2. AUTOCUIDADO 2 Definición
2.2 Estructura del autocuidado
In light of the finding that arbitrarily chosen macroeconomic and financial charac- teristic factors perform overwhelming well in pricing currency portfolio returns, I turn to a second test of their pricing performance.
One additional test for assessing a candidate factor’s pricing capabilities is to test if the factor can price an alternative set of currency portfolios. Those sorted by the characteristic of interest are natural candidates for the alternative portfolios. For example, if a country’s debt-service ratio reallyis important for explaining currency premia, then a factor which captures a country’s debt-service ratio should price interest-rate-sorted portfolios and currency portfolios sorted by the debt-service ratio itself. To capture this approach, I construct sets of five currency portfolios sorted on the characteristics of interest. Specifically, this involves the creation of six sets of ‘theoretical’ test portfolios – three each forAll Countries and Developed Countries– which are sorted by consumption, output gaps and export ratios. I then construct 50 sets of ‘non-theoretical’ test portfolios – 25 each forAll Countries and
Developed Countries – sorted based on the composite and sub-indices of the PRS
data.
In Table 4.6, I record the t-statistic and associated R2 statistic from the second-step of the FMB procedure. I do this for each theoretical characteristic factor and the 25 non-theoretical characteristic factors when the test assets are the currency portfolios sorted by the same characteristic as the factor itself. I find that
noneof the 25HM Laltfactors are statistically significant when tested, with many of
the models now generating negative R2 statistics. I find comparable results across
Developed Countries, where all 25 factors are found to be insignificant.21 Of the
20In Appendix Table C.2, I document the performance of factors constructed based on the sub-
indices of political risk. The equivalent results forDeveloped Countries are provided in Appendix Table C.3.
Variable t-stat R2 Variable t-stat R2
‘Slope’ Risk (LRV) 4.73 85.1% ‘Global Imbalances’ (DCRS) 3.65 84.8%
Theoretically Motivated Characteristic Factors
Variable t-stat R2 Variable t-stat R2
Colacito and Croce 0.38 -84.1% Verdelhan 2.32 50.0% Farhi and Gabaix 0.53 -54.9%
Financial Characteristic Factors
Variable t-stat R2 Variable t-stat R2
Aggregate Financial 0.64 -127% Current Account 0.06 -96.0% Exchange Rate Stability 0.90 -61.8% Debt Service 0.55 -24.6% International Liquidity 0.04 -78.1% Foreign Debt 0.52 -123%
MacroeconomicCharacteristic Factors
Variable t-stat R2 Variable t-stat R2
Aggregate Economic 1.37 56.7% Inflation 0.44 -115% GDP per Capita 0.47 -110% Budget Balance 0.69 -15.8% GDP Growth Rate 0.03 -116% Current Account 0.37 -50.0%
Political Characteristic Factors
Variable t-stat R2 Variable t-stat R2
Aggregate Political Risk 1.35 -19.6% Military in Politics 0.52 -70.6% Government Stability 0.78 -99.4% Religious Tensions 1.44 -0.24% Socioeconomic Conditions 0.44 -50.5% Law and Order 1.68 51.0% Investment Profile 1.27 26.5% Ethnic Tensions 0.32 -26.7% Internal Conflict 1.18 1.23% Democratic Accountability 0.67 9.18% External Conflict 0.74 -60.4% Bureaucracy Quality 1.51 -29.7% Corruption 0.38 -51.0%
Table 4.6: Asset Pricing Tests: Characteristic-Sorted Portfolios. The table presents second stage cross-sectional adjustedR2 and t-statistics from the Fama and MacBeth (1973) procedure. The test assets
are five currency portfolios sorted on the same basis as the factor itself. Each regression contains two risk factors (i) DOL risk and (ii) a characteristic based risk factor. Standard errors are corrected according to Shanken (1992) with optimal lag length according to Newey and West (1987). The sample period is from October 1983 to December 2011. Details of all other data are provided in Section 4.3.
theoretical factors, only the factor reflecting the external habit model of Verdelhan (2010) is significant. Both the factors reflecting the long-run risks model of Colacito and Croce (2013) and the variable rare disasters model of Farhi and Gabaix (2013) are associated with insignificant factor prices of risk and negativeR2 statistics. In contrast, I also provide the results for an alternative characteristic factor suggested by Chapter 3 (Della Corte et al., 2014). This fundamental factor, sorted based on the mixture of a country’s net foreign asset position and currency denomination of foreign debt, is termed the ‘Global Imbalance Factor’. Chapter 3 (Della Corte et al., 2014) demonstrate that the factor can explain the returns of interest-rate- sorted currency portfolios. In Table 4.6, we find this is the only fundamental factor which also prices the set of test-asset portfolios sorted by its own characteristic.
I provide a visual comparison of each factors’ performance, when pricing interest-rate-sorted and characteristic-sorted portfolios in Figure 4.2.22 As previ- ously documented, all but 5 of the non-theoretical factors (all sub-indices of political risk) are significant when the test assets are sorted by interest rates. In contrast, when pricing portfolios sorted by the same characteristic as the risk factor, all the t- statistics fall below 2.0, while 20 of the 25 models generate a negative cross-sectional R2 statistic. The same information is plotted in the right-hand panel of Figure 4.2 for the theoretical characteristic factors.
A reason for the drop in pricing performance of theHM Lalt factors is pro-
vided in Figures 3 and 4. In Figure 4.3, I plot the average returns to the 25HM Lalt
factors, conditional on the distribution of returns toSloperisk (horizontal axes). For a significant number of factors, a monotonic pattern emerges. HighSlope risk (carry trade) returns are often associated with high returns to the HM Lalt factor. Yet,
in Figure 4.4, we find that almost none of the characteristics provide a monotonic pattern in currency excess returns when currencies are sorted into one of five port- folios based on macroeconomic, financial or political risks.23 This finding suggests that many of the characteristics, while associated with variation in currency betas are all, in fact, by-products of some other underlying driving force.