Overall, Table 4.3 shows that the lagged dependent variable itself has the highest association with the dependent variable. However, also the lags of the other variables show a significant, mostly positive, association; albeit less outspoken. 4.4.3 IRFs Building on the panel VAR analysis, we can explore how hypothetical shocks to individual skills are associated with changes in other skills over time, as illustrated in Figure 4.2. This figure represents generalized IRFs.14 It depicts nine graphs, all showing how the skill variables evolve after an exogenous, one standard deviation shock.15 The vertical axis represents the response to this shock in standard deviations, whereas the horizontal axis represents time in periods of half a year, meaning that time period zero refers to the moment of the shock, and time period one to the effect of the shock half a year later. The solid line illustrates the impulse response of a one standard deviation shock to one of the residuals. The main diagonal of Figure 4.2 shows how each skill varies following a hypothetical positive shock to itself. The initial response is around 0.50-0.60 SD. In subsequent periods, the response gradually diminishes. Thus, this pattern shows a substantial change in the skill, with 14We establish that the model meets the stability condition. All eigenvalues are between 0.143 and 0.303, so they lie inside the unit circle. 15Due to computational complexities, the Monte Carlo simulations in the GIRF are limited to only 500 repetitions. This restriction arises from the intensive computational demands associated with running a large number of simulations, especially when dealing with numerous cross-sectional units. While 500 repetitions may provide a reasonable approximation, the results should be interpreted with caution, as a higher number of simulations could yield more precise estimates and confidence intervals. 97
RkJQdWJsaXNoZXIy MjY0ODMw