Turkish Housing Market Dynamics: An Estimated DSGE Model

2.1 Households

Both patient and impatient households maximise

E o ∑ t = 0 ∞ ( β G c ) t z t ( Γ c ln ( c t − ε c t − 1 ) + j t ln h t − τ t 1 + η ( ( n c , t ) 1 + ξ + ( n h , t ) 1 + ξ ) 1 + η 1 + ξ )

(1)

Both decide consumption (c_t), housing (h_t) and hours worked in the consumption goods sector (n_ct) and hours worked in the housing sector (n_ht) to maximise their utility. The subjective discount factor is β; ε measures the consumption habit parameter; η stands for the inverse Frisch elasticity of labour supply; and ξ shows the degree of labour substitution between the sectors. The scaling factor Γ c = ( 1 − ε ) ( 1 − β ε ) guarantees the steady-state value of 1/c for the marginal utilities of consumption. The variables z_t, j_t and τ_t capture intertemporal preferences, the housing preference and labour supply shocks, respectively, and they are as follows:

ln z t = ρ z ln z t − 1 + u z , t

(2)

ln j t = ( 1 − ρ j ) ln j + ρ j ln j t − 1 + u j , t

(3)

ln τ t = ρ τ ln τ t − 1 + u τ , t

(4)

where ρ_i (i = τ or j) denotes the AR parameter; j denotes the housing preference (demand) shock; and u_i,t (i = τ or j) denotes an i.i.d. innovation with mean 0 and variance σ i 2 . Lifetime utility for patient households can be expressed as follows:

c t + k c , t A k , t + k h , t + k b , t + q t h t + p l , t l t − b t = w c , t X w c , t n c , t + w h , t X w h , t n h , t + ( R c , t z c , t + 1 − δ k c A k , t ) k c , t − 1 + ( R h , t z h , t + 1 − δ k h ) k h , t − 1 + p b , t k b , t − R t − 1 b t − 1 π π t + ( p l , t + R l , t ) l t − 1 + q t ( 1 − δ h ) h t − 1 + D i v t − φ t − a ( z c , t ) k c , t − 1 A k , t − a ( z h , t ) k h , t − 1

(5)

where k_c,t, k_h,t and k_b,t denote capital in the consumption and housing sectors and intermediate inputs in the housing sector. Also l_t, b_t, z_c,t and z_h,t represent land, borrowing and capital utilisation rates in the consumption and housing sector. Depreciation rates for capital in both sectors and housing are presented by δ _kc , δ _kh and δ _h , respectively. q_t denotes the house price and p_b,t the price of intermediate inputs. Real rental rates for capital are denoted by R_c,t, for the consumption sector and R_h,t for the housing sectors; R_t shows a riskless nominal return of interest rate; p_l,t and R_l,t denote land prices and the land rental rate; X_wc,t and X_wh,t denote wage markups charged by labour unions; Div _t indicates dividends gained from retailers and labour unions; ϕ_t denotes the adjustment cost parameter; and a(.) denotes utilisation cost of capital.

The impatient households maximise subject to the budget constraint:

c t ' + q t h t ' − b t ' = w c , t ' X w c , t ' n c , t ' + w h , t ' X w h , t ' n h , t ' + q t ( 1 − δ h ) h t − 1 ' − R t − 1 π π t b t − 1 ' + D i v t '

(6)

And in addition to the collateral (borrowing) constraint:

b t ' ≤ m E t ( q t + 1 h t ' π t + 1 R t )

(7)

Where m denotes the LTV ratio.

Equation (7) indicates that impatient households can only borrow the maximum loan ( b ′ t ) , which is up to the LTV ratio (m) times the expected present value of their homes.

