Introduction
The time lags inherent in the transmission of monetary policy require central banks to be forwardlooking. The Bank of Canada actively monitors indicators that contain information about the Canadian outlook, most notably to build economic projections in its Monetary Policy Report.
The literature shows that movements in corporate spreads are an important leading indicator of economic activity.1 For the United States, Gilchrist and Zakrajšek (2012) find that variations in the pricing of corporate credit risk (rather than variations in the probability of default by the issuer) contribute to this predictive ability of corporate spreads. These variations are called the excess bond premium (EBP) and are often associated with investor sentiment or risk appetite in the corporate bond market.
Given that the US and Canadian corporate bond markets differ markedly, we investigate whether Canadian corporate spreads and the EBP also lead Canadian economic activity. Indeed, we find that corporate spreads precede changes in real gross domestic product (GDP) in Canada over the subsequent year. The EBP accounts for most of this property. Further, an unanticipated increase in the Canadian EBP forecasts a deterioration of domestic macroeconomic conditions: a 10basispoint increase results in a fall in both GDP and consumer price index (CPI) of 0.4 per cent and 0.1 per cent, respectively, over three years.
Methodology
Following Gilchrist and Zakrajšek (2012), we initially predict future real GDP growth using past growth and two variables reflecting the stance of monetary policy: the real overnight rate and the term spread (i.e., the difference between the Canadian 10year nominal government bond yield and the 3month treasury bill rate).
We then supplement this baseline forecasting regression with four measures of credit spread, resulting in five separate regressions. The US EBP from Gilchrist and Zakrajšek (2012) is used as our main US credit spread measure. Our three Canadian credit spread measures include Canadian corporate spread and its two subcomponents: the compensation for expected default (CED) and the EBP. These three measures were computed by Leboeuf and Pinnington (2017), who followed the work of Gilchrist and Zakrajšek (2012). The CED is a prediction of the compensation required by investors for the probability of a firm defaulting on its bond obligations. CED fluctuates with market estimates of the default probability of individual firms. The EBP is the residual part of the spread, containing the extra yield that investors demand to hold corporate bonds.
We estimate the forecasting regressions for two GDP forecast horizons, 3 months and 12 months, using monthly data from January 1999 to August 2017. Our specification differs from that of Gilchrist and Zakrajšek (2012) in one aspect: the number of lags is fixed across all regressions to isolate the marginal effect of each credit spread measure in forecasting movements in GDP.2 The Appendix contains detailed regression specifications and results, as well as a description of the variables.
Chart 1: Canadian EBP is highly informative about the outlook for Canadian GDP
Sources: Bloomberg, Haver Analytics, Board of Governors of the Federal Reserve System and Bank of Canada calculationsLast observation: August 2017
The Canadian EBP is highly informative about the Canadian growth outlook
Chart 1 shows the R² to compare the proportion of GDP variations explained by each specification. 3
We find that corporate spread is a leading indicator of future movements in Canadian GDP. Supplementing the baseline regression with the Canadian corporate spread increases the adjusted R² meaningfully—from 12 to 38 per cent at the 3month horizon and from 9 to 35 per cent at the 12month horizon.
The predictive content of Canadian corporate spreads is largely owing to the EBP; the adjusted R² of the regression with the EBP is considerably higher than that of the CED at both horizons. Gilchrist and Zakrajšek (2012) find the same result for the United States. In addition, using the EBP rather than corporate spread actually improves the model’s ability to explain movements in GDP. Perhaps surprisingly, the CED has no additional predictive content for economic activity relative to the baseline regression. Finally, the US EBP also forecasts Canadian economic activity, although the R² are notably lower than with Canadian EBP.
Increases in the EBP point to a deterioration of economic activity
We next investigate the behaviour of the Canadian economy following sudden increases in the EBP. We find these situations tend to be followed by a worsening of economic conditions. This exercise is particularly useful since the EBP is available in real time and predicts future movements in GDP.
Specifically, we use a standard vector autoregression (VAR) that describes the Canadian economy in terms of economic and financial variables. The economic variables are real GDP and CPI. The financial variables are the S&P/TSX equity market Composite Index, 10year Government of Canada (GoC) yields, the overnight rate and the Canadian EBP. Our approach is similar to Gilchrist and Zakrajšek (2012) but differs in two ways: our VAR does not include consumption and investment (unavailable at the monthly frequency), but it does include West Texas Intermediate (WTI) crude oil prices as an exogenous variable.4 We estimate the model using monthly data from January 1999 to August 2017. Additional details on the VAR specification are provided in the Appendix.
Using the model, we carry out an experiment where the EBP unexpectedly rises by 10 basis points. This shock is equivalent to one standard deviation. One interpretation of this experiment is that it shows how the economy and financial markets respond following a sudden increase of investors’ risk aversion. Chart 2 shows the response of each variable within our specified model. The higher EBP is followed by a deterioration of macroeconomic conditions: the levels of GDP and CPI fall by a total of 0.45 and 0.1 percentage points, respectively, over 3 years. Movements in financial variables suggest that EBP shocks are associated with a flight to safety, where investors move capital from more to less risky assets: the S&P/TSX Composite Index declines by about 3 per cent and the 10year GoC yield by 10 basis points. Finally, the overnight rate declines by about 20 basis points in our experiment. This does not mean that the central bank responds to changes in EBP. Indeed, the financial and macroeconomic conditions that follow a surprise change in EBP are those that are associated with a monetary policy response. Note that the aforementioned dynamics are specific to this model.
Chart 2: Responses of macroeconomic and financial variables to a 10basispoint increase in the Canadian excess bond premium
Notes: The charts depict the response of a onestandarddeviation (10 basis points) orthogonalized shock to the excess bond premium. Responses of GDP, CPI and TSX have been accumulated.The 95% confidence interval is shown with dashed lines.
Sources: Bloomberg, Haver Analytics and Bank of Canada calculations
Last observation: August 2017
Conclusion
We show that the excess bond premium drives the predictive content of corporate spreads for Canadian economic activity. Practitioners looking to predict future developments in the Canadian economy should therefore look beyond the standard measure of corporate spread.
The Canadian corporate bond sphere is diverse; it includes financial and nonfinancial firms of varying sizes. Analyzing the information contained in the credit spreads of different sectors is a promising avenue for future research.
Appendix
Table A1 below shows the estimation results from 10 reducedform regressions (equation A1). We estimate the model on monthly data from January 1999 to August 2017 and use three and six lags of GDP for 3 and 12month forecasting horizons, respectively. The coefficients of the constant and lags of GDP are not shown in the table. Ordinary least squares estimated coefficients are reported with NeweyWest standard errors in brackets.
Equation A1: Regression specification
\(ΔGDP_{t+h}=α_{i} \sum_{i}ΔGDP_{ti} + β_{1} Term \, spread_{t} \) \(+ β_{2} Overnight \, rate_{t} + β_{3} Credit \, spread _{t} + ε_{t}, \)
where for a given forecasting horizon \( h≥0, ΔGDP_{t+h} = \frac{1200}{h+1} ln \left( \frac{GDP_{t+h}}{GDP_{t1}} \right) \).
Table A1: Regression results
Table A1: Regression results  

