Page 80 - Contributed Paper Session (CPS) - Volume 3

CPS1947 Hsein K. et al.
supports the view that the term-structure variables are closely linked together
after 1952 but not before.
We compare our results to studies that provide in-sample evidence
regarding stock return predictability. Using the full sample, our finding of
significant predictability is consistent with those of prior studies. Using the
post-1952 period, however, while Campbell and Yogo (2006), Kostakis,
Magdalinos and Stamatogiannis (2015) and Kasparis, Andreou and Phillips
(2015) have found no or weak evidence of predictability in the post-1952
period using linear or non-linear predictive regressions with a single predictor
or multiple predictors without allowing for the presence of cointegration, we
do find strong evidence using bivariate cointegrated predictors.
Furthermore, the results in Table 1 provide ample evidence in favour of
nonlinear predictability of stock returns using some pairs of cointegrated
predictors since the coefficient on the highest power in the polynomial
regression is statistically significant at conventional levels. To illustrate the
approximate form of nonlinearity, Figure 2 plots the predicted value of equity
premium, ̂(̂ −1 ) against ̂ −1 along with the 90 percent pointwise
confidence intervals. This figure shows that the two pairs of lty-tbl and of ep-
dp exhibit a hump-shaped relationship. This empirical finding of nonlinear
predictability using these two pairs of cointegrated predictors highlights a
useful feature of our semi-parametric single index predictive model.

Figure 2: Estimated link function ̂(̂ −1 ) at quarterly frequency

Notes: This figure plots the estimated link function of each pair of cointegrated
predictors. The dashed line shows the approximate 90 percent pointwise
confidence interval and the horizontal line depicts the average quarterly equity

69 | I S I W S C 2 0 1 9