On the production side, wholesale firms produce consumption goods, Y_t, in the consumption goods sector by using labour and capital and hiring intermediate goods from households. Also, new houses, IH_t, are produced by hiring intermediate inputs and land as well as production factor such as labor and capital in the housing sector. They try to solve the following:

max Y t X t + q t I H t − ( ∑ i = c , h w i , t n i , t + ∑ i = c , h w ′ i , t n ′ i , t + ∑ i = c , h R i , t z i , t k i , t − 1 + R I , t l t − 1 + p b , t k b , t )

(8)

The production technologies for goods and new houses are given by:

Y t = ( A c , t ( n c , t ) α ( n c , t ' ) 1 − α ) 1 − μ c ( z c , t k c , t − 1 ) μ c

(9)

I H t = ( A h , t ( n h , t ) α ( n h , t ' ) 1 − α ) 1 − μ h − μ b − μ t ( z h , t k h , t − 1 ) μ h k b , t μ b l t − 1 μ l

(10)

Here, A_ct and A_ht indicate productivity in the consumer goods (non-housing) sector and housing sectors, respectively; α shows patient households’ labour income share; and µ_c, µ_h, µ_b, and µ_l denote production function parameters.

In the model, the consumption sector exhibits price rigidity, while nominal wage rigidities are allowed for both in the consumption and housing sectors. Rigidities stem from labour unions and retailers, which affect the determination of wages and prices depending on market power. Employing the Calvo (1983) pricing mechanism, retailers sell differentiated goods to households by charging a markup, X_t. The Calvo (1983) pricing style enables 1 – θ_π retailers to set prices, while the remaining θ_π retailers cannot set prices in each period; hence, they have to index prices with an elasticity of ι_π, based on the past period inflation rate.

ln π t − ι π ln π t − 1 = β G C ( E t ln π t + 1 − ι π ln π t ) − ( 1 − θ π ) ( 1 − β G C θ π ) θ π ∏ π ln ( X t X ) + u p , t

(11)

X denotes the steady-state price markup; the cost shock with 0 mean and variance σ p 2 is presented by u_p,t. Wage-setting processes are also announced by labour unions similar to the above equation.

It is presumed that monetary policy follows the Taylor rule, which takes into account inflation and GDP growth, with interest rate smoothing to determine the nominal interest rate.

R t = r R R t − 1 + ( 1 − r R ) [ r π π t + r Y ( G D P t − G D P t − 1 ) + r r ─ ] + u R , t − s t

(12)

where G D P t = C t + I K t + q ¯ I H t ; IK _t and IH _t show business and housing investment; and q ¯ and r r ─ denote the steady-state house price and the steady-state real interest rate, respectively. u_R,t denotes an i.i.d. monetary policy shock; and s_t denotes a shock to the inflation target and is represented by the following equation:

ln s t = ρ s ln s t − 1 + u s , t , Ô ø Ω Ô ø Ω u s , t N ( 0 , σ s )

(13)

where ρ_s > 0.

The cleaning conditions for consumption, housing and loan markets are as follows:

C t + I K c , t A k , t + I K h , t + k b , t = Y t − φ t

(14)

H t − ( 1 − δ h ) H t − 1 = I H t

(15)

b t + b t ' = 0

(16)

where C t = c t + c ′ t shows aggregate consumption, H t = h t + h ′ t shows the aggregate stock of housing and IK _c,t = k_c,t – (1 – δ_kc)k_c,t–1 and IK _h,t = k_h,t – (1 – δ_kh)k_h,t–1 are two components of business investment.

3.1 Data

The model parameters are defined by calibration and Bayesian estimation techniques (see An & Schorfheide, 2007; Smets & Wouters, 2007; and for details of the estimation process). The model is estimated by employing Turkish quarterly data for 10 macroeconomic variables ⁴ from 2010Q1 to 2019Q4. The variables are per capita real consumption, per capita real residential investment, per capita real business fixed investment, real house prices, the nominal interest rate, inflation, real wages and hours in the non-housing sector and in the housing sector. We have to begin our sample period in 2010Q1 since this is the first official house price index from CBRT. Our data are drawn from the TURKSTAT, CBRT and TRLIBOR database. We use the Census X12 procedure to de-seasonalise all the variables. Data are detrended by using the Hodrick–Prescott (HP) filter. ⁵

