Commodity cycles and financial instability in emerging economies

Abstract

Commodity-exporting economies display procyclicality with the price of commodity exports. However, the evidence for the relative importance of commodity price shocks for aggregate fluctuations remains inconclusive. Using Russian data from 2001 to 2018 we estimate a small open economy New Keynesian model with a banking system and leveraged domestic firms who default on their unsecured domestic debt. We show that allowing default rates to vary endogenously over the business cycle amplifies the estimated contribution of commodity price shocks. Endogenous default introduces time-varying wedges that amplify the response of commodity price shocks through demand and income effects rather than the relative price effects that are found in the country risk-premium, balance sheet, and financial accelerator channels. We find that the contribution of commodity prices to explaining fluctuations in GDP rises from 2.5 to 33.6% while for deposits and non-performing loans, it increases from 5.3% and 1.6% to 71.3% and 60.4%, respectively.

1 Introduction

The significant fluctuations in emerging economies’ aggregate performance have spurred various theories regarding their root causes. These theories encompass terms of trade and price shocks (Mendoza 1995; Kose 2002), commodity price shocks (Fornero et al. 2016; Fernández et al. 2017; Bergholt et al. 2017; Fernández et al. 2018), productivity shocks (Aguiar and Gopinath 2007; Garcìa-Cicco et al. 2010), and the impact of financial frictions and interest-rate premia (Neumeyer and Perri 2005; Uribe and Yue 2006; Chang and Fernández 2013; Fernández and Gulan 2015). A recent strand of literature focuses on quantifying the influence of commodity prices through financial conditions, exemplified by Drechsel and Tenreyro (2018), who found that commodity price shocks account for 38% of output variation.

This paper investigates the extent to which emerging market business cycles are influenced by commodity prices. Utilizing Russian data from 2001 to 2018, we demonstrate a significant increase in the proportion of output variation attributable to commodity price shocks when financial frictions are integrated into the model’s endogenous structure. With these frictions included, commodity price shocks (and total factor productivity shocks) explain 34% (58%) of output variation, compared to 2% (64%) when excluded. Additionally, the role of investment shocks in explaining GDP, Loans, and Deposits diminishes markedly from 17%, 67%, and 47% to 0%, 16%, and 6% with financial frictions. We also observe a strong ’Dutch Disease’ effect in the Russian economy following a commodity price shock.

Our study of the Russian economy offers dual benefits: Firstly, it has experienced significant economic fluctuations over the past two decades, primarily due to substantial declines in its primary export, oil/gas commodities. Secondly, unlike Latin American emerging economies, Russia maintains a current account surplus, low external debt, and a diverse range of trading partners despite its commodity-focused exports. Middle East And African commodity exporters including Saudi Arabia, Kuwait and Algeria feature similar characteristics over the period 2001–2018, running current account surpluses and being net lenders. Among the CIS countries Uzbekistan is the closest comparison, while Azerbaijan remained a net borrower despite running current account surpluse over 2001–2018.¹ This necessitates a greater dependence on internal mechanisms for increasing real interest rates, achieved here through financial frictions between banking and production sectors.

To understand the interplay between commodity shocks and financial frictions, we estimate a New Keynesian DSGE model for a small, open, commodity-exporting economy, augmented with banking and firm sectors. Firms in the model face idiosyncratic risks and banks adhere to capital requirements, with firm default rates varying endogenously over the business cycle. Our work contributes to three key areas of literature: the role of commodity shocks in emerging market business cycles, the identification of mechanisms for structural shock propagation and amplification. Our findings, related to those of Drechsel and Tenreyro (2018) and Fernández et al. (2018), indicate that oil price shocks dampen domestic demand, increase default expectations, and elevate interest rates. Unlike Drechsel and Tenreyro (2018), we maintain a constant external interest-rate premium and show that the impact of negative foreign oil price shocks is primarily through domestic income effects. Our approach is also closely related to Shousha (2016) who builds a DSGE model of a commodity exporter and examines the relative importance of the working capital, financial accelerator, balance sheet mismatch, and country interest rate channels. Shousha finds that the banking sector is not important as the financial accelerator and balance sheet mismatch channels do not significantly affect impulse responses. On the other hand, the country interest rate and the working capital channels are found to be important. Our result that the financial sector is important is through the default channel that Shousha (2016) misses. Default introduces time-varying wedges that amplify the response of commodity price shocks through demand and income effects rather than only through the relative price effects that are found in the balance sheet and financial accelerator channels of Shousha (2016).

The time-varying wedge that default generates in our model gives rise to cyclicality in the effective cost of debt for borrowers, and through equating this to the marginal product of capital, to investment. Higher default rates drive up the cost of debt and cause investment to decrease. In the sense that negative shocks start this process, our mechanism is similar to that of Martins et al. (2011). In that paper though, the mechanism is dependent on foreign interest rate shocks, whereas our channel through default works through commodity prices as well as other shocks.

Chari et al. (2007) suggest that business cycles can be conceptualized as wedges in the Real Business Cycle model’s endogenous structure. We demonstrate that ’investment wedges’, particularly inefficiencies from domestic firm defaults to banks, are crucial in identifying structural shock impacts. When these wedges are constant, or absent, the transmission of foreign shocks (specifically to foreign oil prices) is undervalued in estimations compared to when financial frictions are considered. This aligns with Justiniano et al. (2010), who emphasized the role of investment shocks in GDP fluctuations but speculate that they may be a proxy for unmodelled financial frictions. Our model, featuring endogenous default, better fits by linking loan dynamics to default rates. Post-oil price shocks, the model predicts a depreciation in the exchange rate, inflation, demand contraction, and a spike in default rates and borrowing costs, leading to a sharp decrease in firm investment and a gradual decline in secured loans due to falling collateral values.

The role of corporate unsecured lending as a significant risk factor in emerging markets has been underscored in studies such as those by Fernández and Gulan (2015), Chang et al. (2017), and Caballero et al. (2018), particularly in relation to the countercyclicality of interest rates and leverage. Our focus extends to domestic firm defaults to local banks in domestic currency, an aspect often overlooked yet critical in these markets. For example, Table 6 in the Appendix shows that uncollateralized debt accounted for over 50% of loans of two major Russian banks. The work of Daniel (2011) reinforces this perspective by highlighting the inherently risky nature of loans in emerging markets, characterized by their volatile access and fluctuating interest rates. This underlines the pivotal role played by domestic firm defaults to local banks, conducted in the domestic currency, in shaping financial stability (Daniel 2011).

