5.3 Robustness
| Dependent variable | ΔSRP Full sample |
ΔSRP LR sample |
ΔSRP Re-weighted |
ΔLRP Full sample |
|---|---|---|---|---|
| (1) | (2) | (3) | (4) | |
| βproducer | 0.092** | 0.108** | 0.167** | 0.171** |
| [0.022] | [0.025] | [0.031] | [0.052] | |
| βlocal | 0.909** | 0.977** | 0.967** | 0.672** |
| [0.029] | [0.038] | [0.057] | [0.107] | |
| βvehicle | 0.700** | 0.744** | 0.698** | 0.230* |
| [0.029] | [0.031] | [0.048] | [0.094] | |
| N | 1,207,100 | 620,100 | 620,100 | 115,000 |
| R2 | 0.013 | 0.021 | 0.083 | 0.151 |
Regressions include unreported HS4-destination fixed effectsand macroeconomic variables as outlined in the main text. Standard errors inbrackets (** ; * denotes significance at the 1%; 5% level respectively).
Potentially, the estimated convergence of βs in the long run could be driven by either (firm- or relationship-level) selection or implicit re-weighting. For example, some relationships which are included in the short-run calculation do not continue beyond the six month threshold required to be included in the long-run analysis. Similarly, some sporadic or seasonal relationships may be observed in the long run, but not be included in the short-run analysis. Additionally, the short-run analysis includes every pair of consecutive trades, such that relationships in which goods are traded monthly will implicitly receive a higher weight in the analysis than those in which trade occurs less regularly.
Table 15 uses short-run data to consider the impact of selection and weighting on estimated ERPT parameters.[26] Column 1 repeats the main short-run specification (column 3 of Table 6 ). Column 2 reestimates this regression restricting the sample to those short-run relationships which (in total) extend beyond the threshold to be included in the long-run analysis. Column 3 then takes this reduced sample and re-weights each observation by 1/N(ΔSRPfcgt), the reciprocal of the number of observed price changes, to give each relationship a total weight of one. Finally, column 4 repeats coefficients from the comparable long-run calculation (column 3 of Table 11 )
Focussing on column 3, the combined effect of selection and re-weighting is to increase estimates of β for both producer and local currency-denominated trades, but leave the coefficient on vehicle currency trade unchanged. While the implicit reweighting between the short-run and long-run calculation largely accounts for the estimated increase in the producer currency coefficient (comparing columns 3 and 4), these factors do not explain any of the apparent change in ERPT for local and vehicle currency groups. Rather, changes in these parameters seem more likely associated with exporters having a greater ability to escape rigidities associated with, eg, explicit contracts in the long run.
Notes
- [26] This test considers only the impact of excluding short-run relationships from thelong-run analysis, not that of excluding sporadic relationships from the short-run analysis.
