows eliminate individual effects and therefore to get rid of possible endogeniety. The results of fixed effect estimation of intermediate model are presented on figure7.

Fig. 7. FE estimates of intermediate model

As it was assumed both regional level and national-level indicators of business cycle appeared to be significant factors of construction activity in Russian regions despite the fact that the absolute values of both coefficients are relatively small.Furthermore this result also supports the statement that residential real estate construction is a pro-cyclical variable. It is also worth noticing that the model better describes intertemporal variation compared to cross-regional probably because the regional-specific effect did not varied a lot across regions but its variation precisely describes dynamics of construction in time and at the same time an overall trend is also supposed to reflect the influence of time effect.

In order to check the model for possible endogeniety consistent fixed-effect estimators should be compared to random-effect estimators. If there is no statistically significant difference between those two sets of estimators then the model predicted values of housing stock per person can be used for the estimation of the structural model.

The results of random effect model estimation are presented on the figure 8 below, this model as a fixed-effect model was checked for heteroskedasticity and as a result robust estimation was implied. Even without testing both sets of estimators for statistical compliance one could notice that the coefficients are almost the same which means that the intermediate model does not cause further endogeniety and the predicted values of endogenous variable (housing stock per person) can be used for estimation of the structural model of Russian regional housing market.

Fig. 8. RE estimates of the intermediate model

The new variable which was extracted out of the reduced-form model was inserted into the whole model under the name lnHSP_hat and as it was assumed earlier this step helped to improve the initial model significantly. The results of the new model assessment are presented in the table 8 below.



Table 8.

The results of the final model estimation

Use of the predicted variable allowed to some extent eliminate the influence of endogeniety and the sign of the coefficient before the amount of housing in the demand equation become negative as it is required by economic theory. So at that point the model can be considered as a suitable for further interpretation and discussion of the results.

It goes without saying that the model stayed significant an all levels and what is more the total quality of the final model improved compared to initial model for both separate equations and the overallmodel (see table 9 below).

Table.9.

Goodness-of-fit of the final model.

Endogenous variables	R^2
Ln_HP	46.02%
Ln_deltaH	12.31%
Overall	46.57%

Therefore it could be concluded that the suggested theoretical framework is basically relevant however housing stock per person which was taken as an indicator of quantity in the demand function appeared endogenous presumably because of omitted variable. For further research of Russian regional housing markets one should use another variable for measurement for quantity of housing, include the indicators connected to residential building land availability and quality of construction measurements in the model or include an additional equation in the model which would describe the connection of the demanded quantity of housing with other variables that participate in the equilibration on the housing market.

Results of the model estimation

The final model almost totally fulfilled the expectations about all the exogenous variables influence on endogenous ones (housing prices and net amount of residential real estate construction). Besides, due to the fact that the model as a whole is also significant according to rejected joint hypothesis about all coefficients being equal to zeros it could also be concluded that the suggested paths of influence of each of the demand-side and supply-side indicators are correctly determined as well.

Structural model estimation approach implies that the signs of the regressions coefficients are dictated mostly by the economic theory and less by the researcher's assumptions. It should be noted that all the estimated coefficients correspond to the economic theory and most of the explored variables appeared to be highly significant. All the components of theoretical demand function - consumption and housing related are equally significant for price determination. However supply function is mostly driven by housing prices rather than cost inflation, because both components of total cost function - labor expenses and cost of capital goods - are significant only at 5% and 10% levels respectively.

This phenomenon can be explained in the following way. Construction companies in reality are not perfectly competitive and therefore they can have some market power to persist marginality of their business at a stable level. Within the framework of the research companies have to reduce construction activity and inflate prices back to the level which would keep their operational margin stable when their total expenses increase.

The mechanism of the interaction between housing prices and conditions on the labor market was described in the paper of (Bover, Muellbauer, and Murphy 1989). Authors found similar evidence of negative connection between wage level and housing construction but the positive influence of labor cost inflation on housing prices on the UK housing market. The wage in their model was incorporated in both demand and in the supply functions through its inverse relationship with unemployment level.

Anyway in order to answer the question about the power of influence of demand-side and supply-side indicators on equilibrium housing price both direct and indirect effects should be studied. Their values are presented in the table 10 below.



Table 10.

Direct, indirect and total effects of exogenous variables on endogenous ones

Direct effect of each variable reflects the influence of a particular exogenous variable on corresponding endogenous variable in the equation which contains both of them. This effect is actually equal to the estimated coefficient in structural themodel. Indirect effect on the other hand reflects the path of influence of exogenous variables from one equation on the endogenous variable from another one. Total effects are the effects that incorporate both direct and indirect ones and allow researcher to observe the influence of all the exogenous factors on all the endogenous ones.