Furthermore, the research by Durbin and Ng (2005) delves into the investors’ perception of country risk as observed through the spreads of emerging market bonds in secondary markets. This perspective is invaluable in understanding the market’s assessment of risk, especially in relation to domestic firm defaults to local banks in domestic currency. It provides a nuanced view of how such defaults are perceived and factored into the broader context of emerging market dynamics (Durbin and Ng 2005).

2 Commodity cycles and empirical regularities

2.1 Data

For estimating model parameters and conducting historical decomposition, we utilized eight data series: GDP, household consumption, CPI (Consumer Price Index), interest rates, total loans, the ratio of non-performing loans to total loans, deposits, and the international oil price. We sourced the data on GDP, consumption, and inflation from Rosstat. The series for the international oil price is based on the Urals oil price, expressed in dollars per barrel. We obtained data for all other series from the Bank of Russia website. Specifically, the total loans series represents loans issued by Russian banks to domestic enterprises. To calculate the amount of non-performing loans, we used the proportion of non-performing loans from the total loans granted by Russian banks to domestic enterprises. We constructed the series for the ratio of non-performing loans to total loans by dividing the non-performing loans by the total loans issued by Russian banks to domestic enterprises. For deposit data, we used the stock of household deposits denominated in domestic currency. The interest rate series we used was the one-day interbank rate in Russia.²

We focus on oil prices as a critical indicator of Russia’s total exports for two main reasons. First, from 2001 to 2018, crude oil was Russia’s foremost export, accounting for approximately 31% of the total export value. Second, international oil prices have a strong correlation with the prices of two other major Russian exports: natural gas and petroleum products. The correlation is notably high, with oil prices and petroleum product prices being 99%, and a 92% correlation between oil prices and natural gas prices. These correlations are based on seasonally adjusted and deflated series, using average USD export prices for these commodities in Russia, with data sourced from the Bank of Russia website. Combined, exports of crude oil, petroleum products, and natural gas represented about 62% of Russia’s total exports during this period. We chose not to include foreign interest rates among our list of observable factors because they are not the primary external influence on Russia’s economic fluctuations. During our study period, Russia maintained a consistent current account surplus, averaging around 5.8% of GDP. The average net international investment position, as per data from the Bank of Russia website, stood at 8.6% relative to GDP from 2001 to 2018, indicating that Russia’s economy was predominantly a net lender. Hence, fluctuations in foreign interest rates were unlikely to be the primary drivers of economic fluctuations in Russia. Additionally, as demonstrated in Table 1, oil prices exhibit a stronger correlation with key economic indicators than foreign interest rates do.

The dataset spans 70 quarters, from the first quarter of 2001 to the second quarter of 2018. We chose the beginning of 2001 as the start of our analysis because, by this time, the impact of the 1998 crisis on the Russian economy had dissipated, and the economy was beginning to benefit from rising international oil prices. To refine the data, we removed the seasonal component. The Gross Domestic Product (GDP), household consumption, total loans, deposits, and international oil prices are presented as seasonally adjusted real figures, using the fourth quarter of 2013 as the base period. The interest rate, Consumer Price Index (CPI), and the ratio of Non-Performing Loans (NPL) to total loans are shown as seasonally adjusted values. Additionally, we calculated the growth rates for aggregate variables and oil prices. This adjustment limits the dataset used for calculating and estimating business cycle moments to the period from the second quarter of 2001 to the second quarter of 2018 (Fig. 1).

Table 1 Business cycle moments Q2 2001–Q2 2018

Fig. 1

Macroeconomic and Financial time series used in estimation

2.2 Business cycle moments

Table 1 represents the key business cycle moments of Russian economy. It summarizes statistics on cross-correlation of GDP growth, consumption growth, oil price growth, real loans growth, real deposits growth, percentage deviations of ratio of NPL to Loans, percentage deviations of annual CPI, percentage deviations of annual domestic interest rate and percentage deviations of annual foreign interest rate. The results indicate that there is a high correlation between consumption and GDP, which corresponds to the correlation of these variables in advanced economies.

The important feature of the Russian business cycle is high correlation between GDP growth and oil price growth as well as between consumption growth and oil price growth. We also observe that there is high correlation between the growth of GDP and real loans as well as real deposits, while annual interest rate and GDP growth are negatively correlated. Another striking feature of the business cycle statistics of Russian economy is strong negative correlation between the growth of GDP and ratio of NPL to Loans. Among others we see that there is negative correlation between oil price and ratio of NPL to Loans, while oil price growth is positively correlated with the growth of real deposits.

Overall, we observe that the dynamics of the variables that represent financial cycle (loans, deposits, NPL to Loans) are strongly correlated with the dynamics of GDP, while the later is positively correlated with the oil price.

2.3 Unsecured credit and loans

The importance of unsecured credit in Russia is reflected in the importance of credit lines as a source of liquidity to firms and loans to early-stage firms who have limited collateral. Table 6 in the Appendix displays point estimates for different types of loans. According to this partial data³ only 17–18% of corporate loans have real estate as collateral. 56–75% of loans are uncollateralized or have financial collateral. The importance of “risky” borrowers in evaluating financial stability was central to the policy debate in the US following the crisis of 2007–08. Aikman et al. (2019) describe how the aggregate loan-to-value ratio on mortgages remained stable in the years leading up to the US crisis, but there was an increase in the concentration of debt among riskier borrowers. build-up in debt was being concentrated at riskier, heavily indebted borrowers was not being adequately picked up (see Eichner et al. (2015)). Unfortunately aggregate statistics on secured versus unsecured credit is not available but we can infer the role that it plays through proxies.