3.2 Calibration

Table 1 lists the calibrated parameters. Some of these rely mainly on Iacoviello and Neri (2010) and Hloušek (2016) and other studies on the Turkish economy (Alp & Elekdağ, 2011; Bari & Şıklar, 2016; Çebi, 2012; Doğruel & Polat, 2015; Sekmen & Şıklar, 2016; Yüksel, 2013). The subjective discount rate for patient households (β) is set to 0.9950, which shows that the average annual real interest rate over the sample is at a steady state. We set (β'), at 0.97, which is the standard value in the DSGE literature (see Iacoviello & Neri 2010; Lee & Song 2015). The discount factor for impatient households, β', is set considerably below β, as the collateral constraint is constantly binding. In line with many studies on Turkey (Alp & Elekdağ, 2011; Doğruel & Polat, 2015; Yüksel, 2013), the markup on price and wages are set at X = X_wc = X_wh = 1.15, which implies a 15 per cent markup in the steady state. The weight on housing in the utility function is set at j = 0.12. The housing stock depreciation rate, δ_h, is chosen to be 0.0125, which reveals a 5 per cent annual depreciation of capital in the housing sector.

We also set the depreciation rates of capital δ_kc = 0.025 and δ_kh = 0.03 for the production and housing sector, respectively. We set µ_c = 0.40 for the capital share in the consumption goods production function parameter. Following Iacoviello and Neri (2010), we set µ_h = µ_b = µ_l = 0.1. One of the calibrated parameters which is important in terms of investigating the impact of monetary policy shocks on the housing market is the LTV ratio (m). We set m = 0.775, which is its average value in Turkey during the period examined. Finally, the correlation persistence of the inflation objective shock, p_s, is set at 0.975.

Table 2 reports the mean and distribution of the prior and posterior for all estimated parameters. For most priors, we follow similar studies (Doğruel & Polat, 2015; Hloušek, 2016; Iacoviello & Neri, 2010; Ng, 2015). The prior distribution of patient households’ income share parameter α is set at 0.65 with a standard deviation of 0.05, following a beta distribution. The prior mean for the habit parameters in consumption ε and ε' is set at 0.50, which is the same as Iacoviello and Neri (2010), Hloušek (2016) and Doğruel and Polat (2015). We impose the prior to the disutility of working η, which is set to 0.5 with a standard deviation of 0.1, in line with Iacoviello and Neri (2010). We assume an inverse elasticity of substitution across hours in the two sectors, following a normal distribution with a mean of 1 and a standard deviation of 0.1. We choose a gamma prior on the capital adjustment cost ϕ with a mean of 10 and standard deviation of 2.5. For the prior mean of the Taylor rule parameters, we impose restrictions on r_R, r_π, r_y of 0.75, 1.5 and 0.25 with a standard deviation of 0.1, respectively. The prior distribution of the Calvo price and wage parameters θ_π, θ_w,c, θ_w,h is beta with a mean of 0.5 and a standard deviation of 0.05. We assume a beta prior to the indexation parameters ι_π, ι_w,c, ι_w,h with a mean of 0.5 and a standard deviation of 0.2.

3.4 Posterior

The estimated posterior distributions of the structural parameters are shown in the right part of Table 2. It indicates that the priors are compatible with the data when considering the posterior means. Moreover, the posteriors indicate that most of the estimated parameters are similar to those in previous studies (e.g., Doğruel & Polat 2015; Hloušek, 2016; Walentin, 2014). The share of labour income for a patient household (α) is estimated to be 67.9 per cent, which also implies that the labour income for an impatient household is around 32 per cent. The values for habit formation in consumption, ε, show moderate degrees (0.28 and 0.45) in both patient and impatient households, respectively. Labour supply elasticity (η) is estimated to be 0.57 for patients and 0.51 for impatient households. The parameters ξ and ξ' are found to be 1.14 and 1.06, respectively, which implies that imperfect substitution occurs among hours in the two sectors. The adjustment cost parameters (ϕ_k,c, ϕ_k,h ) are estimated to be 19.3 and 10.1, respectively. The Taylor rule parameters are estimated as r_R = 0.71, r_π = 1.41 and r_Y = 0.19 and are slightly different from previous studies on Turkey. The estimated values of the Calvo price and wage parameter indicate that prices in Turkey are reoptimised around twice a year on average (θ_π = 0.64) and that the housing sector stickiness (θ_w,h = 0.48) is larger than that in the consumption sector (θ_w,c = 0.49).