This effect appears when the endogenous variable from one of the equation is used for modeling the other endogenous variable. In the estimated model housing prices that were determined within inverse demand equation participated in the determination of net size of dwelling. Therefore the exogenous variables from the demand function such as housing stock per person, aggregate current consumption and the relative costs of buying real estate coefficient indirectly affect net size of dwelling.

One of the most influential housing demand factors in Russian regions is the amount of housing available on secondary marketper one household: when this amount rises by 1 % real estate prices drop by 0.59%. If the direct demand equationwould be constructed instead of reverse demand function the causal relation could be inverted: when housing prices go up by 1% the demand for residential real estate contracts by 1.69%.

This fact reflects the simple idea of the law of demand and the value of the coefficient indicates that demand for housing is price elastic. This result supports estimation conducted by (Mayo, 1981) on state-level data for the USA sample and partly results of the city-level research conducted by (Hanushek and Quigley, 1980) who proved that housing demand is more elastic for relatively expensive objects of residential real estate.

It is not a surprise that such result was obtained for Russian sample because for most people housing is a very valuable asset, often - the most valuable asset they have. So even moderate housing price inflation which is not supported with corresponding increase of personal income can prevent people from buying additional real estate and make them chose other saving or consumption opportunities.

This conclusion can be supported with the positive sign of the estimated coefficient of current consumption (as it was mentioned before that it includes not only perishable goods but some durable goods except housing as well). Anyway, taking into account the fact that Consumption CAPM framework was used for utility function determination and unobservable parameter delta (risk-aversion parameter) was incorporated there it seems impossible to define the exact marginal rate of substitution of additional living space with amount of current consumption.

For now only the existence of significant positive relationship between housing prices and amount of current consumption can be ascertained. However assessment of the model on individual-level panel data would presumably allow estimation of the unobserved risk-aversion parameter and drawing more precise conclusions about the elasticity of current construction by housing prices.

When housing prices raise households make their choice in favor of more current consumption - for example, the individual may choose buying a new car over saving further in order to buy a new flat. This can be explained with high uncertainty about the future - there is a possibility that prices will go down - or inthe general case that expected return on housing can be considerably less compared to mortgage expenses, depreciation, adjustment costs, etc.

These costs of buying an additional living space relative housing return are reflected in the housing coefficient (ln_HC = ). As it was assumed when these expenses grew and were not backed with proportionate housing prices inflation households contracted their consumption of housing at that period, as a result the volume demanded on real estate market dropped and prices moderated further. Therefore it can be concluded that hypothesis H1 about negative significant influence of relative cost of buying housing on priceswas confirmed on Russian sample. However the influence of this indicator is much less compared to current consumption and housing stock per person: when relative costs go up by 1% housing prices will be diminished only by 0.17%.

The direct influence of time-specific and region-specific effects on housing prices appeared insignificant due to the fact that other variables that were included into the equation took over a quite big share of explained variance of prices. However because of endogeniety of housing stock variable time trend and GRP growth were used as instruments for endogenous variable and were proved to be valid and relevant. So their influence on housing prices was accounted for indirectly, therefore the hypothesis H2 could not be accepted unconditionally. It should be said that housing stock is mainly prone to cyclical adjustments and overall economic trends whereas it affects housing prices only through this variable.

At the same time the most influential driver of net residential real estate construction is housing prices themselves. The increase of prices by 1% will encourage constructors to build 0.65% more living spaces. The positive link between these indicators reflects the idea of the law of supply. The value of the coefficient indicates the fact that construction is not price elastic and it is worth mentioning that there are evidences that supply is inelastic in some other mainly developing regions.

Similar results were obtained in papers of (Green, Malpezzi and Mayo, 2005) who studied housing supply elasticity in large cities of different States of the USA. Authors found the evidence that in industrial states of the country elasticity of construction is significantly less compared to agricultural, technological and political centers of the country.

(Caldera and Johansson, 2013) tested cross-country sample for presence of sustainable differences between groups of countries that were combined according to a certain principle (geographical, economic development level, etc.). They found that in countries with many available residential construction land lots and weaker construction regulation price elasticity of housing supply is relatively lower.

The presumable reasons of some kind of insensitivity of construction activity to housing price dynamics are historical (the period under observation is long enough and captures several Soviet Union years, Perestroika years and further recovery of the market).