In Fig. 2a we decompose loan origination across different types of borrowers. We posit that ‘Big firms’ have the ability to pledge physical capital while other types of commercial borrowers cannot.⁴ In the crisis period following December 2014 there are sharp declines in new loans originations in all categories except the loans to large firms. In Fig. 2b we look at the ratio of non-performing loans across borrower types. The large reduction in loans to small and medium firms (SME), and individual entrepreneurs is coincident with a sharp rise in non-performing loan rates. This tells us that these loan types are more similar than others in both the sensitivity to the business cycle and delinquency rates.

Fig. 2

Loan origination and share of non-performing loans. Note: The figure on the left shows year-on-year growth of loan origination (not seasonally adjusted). In the figure on the right, ‘Big’ firms include corporate loans to non-financial and financial organizations of all types of ownership. Financial organizations do not include banks and other credit granting institutions

Table 2 Regressions of non-performing loans on macro variables

We use these stylized facts to motivate the construction of our model where we emphasize the role of unsecured firm credit.

Table 2 is a regression of non-performing loans (NPLs) on deterended macroeconomic variables. We can see that there is a direct correlation between the price of Urals and NPLs but the correlation disappears when GDP, consumption or loans are included. This motivates our reasoning that the endogenous structure of our model should account for non-performing loans because of the way it endogenously interacts with standard macroeconomic variables. The data is on a quarterly frequency from Q2 2001 to Q2 2018 and is calculated as the cyclical component divided by the trend.

2.4 Summary of stylized facts

The stylized facts can be summarized as: 1. Strong correlation of output and consumption. 2. Strong correlation of consumption and output with oil price. 3. Excess volatility of consumption over output. 4. Negative correlation between GDP growth and NPLs. 5. Strong positive relation between GDP growth and loans. 6. Negative correlation of GDP growth and interest rates. We now turn to our model.

3 A NK small open economy model with a banking sector

We now present our small open economy NK model, developed along the lines of Galí and Monacelli (2005) and Gertler et al. (2007) among others. While otherwise standard, our model has two distinguishing features: an explicit optimizing banking sector and the way in which we model endogenous default. Our approach is closest to Martìnez et al. (2020) and Garcès et al. (2023) who estimate a similar model (but with a representative household and heterogeneity of banks) based on Chilean data to study the effects of countercyclical capital buffers and the effectiveness of pandemic economic policies receptively, and for Russia Andreev et al. (2019) that performs macroprudential analysis in a calibrated model with a representative banking and household sector.

Details of the intermediate goods producers, domestically-priced final goods producers (or retails), and capital producers can be found in the Online Appendix.

3.1 Circular flow of funds

Wholesale firms operate over two periods. In the first period, they use their available funds to invest in physical capital. These funds consist of equity from households and both secured and unsecured debt. In the second period, they utilize this capital and labor to produce intermediate non-tradable goods. During this period, firms have the option to default on a portion of their unsecured debt. The amount of secured debt they can take on is limited by collateral requirements.

Capital producers, in their production process, use imported intermediate goods, undepreciated capital, and domestic final goods. The government, owning oil reserves, gains all revenue generated from oil. Banks combine household deposits with their own equity to buy debt issued by wholesale producers, adhering to capital adequacy requirements set by the Monetary Authority.

Inifinitely lived Households own various entities. These include capital producers, non-tradable goods producers, banks, and other firms, except oil producers. Their savings involve bank deposits and investments in domestic and foreign bonds. The Monetary Authority sets the nominal interest rate on domestic bonds, while the Fiscal Authority uses its revenue to purchase non-tradable and imported goods. The circular flow of funds is summarized in Fig. 3.

Fig. 3

Circular flows diagram

3.2 Households

There is a continuum of households who are infinitely lived. They consume both domestically produced goods ($c_{N,t}$) and imported goods ($c_{T,t}$), deriving utility from their overall consumption bundle ($c_t$). The domestic price of imported goods is denoted as $p^{imp}_t$. Households experience disutility from labor ($l^{h}_t$) but receive wages ($w_t$) for their work, which they choose. Additionally, households own all firms in the economy-this includes wholesale and intermediate producers, retailers, and capital producers-as well as banks, with the exception of the oil producer, which is government-owned. They receive profits from these entities.

Households invest in banks and wholesale producers using equity, represented as $e^{bank}_t$ and $e^{w, total}_t$ respectively. The equity for wholesale producers consists of net equity ($e^w_t$) and undepreciated capital received from firms that shut down in the current period ($(1-\tau )p^K_t k^w_t$). This latter component emerges due to the Overlapping Generations (OLG) structure of firms that we employ in our model. Households can save through various means: by making deposits ($d^{h}_{t+1}$), investing in foreign bonds ($B^f_{t+1}$), and purchasing domestic government bonds ($B^{g,h}_{t+1}$). The consumption bundle is:

$$\begin{aligned} c_t = A^c \left[ (\phi ^{h})^{\frac{1}{\nu _c}} c_{N,t}^{\frac{\nu _c - 1}{\nu _c}} + (1-\phi ^{h})^{\frac{1}{\nu _c}} c_{T,t}^{\frac{\nu _c - 1}{\nu _c}}\right] ^{\frac{\nu _c}{\nu _c -1}}, \end{aligned}$$

(1)

where $\nu _c$ is the elasticity of substitution between domestic and foreign goods. Budget Constraint of a Household:

$$\begin{aligned}{} & {} {d}^{h}_{t+1} + p^{imp}_t c_{T,t} + c_{N,t} +e^{w,total}_t +e_t^{bank} + Q_t B^{f}_{t+1} + B^{g,h}_{t+1} \nonumber \\{} & {} \le {(1+r^d_t)}{{d}_{t}^{h}} + Q_t B^{f}_{t} (1+r^{f}_t) + B^{g,h}_{t} (1+ r^b_t) + w_{t} l_{t}^{h} + (1-\theta ^w) \bar{\Pi }_{t}^{w} + \theta ^w \underline{\Pi }_{t}^{w} \nonumber \\{} & {} \quad + {\Pi }_{t}^{bank} +\Pi _t^{cap}+\Pi _t^{ret} +\Pi _t^{exp} - adj^h_t \end{aligned}$$