Forecast horizon: 3 months  Forecast horizon: 12 months  
Dependent variable: Real GDP 
(1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  (10)  
Term spread  0.496^{**} (0.02) 
0.411 (0.12) 
0.512^{**} (0.02) 
0.443^{*} (0.05) 
0.805^{***} (0.00) 
0.589^{***} (0.00) 
0.515^{***} (0.00) 
0.626^{***} (0.00) 
0.523^{***} (0.00) 
0.772^{***} (0.00) 

Overnight rate  0.177 (0.28) 
0.385^{*} (0.10) 
0.212 (0.34) 
0.272 (0.17) 
0.372^{**} (0.04) 
0.003 (0.98) 
0.364^{**} (0.04) 
0.062 (0.63) 
0.329^{**} (0.03) 
0.110 (0.40) 

Canadian corporate spread  3.169^{***} (0.00) 
2.100^{***} (0.00) 

Compensation for expected default  0.560 (0.79) 
1.100 (0.52) 

Excess bond premium 
4.056^{***} (0.00) 
2.917^{***} (0.00) 

Excess bond premium (US) 
2.239^{***} (0.00) 
1.321^{***} (0.00) 

Adj R²  0.12  0.38  0.12  0.47  0.44  0.09  0.35  0.09  0.52  0.36  
Akaike Information Criterion  954  884  956  851  861  753  687  754  626  683 
Equation A2: Vectorautoregression specification
$$y_{t}=Φ_{0}+ \sum_{i=1}^{2} Φ_{i} y_{ti} + Θx_{t}+ε_{t},$$ where \( y_{t}\) is the vector of endogenous variables, \( Φ_{0}\) is a constant vector, \( Φ_{i}\) are coefficient matrices of lag i, \( x_{t}\) is the exogenous variable (i.e., WTI crude) and \( ε_{t}\) is a vector of white noise innovations.Table A2: Variables and data sources
Table A2: Variables and data sources  

Data  Source  Computation and units 
Reducedform regressions  
Monthly real GDP  Statistics Canada  Logdifference, at basic prices 
10year Government of Canada yield  Statistics Canada  In percentage points 
3month treasury bill rate  Statistics Canada  In percentage points 
Overnight rate  Statistics Canada  In percentage points 
Canadian corporate spreads  Leboeuf et al. (2017)  In percentage points 
Canadian compensation for expected default  Leboeuf et al. (2017)  In percentage points 
Canadian excess bond premium  Leboeuf et al. (2017)  In percentage points 
US excess bond premium  Gilchrist et al. (2012)  In percentage points 
VAR  
Headline consumer price index  Statistics Canada  Logdifference 
Monthly real GDP  Statistics Canada  Logdifference, at basic prices 
Canadian excess bond premium  Leboeuf et al. (2017)  In percentage points 
Overnight rate  Statistics Canada  In percentage points 
10year Government of Canada yield  Bloomberg Finance L.P.  In percentage points 
S&P/TSX Composite Index  Bloomberg Finance L.P.  Logdifference 
WTI crude oil prices  Bloomberg Finance L.P.  Logdifference (US dollars) 
 Among others, Gilchrist and Zakrajšek (2012), Gilchrist and Mojon (2017) and Bedock and Stevanovic (2017). [←]
 Gilchrist and Zakrajšek (2012) choose the optimal number of lags for each regression using the Akaike Information Criterion (AIC). For robustness, we test their method and obtain qualitatively identical results. [←]
 AIC is presented in the Appendix and is consistent with the adjusted R² values. [←]
 We include WTI to account for the importance of the oil sector in the Canadian economy. [←]
References
Bedock, N. and D. Stevanović. 2017. “An Empirical Study of Credit Shock Transmission in a Small Open Economy.” Canadian Journal of Economics/Revue canadienne d'économique 50: 541–570.
Gilchrist, S. and E. Zakrajšek. 2012. “Credit Spreads and Business Cycle Fluctuations.” American Economic Review 102 (4): 1692–1720.
Gilchrist, S. and B. Mojon. 2017. “Credit Risk in the Euro Area.” The Economic Journal 128 (608): 118–158.
Leboeuf, M. and J. Pinnington. 2017. “What Explains the Recent Increase in Corporate Bond Spreads?” Bank of Canada Staff Analytical Note No. 20172.