The priors and posteriors for the shock process are presented in Table 3. We impose an inverse gamma prior to the standard deviation of each shock, with a mean of 0.001 and a standard deviation of 0.1. We set a beta distribution with a prior mean of 0.8 and a standard deviation of 0.1 for the AR (autoregressive) coefficients. It shows that the stochastic process for exogenous innovation is quite persistent. The shock coefficients vary between 0.61 and 0.97. The housing preference (ρ_AJ) and housing technology (ρ_AH) shocks are estimated to be highly persistent, at 0.97 and 0.93, respectively. These findings are similar to those in Iacoviello and Neri (2010), Walentin (2014) and Hloušek (2016). Monetary policy (ρ_AC) and business investment (ρ_AK) shocks are moderately persistent, with AR (1) coefficients of 0.70 and 0.82, respectively.

OPEN IN VIEWER

Table 3.

Shock Parameter Estimations

Parameters	Priors			Posteriors
	Distribution	Mean	SD	Mean	2.5%	97.5%
ρ AC	Beta	0.8	0.1	0.701	0.599	0.806
ρ AH	Beta	0.8	0.1	0.936	0.892	0.980
ρ AK	Beta	0.8	0.1	0.824	0.696	0.968
ρ AJ	Beta	0.8	0.1	0.978	0.960	0.996
ρ AZ	Beta	0.8	0.1	0.616	0.439	0.804
ρ AT	Beta	0.8	0.1	0.621	0.498	0.752
σc	Inv. Gam	0.001	0.01	0.039	0.030	0.046
σH	Inv. Gam	0.001	0.01	0.031	0.025	0.038
σk	Inv. Gam	0.001	0.01	0.023	0.017	0.028
σj	Inv. Gam	0.001	0.01	0.058	0.033	0.084
σR	Inv. Gam	0.001	0.01	0.056	0.004	0.007
σz	Inv. Gam	0.001	0.01	0.054	0.039	0.068
στ	Inv. Gam	0.001	0.01	0.082	0.061	0.102
σp	Inv. Gam	0.001	0.01	0.036	0.024	0.047
σs	Inv. Gam	0.001	0.01	0.001	0.001	0.003
σn,h	Inv. Gam	0.001	0.01	0.156	0.119	0.192
σw,h	Inv. Gam	0.001	0.01	0.024	0.019	0.029

Source: The authors.

4.1 Impulse Response Analysis

The behaviour of the model in response to shocks is investigated under different LTV values. There are three different specifications to observe the effect of macroprudential policies on the housing market. In the first, the LTV ratio is set to 0.775, which is called the benchmark specification. Secondly, the LTV ratio is set to 0.95, the high collateral specification, which allows impatient households more borrowing opportunities. Finally, to observe the implications on impatient households which are not allowed to access the financial market, LTV is set to 0 in the no collateral specification. In the analysis of macroprudential policies, a high LTV value of the housing market is generally considered procyclical and a tool that increases consumption and thus enhances output (Bouda & Formanek, 2014).

We conduct the impulse responses analysis under three different shocks, which are housing preference, housing technology and monetary policy shocks, to observe the behaviour of real consumption, residential investment, business investment, house prices, inflation, interest rate and output. All the values are expressed as the percentage deviation from the steady-state value. Note that a period in the impulse response graphs represents a quarterly period.