(2)

where $Q_t$ is an exchange rate, $e^{w, total}_t = (e^{w}_{t}+(1-\tau )p^K_{t} k^w_{t})$, and $adj^h_t$ are the adjustment costs of household.⁵ Households maximize their discounted utility s.t. their BC:

$$\begin{aligned}{} & {} \max _{c^{h,imp}_t, c^{h,N}_t, e^{w,total}_t, e_t^{bank}, {d}^{h}_{t+1}, w_{t}, B^f_{t+1}, B^{g,h}_{t+1}} {\frac{(c_0^{h})^{1-\sigma }}{1-\sigma }-\theta ^{h}\frac{(l^{h}_0)^{1+\gamma ^{h}}}{1+ \gamma ^{h}}} \\{} & {} \quad \qquad + \sum _{t=1}^{\infty }(\beta ^{h}_{t-1})^t \left[ {\frac{(c_t^{h})^{1-\sigma }}{1-\sigma }-\theta ^{h}\frac{(l^{h}_t)^{1+\gamma ^{h}}}{1+ \gamma ^{h}}}\right] \end{aligned}$$

Households offer their labor in a monopolistically competitive market, where they can optimally choose their wages. These wages may be subject to future revisions with a probability of $1-\theta ^{pw}$. This construct of nominal wage rigidity leads to the labor supply adjusting to meet demand, similar to how firm output reacts to demand in the presence of nominal price stickiness. The demand for an individual’s labor becomes a function of the overall labor demand, the aggregate wage, and the individual’s wage. The specifics of this labor supply and wage-setting problem are standard and further details can be found in the Online Appendix.

3.3 Wholesale producer

Wholesale producers have a lifespan of two periods. In their first period, all firms are identical and each firm obtains equity from households (HH) and issues both secured ($\mu ^{w,s}_{t+1}$) and unsecured ($\mu ^{w,u}_{t+1}$) debt to banks. This funding is used to purchase capital ($k^w_{t+1}$) at the price $p^K_t$. In the second period, each firm realizes its productivity level ($A_t$), which can be either high ($\bar{A_t}$) or low ($\underline{A_t}$). Depending on their productivity, firms decide the amount of labor ($l^w_t$) they wish to employ. We assume that a portion of firms ($1-\theta _w$) are “lucky” and achieve high productivity, while the remaining firms ($\theta _w$) are “unlucky” and experience low productivity. Thus, firms are identical before realization but differ afterwards.

When firms opt for secured borrowing, they are bound by a collateral constraint. This constraint limits the repayment amount to not exceed the expected value of the firm’s undepreciated capital in the next period. This expected value is calculated using a collateral discount ($coll$). Firms also have the option to default on a part of their unsecured debt. The rate at which they default ($\delta ^w_{t}$) is referred to as the ‘loss given default.’

The total production is given by a constant returns to scale production function:

$$\begin{aligned} y_{t}^{j}=A_{t}^j (k_{t}^{j})^{\alpha }(l_{t}^{j})^{1-\alpha }. \end{aligned}$$

(3)

The first period budget constraint of a firm, in real terms, takes the form:

$$\begin{aligned} \begin{aligned} p_{t}^{K} k^{w}_{t+1}+T^w + adj^{w}_t = {\mu _{t+1}^{w}}+ e^{w,total}_t, \end{aligned} \end{aligned}$$

(4)

where total debt issued is $\mu ^w_{t+1} = \mu ^{w,u}_{t+1} + \mu ^{w,s}_{t+1},$ and $adj^{w}_t$ is the adjustment costs of the firm.⁶

Collateral constraint of a firm takes the form:

$$\begin{aligned} {{\,\mathrm{\mathbb {E}}\,}}(1+r^{w, s}_{t+1}) {\mu }^{w, s}_{t+1} \le coll (1-\tau ) k_{t+1}^w {{\,\mathrm{\mathbb {E}}\,}}p_{t+1}^K. \end{aligned}$$

(5)

In the second period a firm receives profit:

$$\begin{aligned} \Pi _{t+1}^w =&p_{t+1}^{w} A_{t+1}^w (k_{t+1}^w)^{\alpha }(l_{t+1}^w)^{1-\alpha } \ - {(1-\delta _{t+1}^w) \mu _{t+1}^{w, u}}(1+r^{w, u}_{t+1}) - \mu _{t+1}^{w, s} (1+r^{w, s}_{t+1}) \nonumber \\&- w_{t+1} l_{t+1}^{w} - \frac{\Omega _{t+1}^w}{1+\psi } \Big ({\delta _{t+1}^{w} \mu _{t+1}^{w, u} (1+r^{w, u}_{t+1})} \Big )^{1+\psi }+p^K_{t+1}k_{t+1}^w(1-\tau ). \end{aligned}$$

(6)

So, depending on the level of technology firm’s profit can either be $\bar{\Pi }_t$ or $\underline{\Pi _t}$. $\Omega _{t}^w$ is a macro variable that represents the aggregate credit conditions.⁷ It evolves according to:

$$\begin{aligned} \Omega _t^w=\Omega _{ss}^w \left( \frac{ \mu ^{w, u}_{ss} (1+r^{w, u}_{ss})}{GDP_{ss}}\right) ^{\omega } (\delta _{ss}^w)^{\gamma } \left( \frac{GDP_{t}}{ \mu ^{w, u}_{t} (1+r^{w, u}_{t})}\right) ^{\omega } \frac{1}{(\delta _t^w)^{\gamma }}. \end{aligned}$$

(7)

$\Omega _{t}^w$ varies with the aggregate debt, but individual firms do not internalize how their borrowing decisions affect the aggregate credit conditions.

$$\frac{\Omega _{t+1}^w}{1+\psi } \Big ({\delta _{t+1}^{j} \mu _{t+1}^j(1+r^{w,u}_{t+1})}\Big )^{1+\psi }$$

is the pecuniary cost of the loss given default (cost of negotiating the debt).

Firms solve:

$$\begin{aligned} \max _{k_{t+1}^w, \mu _{t+1}^{w,u}, \mu _{t+1}^{w,s}, l_{t+1}^w, \delta _{t+1}^w} \mathbb {E}_t \beta _t^{h} \lambda _{t+1}^{h} \Big [\Pi _{t+1}^w \Big ] \end{aligned}$$

(8)

subject to 4, 5, and 6.