Real estate built in the Soviet Union was practically unified - within one region and between different regions there were almost no differences in construction style, so people had no choice but to live in standard apartments. It is also should be mentioned that many people that days lived in the halls of residence which were provided by government.

After the toughest part of the transitional period in Russian economy privately-owned companies started building up regions with constructions, which were distinguishable from Soviet style of housing construction in order to cover the free market share and fulfill appeared demand for better housing practically regardless price situation.

At that time the quality of construction became higher, buildings taller and placed with higher density in the most demanded parts of the regions due to the fact that market became competitive. And those people who could afford buying a new apartment created demand on primary market of real estate whereas those who could not afford a primary real estate could buy a flat on secondary market, which as a result pushed prices up.

This in turn encouraged more construction however by the time when price instead of market share became the matter regional market were already saturated to some extent and finding unsatisfied demand became the bigger problem. So after the period of construction boom prices also could be overshadowed by other factors such as for instance availability of suitable residential construction land lots, rising regulatory requirements and etc.

Among other factors that were proved to be important drivers of housing construction are construction costs. Within the framework of the developed theoretical model they were assumed to be consisted of labor costs which were approximated with average wage level and capital costs which included expenses for materials, machinery, design, etc. Both of these type of costs affect negatively construction activity which complies with common sense and economic theory. So the hypothesis H3 can be confirmed.

The value of the coefficients shows that more influential factor is capital cost inflation because when it goes up by 1% the prices will drop by 0.35% whereas the increase of workers' wage by 1% will diminish housing prices by only 0.085%. Despite the fact that wage is more statistically significant and rose more quickly compared to capital costs it should be noted that these costs account for more than 80% of the total cost of construction. Therefore even moderate inflation of their value can lead to considerable contraction of their operational marginality or to increase of the prices on primary real estate market and corresponding decrease of demand.

It should be noticed that as in the case of risk-aversion parameter that was included intodemand function the parameter alpha which reflects the marginal rate of substitution of capital costs by labor in total costs function is also unobservable. Due to the lack of company-level data on construction companies of each region about the structure of their expenditures this parameter was not estimated separately of the final coefficients of the structural model. This can be attributed to the shortcomings of the model.

Among variables that affect housing construction activity indirectly (through their influence on housing prices) the most influential factor is housing stock per person. This variable has significant negative influence on construction activity when the indicator rises by 1% net size of dwelling fall by 0.38%. Therefore it could be concluded that when housing market is insufficiently saturated with living spaces construction companies build up more actively in order to gain market share and cover potential demand. And vise-a-versa when most areas suitable for residential construction had been already built up companies moderate their activity or move it to other regions where market is relatively free.

The influence of current consumption on the net size of dwelling is positive and significant: when current consumption grows by 1% construction activity increases by 0.15%. As was observed earlier current consumption and housing prices are directly related - when housing prices inflate households postpone housing consumption and chose other consumption opportunities. So far it could be concluded that the higher level of current consumption means higher housing prices,and current situation on the market in its turn promotes formation of positive expectation by construction companies and by this stimulates construction activity as well.

Cost of buying residential real estate relative housing returns negatively influence net construction. The indirect effect of this indicator is the most moderate among all of the demand-side variables: when this ratio increases by 1% the net size of dwelling drops by only 0.11%. And this result seems to be pretty straightforward because this variable reflects relative cost of buying for households not for construction companies. Due to the fact that indicator negatively influences housing prices it acts in opposite way compared to current consumption and facilitates the formation of negative expectations about future prices and depresses housing construction activity.

All things considered it could be concluded that all the demand-side and supply-side factors that were included in to the theoretical model of Russian regional housing markets some way participated in equilibrium formation process. Besides due to the fact that the estimated structural model as a whole appeared to be significant all the paths of influence of all indicators can be believed as reliable and therefore used not only for further research but also for studying the regulatory effects on the market.

However except discussion of direct and indirect effects of different micro and macro indicators on housing demand and supply one of the most interesting implication of this research is that the model predicts equilibrium states of the system, because all the coefficients were extracted out of interaction between prices estimated within demand function and amount of construction dependent on these prices. The analysis of residuals of each equation of the model will allow concluding about was the equilibrium on housing market in Russian regions persistent at each moment of time in past sixteen years.

Due to the fact that the sample covers a really long period which includes at least two whole business cycles, two severe financial crises and a period of boom on housing market it is a question of special interest about the rationality of economic agents that moved prices that high or that low regarding their fundamentally justified level. It should be mentioned that by fundamental factors hereonly those factor included into the model were meant.