Housing Preference Shock

Figure 2 shows the impacts of a housing preference ⁶ (housing demand) shock on macroeconomic variables. As a result of the increase in housing demand, house prices go up. In all three specifications, housing demand shocks have a positive effect on house prices. This shock enhances the possibility of borrowing by households who want to use their houses as collateral, thus facilitating their borrowing and consumption under the high collateral scenario. Since consumption by impatient households is higher, this has a positive effect on total consumption, even if patient consumption contracts (Lee & Song, 2015). Also, while residential investment increases after a housing preference shock, business investment declines in the no collateral and high collateral specifications, since resources in the economy shift from business investment to housing investment in the short term. The increase in consumption and housing investment leads to an increase in economic output.

An increase in the value of the collateral as a result of the surge in housing prices triggers the consumption of impatient households (as in Doğruel & Polat, 2015; Iacoviello & Neri, 2010; Walentin, 2014) and creates an inflationary pressure in the economy, ⁷ causing the monetary authority to increase the interest rate. However, Figure 2 shows that the increase in the interest rate in the high collateral scenario is higher than in the other scenarios. As a result, the negative effect of the high interest rate on output is greater in the high collateral scenario. Consequently, housing demand shocks in the model create a procyclicality movement among the examined variables, owing to the collateral effect that strengthens the relationship between housing prices and consumption.

OPEN IN VIEWER

Figure 2.

Source: The authors.

Monetary Policy Shock

Figure 3 shows the responses to positive monetary policy shocks, which increase the nominal interest rate under the three scenarios. It indicates that the patterns of the impulse responses are very similar and a decline in all items of output (i.e., consumption and residential and business investment) and in house prices. The decrease in house prices following a contractionary monetary policy shock is in line with most of the empirical housing literature (Assenmacher-Wesche & Gerlach, 2008; Bjørnland & Jacobsen, 2010; Elbourne, 2008). Under the high collateral and benchmark scenarios, the consumption of patient households rises temporarily, but the consumption of impatient households decreases for all three scenarios, since the drop in house prices brings down their borrowing capacity. Another outcome of the monetary policy shock is that an increase in the interest rate causes the inflation rate to decrease by 0.5 per cent. This finding can be interpreted as indicative of the efficiency of monetary policy in controlling inflation.

OPEN IN VIEWER

Figure 3.

Source: The authors.

Moreover, an increase in the interest rate in Turkey leads to a reduction in consumption and investment, and therefore in economic output, which is consistent with the conventional monetary transmission mechanism in economics. The effect of monetary policy shocks in the three scenarios reveals striking results. After an increase in m from 0.775 to 0.95, the relative decline in inflation is less than the decrease in consumption and economic output. This reveals two outcomes. First, that monetary policy shocks work through the collateral channel when the effect of high LTVs is considered, and the effects of the shocks are further strengthened. Second, the response of real variables, such as consumption and output, is higher than the response of inflation. Based on these findings, it can be said that a contractionary (tight) monetary policy aimed at lowering inflation may lead to relatively larger decreases in real variables such as consumption and investment compared to inflation when the LTV ratio is high. The response of residential investment to monetary policy is negative. Wage rigidities in the housing sector play an important role in explaining this, as it makes housing investment very responsive to monetary policy shocks. As a result of higher interest rates (contractionary monetary policy), housing demand and house prices go down while wages in the housing sector remain constant. This leads to a fall in housing investment because of decreasing profitability.

Housing Technology Shock

Figure 4 shows the responses of nine variables to a housing technology shock. Since this shock relates to the marginal efficiency of producing housing, it is called as a housing-specific technology shock. Note that a positive shock to housing-specific technology will reduce the price of building houses. This shock in the housing sector leads to a considerable rise in residential investment. Since it represents housing supply, house prices fall due to technological progress over a long-term horizon. The consumption of impatient households, which enable the possibility of borrowing through the value of their housing, decreases as a result of a decrease in house prices. However, the decline in the consumption of impatient households has been the main driver of the decline in consumption even though consumption of patient households has slightly increased in the high collateral scenario. As expected, the nominal interest rate goes down and the inflation rate also falls in response to housing technology shocks. These findings are in line with those of Iacoviello and Neri (2010), Walentin (2014), Hloušek (2016) and Doğruel and Polat (2015).

OPEN IN VIEWER

Figure 4.