We assume that firms can only issue non-state-contingent nominal bonds to banks, or, equivalently, nominally riskless loans are obtained from banks. Each firm may choose to renege on some of their debt obligations, but then suffer a renegotiation cost proportional to the scale of loss given default. As firms vanish after their second period of life, their ability to liquidate assets and pay dividends to shareholders is predicated on successfully negotiating their existing debt burden. In this sense, the decision on how much of their debt to default on is strategic.

This cost effectively creates a borrowing constraint and stems from Shubik and Wilson (1977) and Dubey et al. (2005) and applied in Tsomocos (2003), Goodhart et al. (2005), Goodhart et al. (2006) and Goodhart et al. (2018). We follow Goodhart et al. (2018) by introducing a macrovariable that governs the marginal cost of renegotiating debt (default), termed ‘credit conditions’. This reflects changing motivations and incentives of debtors to make the necessary sacrifices to repay their obligations, as emphasized by Roch (2016). Ultimately the pecuniary cost of default methodology and credit conditions variable allows us to calibrate the model to realized average loss given default rates (fraction of firms who default times loss given default, or, equivalently, total non-performing loans rates on bank lending). We believe that this approach has valid economic grounds and argue that credit conditions can be adequately captured by an appropriate state variable in order to describe the relationship between loan delinquencies and the capital stock. Meanwhile the debtor firm takes the credit conditions variable as given since creditors are capable of imposing institutional arrangements that are non-negotiable.

3.4 Banking sector

Banks are also two-period lived. New-born banks are capitalized with equity ($e_t^{bank}$). They accept deposits from households ($d^{bank}_{t+1}$), extend secured ($\mu ^{bank,s}_{t+1}$) and unsecured ($\mu ^{bank,u}_{t+1}$) loans to firms. The first period budget constraint of a bank is given by

$$\begin{aligned} \mu _{t+1}^{bank} = d_{t+1}^{bank} +e_t^{bank}- adj^b_t, \end{aligned}$$

(9)

where the total assets of the bank are given by $\mu _{t+1}^{bank} = \mu _{t+1}^{bank,s} + \mu _{t+1}^{bank,u}$ and the adjustment costs are $adj^{b}_t$.⁸ The capital adequacy ratio is defined as the ratio of bank capital to risk weighted assets net of reserves ($r w a_t^{bank}$):

$$\begin{aligned} k_{t}^{bank}= \frac{e_t^{bank}}{r w a_t^{bank}}=\frac{e_t^{bank}}{( \bar{r w} \mu _{t+1}^{bank,u}+ \bar{r w} \mu _{t+1}^{bank,s})} \end{aligned}$$

(10)

Banks then choose how much of secured and unsecured debt to lend out to firms:

$$\begin{aligned} {\Pi }_{t+1}^{bank}= & {} [\theta ^w (1+r^{w,u}_{t+1}) {(1-\delta _{t+1}^{w}) {\mu }^{bank,u}_{t+1}} +(1-\theta ^w) (1+r^{w,u}_{t+1}) {\mu }^{bank,u}_{t+1} \nonumber \\{} & {} \quad + (1+r^{w,s}_{t+1}) {\mu }^{bank,s}_{t+1} -[(1+r_{t+1}^d) d_{t+1}^{bank} ], \end{aligned}$$

(11)

where $r^{w,u}_t$ and $r^{w,s}_t$ are unsecured and secured lending rates. We also assume that only “unlucky” firms default on their unsecured borrowing. Given $\left\{ \delta _{t+1}^{w}, r^{w,u}_{t+1}, r^{w,s}_{t+1},\right.$$\left. r^d_{t+1} \right\}$, banks maximize:

$$\begin{aligned} \begin{aligned} \max _{\mu _{t+1}^{bank,u},\mu _{t+1}^{bank,s}, {d}_{t+1}^{bank}} \mathbb {E}_{t} \beta ^{h}_{t} \frac{({ \Pi _{t+1}^{bank}})^{1-\varsigma _{bank}}}{1-\varsigma _{bank}} - a_{cap} 0.5 [k^{bank}_{t}-\bar{k}^{bank}]^2 \end{aligned} \end{aligned}$$

(12)

3.5 Government

3.5.1 Fiscal authority

Government gets all revenue ($p_t^{o,dom} O_t$) from oil export ($O_t$). Government spends its funds on the domestically produced final goods ($G_t$) and imported goods ($G^{imp}_t$), can save or borrow through the domestic government bonds ($B^g_t$) and receives net taxes from agents in the economy. The Government Budget Constraint:

$$\begin{aligned} \begin{aligned} G_t + p_t^{imp} G_t^{imp}+B^g_{t}\frac{(1+i^b_{t})}{1+\pi _t}= B^g_{t+1} +p_t^{o,dom} O_{t} + T^{w} \end{aligned} \end{aligned}$$

(13)

3.5.2 Monetary authority

The Central Bank controls the interest rate $i^b_t$ according to the following rule:

$$\begin{aligned} \frac{1+i^b_t}{1+i^b_{ss}}=\left( \frac{1+i^b_{t-1}}{1+i^b_{ss}}\right) ^{\rho _{i}} \left( \frac{1+\pi ^{cpi}_t}{1+\pi ^{cpi}_{ss}} \right) ^{1+\rho _{\pi }} \left( \frac{GDP_t}{GDP_{ss}} \right) ^{\rho _{gdp}} \varepsilon _t^i, \end{aligned}$$

(14)

where $\varepsilon _t^i$ is a monetary policy shock that follows AR(1) process. The CPI inflation is defined as:

$$\begin{aligned} 1 + \pi _t^{cpi} = (1+ \pi _t) \frac{r^{cpi}_t }{r^{cpi}_{t-1}}, \end{aligned}$$

(15)

where $r^{cpi}_t$ is measured as $r^{cpi}_t = p^{imp}_t T^{weight}_t + (1-T^{weight}_t),$ with $T^{weight}_t$ being defined as: $T^{weight}_t = \frac{c_{T,t}}{c_{T,t} + c_{N,t}}.$

3.6 Wedges and financial frictions: the default channel

In this discussion, we explore two versions of our model, each addressing different sources of financial inefficiency: the collateral constraint, and the dead-weight cost associated with loss given default. The first version examines “wedges” or inefficiencies that arise from financial frictions and vary over time. We refer to this as the “endogenous financial frictions” scenario. The second version, known as the “exogenous financial frictions” scenario, features constant “wedges” throughout the business cycle. Detailed derivations of these models can be found in the Online Appendix.

