Does local and Euro area sentiment matter for sovereign debt markets? Evidence from a bailout country

ABSTRACT

Does sentiment impact the sovereign debt markets? This article investigates whether lagged domestic and Euro area irrational sentiment (optimism or pessimism unwarranted by fundamentals) predicts future sovereign bond spreads, in Portugal, between January 2000 and December 2013. We find that domestic and Euro area sentiment negatively forecasts total return spreads and that this effect is stronger during the bailout period. Also, we find that the business sentiment is even most noticed. Therefore, Portuguese sovereign debt market is prone to the influence of investors’ sentiment.

KEYWORDS:

• Investor sentiment
• sovereign debt
• bailout

JEL CLASSIFICATION:

• G10
• G12
• G15

Notes

¹ As an example of explicit sentiment measures is worth notice the Sentiment Index provided by the American Association of Individual Investors and Investors Intelligence (used by Fisher and Statman 2000; Brown and Cliff 2005; Verma and Soydemir 2006; Verma, Baklaci, and Soydemir 2008) and the Consumer Confidence Index (Qiu and Welch 2006; Lemmon and Portniaguina 2006; Schmeling 2009; Zouaoui, Nouyrigat, and Beer 2011).Proposals for implicit proxies include, for example, the closed-end fund discount (e.g. Lee, Shleifer, and Thaler 1991), the number of new initial public offerings (e.g. Lee, Shleifer, and Thaler 1991) and the put-call trading volume ratio (e.g. Wang, Keswani, and Taylor 2006). Based on implicit measures, composite indices have been also proposed, see Glushkov (2006) and Baker and Wurgler (2006, 2007) for examples of this approach. However, these variables are mainly related to the stock markets.

² The ESI is a composite indicator made up of the individual components of the following indicators: industrial confidence indicator; services confidence indicator; consumer confidence indicator; construction confidence indicator and retail trade confidence indicator. This indicator is calculated and provided by the European Commission services (DG ECFIN) at the country level and at the aggregate level (European Union and Euro area).

³ We collect data from World Bank and IMF Coordinated Portfolio Investment Survey (CPIS), and we cross the categories of data in order to know who the main holders of Portuguese sovereign bonds are. Table A1 presents the results of this work.

⁴ Fernandes, Gonçalves, and Vieira (2013) show the impact of sentiment in the Portuguese stock market.

⁵ ‘Benchmark indices are based on single bonds. The bond chosen for each series is the most representative bond available for the given maturity band at each point in time. Generally, the benchmark bond is the latest issue within the given maturity band; consideration is also given to yield, liquidity, issue size and coupon. The index constituents are reviewed at the beginning of each month, and any changes made at that time. Constituents are then fixed until the start of the following month (Datastream definition).

⁶ The consideration of global factors was motivated by the fact that Portugal is an open economy. The principal component analysis reduces the likelihood of multicollinearity and summarizes information that is extracted from the macroeconomic variables into a few principal components, retaining most of the information in the original variables.

⁷ Later these systems of equations were re-estimated using Euro area investor sentiment. That is, we jointly estimate regression equations 1b, 1c, 2b and 2c for forecast horizons of 1, 3, 6 and 12 months in a systems of equations, designated as 3b, 3c, 4b and 4c in Tables 4 and 6, and perform similar tests for the coefficient of the Euro area investors’ sentiment. The new equations are not presented to save space.

⁸ Doing some preliminary estimations and tests, we have found important effects of the Euro area sentiment on bonds spreads during the bailout, as we show in the results. So, we chose to include the interaction term between the Euro area sentiment and the dummy.

⁹ A description of the variables used in the construction of the factors, as well as their data sources are presented in Table 1. We have not included credit rating because this is a measure relatively unchanged on time series and more appropriate for cross-sectional analysis. Additionally, Afonso, Gomes, and Rother (2011) find that credit ratings are widely determined by macroeconomic and fiscal variables.

¹⁰ Some results are not included to save space, but are available upon request.

¹¹ Note that data on Euro area long-term interest rates is based on yield observations of actively traded government bonds in euros (with around 10 years’ residual maturity) of the countries of the Euro zone, weighted by the stock of bonds issued in euros (OECD data definition).

¹² In all regression processes, we use an estimator that is robust to both heteroscedasticity and autocorrelation called HAC (heteroscedasticity autocorrelated consistent) proposed by Newey and West (1987, 1994).

¹³ The levels of the bivariate correlations between the two sentiment proxies are low (Table A2). Additionally, we analyse the values of variance inflation factor (VIF) and these values are lower than 2 (not included to save space, but are available upon request).

¹⁴ The series of the principal components are transformed into rates of change by the formula ln(PC_t/PC_t–1), where PC_t is the principal component extracted in month t and PC_t–1 is the principal component extracted in month t – 1. This procedure minimizes any autocorrelation problem that comes with using the time series of variables.

¹⁵ We estimate the regression equation 2c, after, we re-estimate the same regression, but omitting the sentiment variables (i.e. only with the controls) and then compute de difference between the adjusted R² of both specifications. We perform the same procedure for regression equations 2a and 2b and find that local (Euro area) sentiment adding 3.8% (4.9) of predictive power relative to the other predictors (Table 3). The results of these regression estimations are not included to save space, but are available upon request.

¹⁶ The statistics of the same kind of test that we perform for the coefficient of the Euro area investors’ sentiment (in the systems equations 3b and 4b) also suggest the hypothesis that the sentiment coefficients are equal to zero should be rejected at the significance level of 1%.

¹⁷ Similarly, we test the hypothesis that the sentiment coefficients (domestic and Euro area) are simultaneously equal to zero, and the results show that this hypothesis should be rejected at the significance level of 5%.

¹⁸ The BCI derive from business surveys and reflects the firm’s assessments and expectations about production, selling prices, employment among other variables. More details about this indicator and about the corresponding data are provided in Table 1.

¹⁹ The levels of the bivariate correlations between the two business sentiment proxies are low (Table A2). We also analyse the values of VIFs and these values are lower than 2 (not included to save space, but are available upon request).

²⁰ See note 7.

²¹ Coefficients of the variables Sent(BCI) and Sent EA(BCI_EA) in all the systems equations 3a–3c and 4a–4c whose results are presented in Table 6.

²² The yields spreads was computed as difference between the Portuguese government bond yield and German government bond yield. Data came from Datastream.