Source: The authors.

The impact of the technology shock in the housing sector on consumption shows that the collateral channel works effectively. While the technology shock decreases house prices and borrowing opportunities for impatient households, at the same time, it decreases inflation. For this, the response of the collateral effect on inflation and house prices is in the same direction; monetary policy, housing preference and housing technology shock are significant. Since the permanence parameter of the housing technology shock has a very high AR (1) autoregressive process, 0.93, it is observed that the effect of the shock is long.

The findings from variance decompositions are shown in Table 4, which indicates how much of the forecast error variance of each variable can be explained by exogenous shocks to other variables. There are nine exogenous shocks to the model economy: u_c —consumption goods shocks; u_e —monetary policy shocks; u_h —housing technology shocks; u_j —housing preference shocks; u_k —business investment shocks; u_z — intertemporal preference shocks; u_π—labour supply shocks; u_p—cost-push shocks; and u_s—inflation-targeting shocks.

In all periods, housing technology, housing preferences and labour supply shocks largely account for residential investment fluctuations. In the long term (20-quarter horizon), housing-related shocks explain 68.5 per cent of the variation in residential investment. In the very short term (1-quarter horizon), movements in house prices are largely explained by housing preference shocks (28.9%) and cost-push shocks (44.7%). The shock on consumption goods does not have any effect on the movement in house prices: in the very short term, it explains 9.9 per cent of the movement in house prices, while in the long term, it accounts for 7.5 per cent of variations in house prices. In the medium term (8-quarter horizon) and long term, the main contributors to house price fluctuations stem from house preference shocks, which accounts for 54.7 per cent of house price fluctuations for 8-quarters and 67.5 per cent for 20-quarters. These results indicate that the demand for housing is the predominant driving force of fluctuations in house prices in Turkey. ⁸ It is seen that consumption good shocks and cost shocks affect not only nominal variables, but also real variables (i.e., the GDP) in the very short term. However, the impact of the cost shock on GDP decreases in the medium term and long term, while the impact of the consumption goods shocks increases, especially on GDP.

Another result is that the effect of a monetary policy shock on movements in house prices is limited, compared to a housing demand shock. This shows that the variation on consumption is determined by intertemporal preference shocks (36.2%), cost-push shocks (21.1%) and consumption goods shocks (19.1%) over the long-term period. Fluctuations in the GDP are not well explained by housing preference shocks, but approximately 90 per cent of GDP is explained by the shocks in the consumption goods sector (20.1%), business investment (21.5%), cost-push (25.5%) and labour supply (23.1%). At the end of 20-quarters, both housing technology (2.3%) and housing preference (1.6%) shocks account for less than 4 per cent of the variance in GDP and less than 1 per cent of consumption and inflation variations, respectively. The variance decomposition results reveal that the shock from labour supply is one of the most important shocks accounting for variations in consumption, investment and GDP. This shows that Turkey’s economy, with its abundant labour force, and despite the drawbacks of technology, has for many years depended on developments in the highly labour-intensive industry structure. Therefore, a change in the supply of labour has a significant impact on the entire economy. Also, it is observed that cost-push shocks are an important driver of business cycle fluctuations in Turkey, whereas the central bank’s inflation-targeting shocks do not play any significant role.

The variance decomposition analysis (Table 5) reveals the amplification mechanism of the LTV ratio under three different collateral constraints (i.e., high, benchmark and no-collateral) on consumption, output and inflation, by considering the monetary policy shock. We can easily compare these variance decomposition results to those under the no-collateral, benchmark and high collateral scenarios. In the case of the high collateral scenario, the effect of monetary policy shocks, mainly on consumption (more than 2 times), is higher than on inflation and output. In the case where impatient households are excluded from the loan market (m = 0), fluctuations in the variables are moderately explained by monetary policy shocks. However, there is little difference in the benchmark situation. This indicates that the high collateral scenario considerably amplifies the effect of the monetary policy shock. This amplification is strongest in consumption, followed by inflation and then output, which is in line with findings for other countries (Hloušek, 2016; Walentin, 2014). Therefore, using the LTV ratio as an efficient macro prudential tool to reduce the number of adverse effects that arise from housing policy may stabilise the fluctuation of business cycles (Bruneau et al., 2018).