As we found the wedge from the collateral constraint not to be significant, we focus on the wedge arising from the dead-weight cost of default. In the endogenous financial frictions model, firms strategically decide the proportion of debt they wish to default on, denoted as $\delta ^w_{t}$. In the endogenous financial frictions case, the optimality condition of the firm w.r.t. the default rate at time t is:

$$\begin{aligned} \delta ^w_t = \frac{1}{\Omega ^w_t \mu ^{w,u}_{t} (1+r^{w,u}_{t})} \end{aligned}$$

(16)

and results in the effective first order condition for debt of:

$$\begin{aligned} \lambda ^{h}_{t+1} \beta _{t}^{h} ((1+r^{w,u}_{t+1})(1- \delta ^w_{t+1}) + (1+r^{w,u}_{t+1})\delta ^w_{t+1} )= \lambda ^{w}_{t} (1 - a^{w,u} (\mu _{t+1}^{w,u} - \mu ^{w,u}_{ss})). \nonumber \\ \end{aligned}$$

(17)

The time-varying wedge that the cost of default generates is $(1+r^{w,u}_{t+1})\delta ^w_{t+1}.$ In the exogenous financial frictions model, this wedge is taken at the steady state level. We apply the previously mentioned first order condition while keeping $\delta ^w_t$ at its steady-state level. This results in the effective first order condition for debt of:

$$\begin{aligned} \lambda ^{h}_{t+1} \beta _{t}^{h} ((1+r^{w,u}_{t+1})(1- \delta ^w_{t+1}) + (1+r^{w,u}_{ss})\delta ^w_{ss} )= \lambda ^{w}_{t} (1 - a^{w,u} (\mu _{t+1}^{w,u} - \mu ^{w,u}_{ss})).\nonumber \\ \end{aligned}$$

(18)

The constant wedge with exogenous costs of default is $(1+r^{w,u}_{ss})\delta ^w_{ss}.$ This approach implies that while the default loss remains constant throughout the business cycle, the premium or wedge linked to default still fluctuates. This methodology enables us to separate the impact of changes in the default rate (thus highlighting the significance of incomplete markets) from the influence of the wedge, which relates to the borrowing constraint. This mechanism is what we call the default channel and is different from the balance sheet or financial accelerator channels of papers such as Shousha (2016) we macroeconomic conditions feed into the marginal cost of debt through the credit conditions variable.

The difference between the two wedges above corresponds to the “investment wedge” in the terminology of Chari et al. (2007). The last equation shows that moving over the business cycle loss given default rates create a wedge for unsecured borrowing. When we substitute Eq. 16 into 17 and recall the definition of $\Omega ^w_t$, it becomes evident that the wedge is influenced by the debt-to-GDP ratio. By correlating these factors with the investment wedge, we enhance our model’s accuracy, enabling the oil price shock to have a direct impact on investment and, consequently, GDP.

4 Estimation and simulation

4.1 Observables

Our model employs Bayesian Estimation techniques to analyze two scenarios: endogenous financial frictions and exogenous financial frictions. This analysis is based on eight data series: GDP growth rates, household consumption growth rates, percentage change in CPI inflation, percentage change in interest rates, total loans growth rates, household domestic currency deposits growth rates, percentage change in the ratio of non-performing loans to total loans, and growth rates of international oil prices. For interest rate data, we utilize the Moscow Inter bank Average Credit Rate (MIACR). The study period spans from Q2 2001 to Q2 2018. We sourced data from the Federal State Statistics Service of the Russian Federation and the Bank of Russia. Specifically, quarterly consumption and output data were obtained from the Federal State Statistics Service of the Russian Federation.⁹ Other data series were taken from Bank of Russia.¹⁰ The key descriptive statistics of the data used are represented in the Table 1.

Details of our measurement equations and the variables used in them can be found in the Online Appendix.

4.2 Shocks

The model contains fourteen exogenous variables, six of them are structural shocks that follow AR(1) process and eight are measurement errors, one for every observable. The structural shocks included in the model are: international oil price shock, monetary policy shock, total factor productivity shock, shock to foreign bond interest rate premia and saver time-preference shock. The equations for the shocks are found in the Online Appendix.

The rest of the shocks are the measurement errors that correspond to each of the observables:

$\varepsilon ^{me}_{p,o}$, $\varepsilon ^{me}_{gdp}$, $\varepsilon ^{me}_{cons}$, $\varepsilon ^{me}_{\pi ^{cpi}}$, $\varepsilon ^{me}_{i^b}$, $\varepsilon ^{me}_{l}$, $\varepsilon ^{me}_{npl}$, $\varepsilon ^{me}_{dep}$. The measurement errors are mean-zero with a variance set to 10% of the variance of the corresponding data series. By doing this we follow the approach used in Adolfson et al. (2013). As a result, each of the observables could be explained by no more than six shocks: five structural shocks and the corresponding measurement error.

4.3 Calibrated parameters

Calibrated values are given in Table 7. Household’s time-preference parameter $\beta$ is set to yield in the steady state an annual risk-free rate of about 9.4 percent which corresponds to the average Russian government bond yield for the period we consider. Loss given default value $\delta ^f$ is also set in accordance with the Russian data. Capital requirement for banks $k^{bank}$ corresponds to the Russian capital requirement for big banks. The depreciation rate $\tau$ is set to yield an annual depreciation rate of 10 percent. The fraction of firm’s that default $\theta _f$ is calibrated to the Russian banks’ statistics on firms’ default. Other parameters are calibrated to yield the steady state ratio of aggregate consumption to GDP of about 54 percent as well as the steady state size of the oil sector in the economy of about 39%.

The parameter values that we use for our calibration are close to those used or estimated in other models of the Russian economy. For instance, the depreciation rate corresponds to the rate used in Malakhovskaya and Minabutdinov (2014). As follows from Malakhovskaya and Minabutdinov (2014), estimated value of household risk aversion for Russian economy is 1.015. In Polbin (2014) the estimated mean value of household risk aversion is close to its prior value $-$ 1.19.