Firstly the analysis of supply equation residuals were calculated as a difference between empirical values of net size of dwelling and those estimated within the model. The figure 9 below represents supply equation residuals.

Fig.9. Residuals of supply equation.

One could noticed that even though residuals are highly dispersed for different regions which one more time supports the idea that regions are highly heterogeneous in their economic development they majorly do not have any trend in time and seem to be quite stable. There are almost no significant deviations from average value for each region and all the values lie in the close neighborhood of zero. This result tells that the model quite precisely described variation of net construction variable and data fits theoretical models pretty well. So it could be concluded that supply for most regions was driven by fundamental factors even during periods of economic instability and even the period of construction boom was consistent with rational assumptions theory.

Demand equation residuals were calculated the same way as supply equation residuals as a difference between empirical market data and forecasted within the model data. The graph of their dynamics over time is presented below.

Fig.10. Residuals of supply equation

This graph is much more interesting because residuals are volatile not only over different regions but also they are highly volatile in time. Taking into account deviations from zero there are obvious peaks and bottoms that reflect booms and bursts on housing market in Russia.

The first peak reflects the default of Russia in 1998 when there was hyperinflation and prices rocketed up very quickly but normalized within a year after that and even fall too much by the early 00's. The drop in real income of citizens and weak demand contributed to the drawdown of prices. Considering the start of construction boom which created the situation of oversupply on the housing market prices fall unexpectedly low and some sort of anti-bubble existed.

However along with economic recovery in the country the housing market recovered too.The following years were a period of rapid growth in many industries including construction itself and related to it. Real income of households increased and their savings and particularly investment in housing increased as well. So housing market reached its equilibrium in mid-00's but it persisted not long because the inflation of housing prices continued up to the crisis of 2008-2009.

According to the model and those fundamentals on which it is based there was a real estate bubble that time, because rapid growth of housing prices was not supported with the corresponding rise of real disposable income, drop of relative cost of buying additional housing or shortage of housing supply. Therefore it can be suggested that bubble was caused by irrational behavior of households that experienced some kind of money illusion. The link between money illusion and housing prices was established in the paper of (Shafir, Diamond and Tversky, 1997).

Overheated market could not persist for a long time and eventually in 2008 it burst and prices experienced serious drawdown compared to their peak values. The bubble had been over by 2010 in most regions. Several of them, presumably regions with lowest level of income per person even experienced an anti-bubble again. After 2010 prices stayed at their equilibrium level according to the developed model of supply and demand on regional housing market.

To sum up, during the period in observation Russian regional real estate experienced several bubbles and even anti-bubbles which implies that price dynamics during that time could not be explained with those factors that were included in the model. The source of these bubbles is presumably the irrational behavior of households.

Conclusion and discussion

regional housing market

All things considered quite satisfying results were obtained - the model of equilibrium on Russian regional housing market was constructed based on economically justified assumptions and was proved overall significant. Therefore it could be concluded that the aim of the study stated in the very beginning was reached. Besides, not only overall model appeared to be relevant but each equation and both demand-side factors and supply-side factors used for modeling equilibrium states are significant as well.

It was proved on empirical data covering a long period of time that housing prices are heavily dependent on the amount of living spaces available at the market, other alternatives of consumption or savings and historical performance of residential real estate as an asset class relative to the cost of buying it such as mortgage expenses, depreciation and alternative rate of return on financial assets.

However the initial indicator of housing demand - housing stock per person - was an endogenous variable presumably connected to the availability of spare residential land lots. It was instrumented with indicators of regional and national business cycle, whereas these variables were excluded out of the initial model estimation because the insignificant direct impact on housing prices. Anyway it was proved that they participate in housing price determination through their connection to housing construction activity.

This result one more time supports the idea presented by(Leamer, 2007) that housing market is highly connected with overall economic situation. Therefore it should be noted that in order to avoid the problem of endogeniety the additional indicators of residential land market need to be included into the modeling or a different than housing stock per person indicator should be used.

Construction activity in each region in its turn mainly orients on expected housing prices that were assumed to be based on observed current prices. A little bit less influential both by statistical significance and the absolute value of the regression coefficient are cost components: labor-related and capital-related. But at the same time the power of capital goods inflation influence is much higher compared to wage inflation. It can be explained by the fact that expenses of the construction companies on materials, machinery, design, etc. amount up to 80% in overall construction costs and even moderate inflation of these costs can have significant impact on marginality of the business. This result particularly supports the estimates conducted by (Gyorko and Saiz, 2006) who studied the influence of cost composition on housing supply.