OPEN IN VIEWER

Table 5.

The Importance of Collateral Constraint: The Monetary Policy Shock for Variance Decomposition (%)

Scenario	Consumption	Output (GDP)	Inflation
High collateral (m = 0.95)	10.4	5.32	7.46
Benchmark (m = 0.775)	4.28	3.97	4.26
No collateral (m = 0)	2.75	4.14	2.75

Source: The authors.

Similarly, housing demand shocks have the same effect on the variables (Table 6). The analysis shows that the spillover effect from the housing sector to the other side of the economy is not insignificant, and that the determination of the LTV ratio is important for the rest of the economy. The impact of the housing preference shock is very high, particularly in terms of consumption and inflation, in the case of the high collateral scenario. With a higher positive preference shock, real house prices rise, and this leads to an increase in the collateral capacity of impatient households which in turn boosts borrowing and consumption.

OPEN IN VIEWER

Table 6.

Importance of Collateral Constraint: The Housing Preference Shock for Variance Decomposition (%)

Scenario	Consumption	Output (GDP)	Inflation
High collateral (m = 0.95)	5.91	4.85	1.85
Benchmark (m = 0.775)	0.30	2.67	0.21
No collateral (m = 0)	0.06	2.72	0.02

Source: The authors.

Figures 5 and 6 depict the historical shock decomposition of real house prices, residential investment and GDP according to the structural shocks in the model over the period 2010Q1 to 2019Q4. In line with the variance decomposition results, the historical shock decomposition indicates that housing preference shocks are the main drivers of real house prices in Turkey. Since 2013, housing preference shocks have contributed most to changes in house prices, followed by consumption goods sector shocks. The historical shock decomposition further suggests that the contribution of monetary policy has become more important in recent years. The results of the historical shock decomposition of housing prices are similar to those noted by Iacoviello and Neri (2010), contrary to Walentin (2014) which shows that that real housing price fluctuations are negatively affected by monetary policy shocks for Sweden. Further, Ng (2015) and Lee and Song (2015) find that housing preference shocks play a key role in house prices for China and Korea, respectively.

OPEN IN VIEWER

Figure 5.

Source: The authors.

Although housing investments have generally followed an increasing trend in the period 2010–2019, there has been a rapid contraction after 2018 (Figure 6). Shocks from housing technology and housing preferences almost equally account for the fluctuations in residential investment after 2014, as indicated by the results of the variance decomposition. Housing technology shocks were almost the only determinant factor until 2013, except for weak impacts from shocks in the consumer goods sector and monetary policy. After 2013, housing preference shocks play an increasing role in explaining the variance in housing investments. Similar to our findings, Ng (2015) finds that housing preference and housing technology shocks are equally important for residential investments, as do Doğruel and Polat (2015) who suggest that housing technology shocks have a dominant effect on residential investments for the period 2010 to 2014 in Turkey.

OPEN IN VIEWER

Figure 6.

Source: The authors.

In this section, we evaluate the effect of traditional monetary policy on helping to stabilise house prices. For this, we extend the Taylor rule to the case in which CBRT responds to fluctuations in house prices. In literature, there is no consensus on whether central banks should respond to housing prices or not. However, in recent years, there have been several studies that analyse the extended Taylor rule by incorporating asset and house prices in the central bank’s response function (Aspachs-Bracons & Rabanal, 2011; Finocchiaro & Heideken, 2013; Fuhrer & Tootell, 2008; Gupta & Sun, 2020; Kannan et al., 2012; Notarpietro & Siviero, 2015; Sun & Tsang, 2014). However, so far, there has been no study which takes into account house prices in Turkey.

Equation (17) is an extension of Equation (12) with house prices, which represents the standard monetary policy according to the Taylor rule; nq_t and r_q denote nominal house prices and the responsiveness of changes in house prices in the reaction function of the central bank, respectively.