4.4 Estimated parameters

Table 9 shows the results of the Bayesian Estimation of the model for the two cases: endogenous financial frictions and exogenous financial frictions, where the negotiated unsecured loan repayment amount $1- \delta$ is constant. The main difference in the estimation lies in the adjustment costs, in particular, bank’s adjustment costs to secured lending and capital producer’s adjustment costs to investment. The values of these two adjustment costs are much higher in the exogenous financial frictions case, which means that they add additional frictions into the model to match the data.

The central result of our estimation is presented in Table 3. From this table we can see that the marginal likelihood for the model with endogenous financial frictions is higher (1213 vs. 942).

Table 3 Marginal likelihood

Table 4 Business cycle moments Q2 2001–Q2 2018

4.5 Theoretical moments and error variance decomposition

Table 4 shows the business cycle moments of the simulated economy. In Table 5 we can see that in the endogenous financial frictions case 33.6% of the the variation in GDP is explained by the oil price shock ($\epsilon ^{p,o}$) compared to 58.3% for the TFP shock ($\epsilon ^a$ ), while in the exogenous financial frictions case the contribution of oil falls to 2.49% while the contribution of TFP rises to 63.8%.¹¹

The contribution of investment shocks to explain all the variables declines and in some cases dramatically when we move from the exogenous to endogenous case. For GDP it falls from 17.3 to 0.3% while for Loans (Deposits) it falls from 66.8% (47.3%) to 15.8% (5.8%). Justiniano et al. (2011) speculate that investment shocks may be a proxy for financial frictions, and here we see that the role of the investment shock in explaining fluctuations in non-performing loans (i.e. the spread for lending to firms) falls from 35.9 to 7.8%. The shock to the discount factor ($\epsilon ^{\beta ,h}$), criticized by Chari et al. (2007) and Chari et al. (2009) as not being truly structural, falls in its contribution to the variance of variables when moving from the exogenous financial frictions to the endogenous financial frictions case. In particular, for Loans (Deposits), the contribution of the discount rate shock goes from 18.9% (8.7%) in the exogenous financial frictions case to 3.6% (1.4%) in the endogenous case. Coupled with the better fit for the non-performing loans rate, the larger contribution of the observed shock series gives a clearer role for policy actions to depend on these shocks.

Table 5 Error variance decomposition: endogenous and exogenous financial frictions cases

4.6 Impulse response functions

Figures 4 and 5 represent IRFs for a positive oil price shock, while Figs. 6 and 7 represents IRFs for a positive TFP shock. Each figure represents the response of the variables under endogenous financial frictions and exogenous financial frictions cases.

Increased productivity under positive TFP shock expands production and real wages that in turn forms higher demand for consumption of imported goods. The rise in consumer imports leads to currency depreciation that rises the cost of capital production through the imported component of capital investment. As the case with exogenous default implies exogenous collateral constraint as well, the secured lending effectively doesn’t change, making firms more financially constrained.

Fig. 4

IRFs to a positive oil price shock, macro variables

Fig. 5

IRFs to a positive oil price shock, financial variables

Fig. 6

IRFs to a positive TFP shock, macro variables

Fig. 7

IRFs to a positive TFP shock, financial variables

Under endogenous default, as the relative price of capital shifts upwards, the collateral constraint is relaxed and the quantity of secured debt issued increases immediately. As the price of capital falls back to its steady state value, firms switch their issuance of debt towards unsecured loans. The higher profitability of the production sector results in an improvement in credit conditions and a sharp decline in the rate of non-performing loans. Relaxed collateral constraint and reduction in NPL ratio amplifies the response of aggregate lending to the economy, resulting in higher GDP and aggregate consumption levels in the transition. The response of inflation reflects the lower real price of domestic output which dominates the currency depreciation, resulting in inflation declining and a decline in the nominal interest rate.

In the case with endogenous financial frictions, a shock to the foreign oil price causes a sharp appreciation in the exchange rate,¹² causing a corresponding large increase in imports. The stronger exchange rate causes a reduction in the cost of imported goods for capital goods, and hence a fall in the price of capital. The decline in the price of capital reduces the ability to issue secured debt, which is substituted by issuing unsecured debt. In the expectation of future currency depreciation, households switch from domestic savings in equity to foreign bonds which is used to finance imported consumption and resulting in lower labor supply in subsequent periods. This causes a decline in the production of domestic non-tradables in the medium term and is evidence of a Dutch-disease type effect in Russia: an increase in the tradable sector causes the non-tradable sector to contract via the price of inputs, here labor.¹³ The decline in the interest rate on unsecured debt causes credit conditions to improve and non-performing loans rate to decline. Our evidence for this effect is consistent with Malakhovskaya and Minabutdinov (2014), but contrasts Kreptsev and Seleznev (2017) and Kozlovtceva et al. (2019). This effect is pronounced in our model because of the strong substitution between domestic and foreign consumption goods driven by the high elasticity of the real exchange rate with respect to the dollar price of oil. One reason is that our foreign interest rate doesn’t depend explicitly on the dollar oil price as in the case of Kreptsev and Seleznev (2017) and Kozlovtceva et al. (2019). This means that as our foreign interest rate does not decrease when oil price increases, households have a greater incentive to accumulate foreign assets and sustain their consumption of imports in the future. Another reason for our stronger Dutch-disease type effect is that oil revenue is given directly to government who spends it, and as a result aggregate demand directly depends strongly on the domestic price of oil which falls due to a strongly appreciating exchange rate. In practice government spending will not adjust as much, however in our model government spending substitutes for a hand-to-mouse consumer whose consumption depends directly on domestic currency oil revenues.

5 Concluding remarks

Since the Global Financial Crisis policy makers in emerging economies focused on novel, macroprudential tools to maintain both price and financial stability. These tools mitigate the domestic effects of external shocks. Since the effectiveness of policy tools depends on the shock, discerning which shocks drive business cycle dynamics becomes as important to understand as which financial frictions amplify them. Through the lens of an estimated financial frictions augmented, small open economy New Keynesian model we show that the contribution of commodity price shocks to output fluctuations depends qualitatively and quantitatively on the inclusion of financial frictions. For the specific Russian case we estimate, commodity price shocks are amplified by domestic credit market conditions.