R t = r R R t − 1 + ( 1 − r R ) [ r π π t + r Y ( G D P t − G D P t − 1 ) + r r ─ + r q ( n q t − n q t − 1 ) ] + u R , t − s t

(17)

Table 7 shows the estimation results of the parameters in the extended Taylor rule, in which only the parameters of the Taylor equation are altered by keeping all other parameters in the model constant. We set 0 prior means with 0.1 standard deviations for r_q. Such an assignment is consistent to estimate the model to understand the importance of the coefficient.

Table 7 shows the mean and standard deviation of the prior and posterior distributions, both for the benchmark Taylor rule model [in which the central bank does not pay attention to house prices (r_q = 0) and the extended Taylor rule model (r_q 0) in Turkey. In the extended model where the interest rate reacts to house price inflation, the posterior mean of r_q is 0.13. It can be inferred from our main result that house price fluctuations do not play a substantial role in CBRT’s reaction function over the sample period as expected. CBRT has reacted less aggressively to increases in house prices than to inflation (1.43) and output fluctuations (0.17). The posterior means of the other Taylor rule parameters (r_R, r_π, r_Y) are similar to the benchmark Taylor rule models, which show the robustness of the two models estimated.

OPEN IN VIEWER

Table 7.

Priors and Posteriors of the Extended Taylor Rule

Parameter	Prior			Posterior rq = 0			Posterior rq ¹ 0
	Dist.	Mean	S.D.	Mean	% 2.5	% 97.5	Mean	% 2.5	% 97.5
rR	Beta	0.75	0.1	0.71	0.65	0.77	0.69	0.62	0.75
rπ	Normal	1.50	0.1	1.41	1.26	1.58	1.43	1.27	1.59
rY	Normal	0.25	0.1	0.19	0.06	0.31	0.17	0.04	0.28
rq	Normal	0	0.1				0.13	0.03	0.23

Source: The authors.

Unfortunately, we cannot compare our results with the other studies that consider Taylor rules with house prices in Turkey. However, the parameters for the US Fed, the Bank of England (BoE) and Bank of Japan (BoJ) are found to be 0.36, 0.16 and 0.26, respectively, in Finocchiaro and Heideken (2013). Recently, Gupta and Sun (2020) found that the house price responsiveness parameter for South Africa was 0.58. When we compare that parameter to the other important central bank’s reaction functions, we see that the CBRT’s response parameter (r_q) to house price growth is lower than that for the FED, BoJ and South African Reserve Bank. Our findings make a significant contribution to the empirical literature on estimating DSGE models for all developing countries, including Turkey. Moreover, they enable us to examine the shocks behind housing market dynamics in those countries.

Table 8 presents the unconditional variance decomposition results of the model estimated by the extended Taylor rule. When we consider the variance decomposition findings, which are very similar to the results obtained in the previous part of the study, 62.2 per cent of the fluctuations in house prices are explained by housing demand shocks. The housing preference (30.7%) and housing technology shocks (41.4%) account for about 72.1 per cent of the variations in residential investment. The variation in GDP is mostly explained by consumption goods (19.1%), labour supply (20.8%), cost-push (25.8) and business investment (23.2%) shocks.

OPEN IN VIEWER

Table 8.

Variance Decomposition of the Extended Taylor Rule

	uc	ue	uh	uj	uk	up	us	ut	uz
Consumption	20.7	3.3	0.1	0.1	8.1	26.7	0.1	17.1	23.8
Residential investment	0.3	1.4	41.4	30.7	0.1	1.8	0.1	18.3	5.9
Business investment	13.8	2.5	0.1	0.5	42.1	19.2	0.1	10.4	11.4
House prices	6.7	1.3	9.8	62.2	1.9	9.1	0.1	3.9	5.1
Interest rate	17.3	6.4	0.1	2.1	0.7	38.4	0.1	12.8	22.1
GDP	19.1	3.6	2.1	2.2	23.2	25.8	0.1	20.8	3.1

Source: The authors.