Appendix

See Tables 6, 7, 8, and 9.

Table 6 Corporate loans in Russia: secured and unsecured

Table 7 Calibrated parameters and ratios

Table 8 Steady state values of variables

Table 9 Estimated parameters for endogenous and exogenous financial frictions cases

1.1 Equilibrium

Given the exogenous shocks, equilibrium is a sequence of prices and quantities such that each agent in the economy maximizes her value, and all markets clear. In particular, market clearing condition for labor requires:

$$\begin{aligned} l^{h}_t = l^w_t \end{aligned}$$

(19)

Market clearing for secured loans:

$$\begin{aligned} \mu ^{bank,s}_{t+1} = \mu ^{w,s}_{t+1} \end{aligned}$$

(20)

Market clearing for unsecured loans:

$$\begin{aligned} \mu ^{bank,u}_{t+1} = \mu ^{w,u}_{t+1} \end{aligned}$$

(21)

Market clearing for deposits:

$$\begin{aligned} d^{h}_{t+1} = d^{bank}_{t+1} \end{aligned}$$

(22)

Market clearing for domestic bonds:

$$\begin{aligned} B^{g}_{t+1} = B^{g, h}_{t+1} \end{aligned}$$

(23)

Market clearing for domestic output:

$$\begin{aligned} Y^{ret}_t= & {} c_{N,t} + i_{N,t+1} + G_t + \theta ^w \frac{\Omega _{t}^w}{1+\psi } \Big ({\delta _{t}^{w} \mu _{t}^w(1+r^{w,u}_{t})}\Big )^{1+\psi } \nonumber \\{} & {} + adj^h_t + adj^w_t + adj^{b}_t \end{aligned}$$

(24)

Household’s time-preference variable $\beta ^{h}_t$ is defined as:

$$\begin{aligned} \beta ^{h}_t = \beta ^{h} \varepsilon ^{\beta , h}_t. \end{aligned}$$

(25)

Domestic price of an imported good is:

$$\begin{aligned} p^{imp}_t = Q_t p^{imp, \star }, \end{aligned}$$

(26)

where $p^{imp, \star }$ is an international price of an imported good and we assume it to be constant and $Q_t$ is a real exchange rate. Domestic price of commodity good (oil) is:

$$\begin{aligned} p^{o, dom}_t = Q_t p^{o, \star }_t, \end{aligned}$$

(27)

where $p^{o, \star }_t$ is an international price of commodity good and it is defined as:

$$\begin{aligned} p^{o, \star }_t = p^{o, \star } \varepsilon ^{p,o}_t. \end{aligned}$$

(28)

So, the international price of oil is a product of some constant oil price $p^{o, \star }$ and its shock process $\varepsilon ^{p,o}_t$, which follows AR(1) process. Interest rate on the foreign bonds also subject to the shock, which we call ”foreign interest rate shock“. So, interest rate on foreign bonds is defined as:

$$\begin{aligned} r^f_t = r^f + \varepsilon ^{r,for}_t, \end{aligned}$$

(29)

where $r^f$ is some constant interest rate on foreign bonds and $\varepsilon ^{r,for}_t$ is a shock process for interest rate on foreign bonds that follows AR(1) process. We assume that the technology levels of ”lucky“ and ”unlucky“ firms are correspondingly $\bar{A_t^j}$ and $\underline{A_t^j}$.

$$\begin{aligned} \bar{A_t^j} = A_t \bar{A^j}, \end{aligned}$$

(30)

where $\bar{A^j}$ is some constant and

$$\begin{aligned} \underline{A_t^j} = A_t \underline{A^j}, \end{aligned}$$

(31)

where $\underline{A^j}$ is some constant with $\bar{A^j}> 1 > \underline{A^j}$.

The real interest rate on domestic government bonds is defined as:

$$\begin{aligned} 1+r^b_t = \frac{1+ i^b_{t-1}}{1+ \pi _t}. \end{aligned}$$

(32)

We define real GDP in the model as follows:

$$\begin{aligned} GDP_t = Y_t^{ret} + p_{ss}^{o,dom} O_t \end{aligned}$$

(33)

Aggregate real consumption in the model is defined as:

$$\begin{aligned} cons_t = p_t^{imp} c_{T,t} + c_{N,t} \end{aligned}$$

(34)

In the data the procedure of calculating GDP and its components in constant prices includes two key approaches: the reevaluation of GDP and its components in the previous periods prices using the indexes of volume and through the direct division of current nominal values by the change in the price index. So, given that model variables are in real prices, consumption and GDP could be measured either in constant real prices or in changing real prices. In our model we measure real GDP in constant real prices, while we measure consumption in changing real prices.

The empirical relevance of our credit conditions variable.$\Omega ^w_t$ is constructed to be falsifiable. If it is not a valid description of the relevant dead-weight costs of default, then the estimated values of parameters $\omega$, $\gamma$ and $\psi$ should be estimated to be close to zero. Suppose that $\omega$, $\gamma$ $\rightarrow$ 0. Then from Eq. (7), $\Omega ^w_t$ $\rightarrow$ $\Omega ^w_{ss}$. $\Omega ^w_{ss}$ is determined from equation:

$$\begin{aligned} \Omega ^w_{ss} (\delta ^w_{ss} (1+r^{w,u}_{ss}) \mu _{ss}^{w,u})^{\psi } = 1. \end{aligned}$$

(35)

From Eq. (35) follows that as $\psi \rightarrow 0$, $\Omega ^w_{ss} \rightarrow 1$. Then we have that:

$$\begin{aligned} (\delta _{t}^w (1+r^{w,u}_{t}) \mu _{t-1}^{w,u})^{\psi } = 1. \end{aligned}$$

(36)

From (36), at $\psi = 0$, this optimally condition holds true which for all choices of $\delta _{t}^w$ and implies that $\delta _{t}^w$ stays close to its steady state level along a stable unique path. However, as all the estimated values of these parameters are different from zero, we can say that both aggregate credit conditions variable and the cost of negotiating the debt are important for matching the movement of the observed data series.