114
Onal
Table 3A.3 Individual Augmented Dickey-Fuller Test Statistics
Cashew Coffee
Cotton
Tea Tobacco
Mozambique Tanzania Kenya Tanzania Uganda Vietnam China India Zambia Kenya Tanzania Tanzania
ln(Qt)
ln(Pt)
With trend ln(Qt) ln(Pt)
Δln(Qt)
Δln(Pt)
–2.57 –1.09 –1.27 –3.59*** –2.69* –1.88 –1.96 –0.35 –1.49 –1.19 –1.31 –0.54
–2.85* –2.64* –1.08 –1.94 –2.68* –1.70 –2.58* –1.92 –1.69 –0.86 –1.62 –2.45
–2.81 –2.09 –3.52** –3.20* –2.52 –1.41 –2.92 –2.43 –3.00 –2.66 –3.25* –2.63
–4.33*** –5.59*** –6.27*** –7.16*** –5.72*** –4.62*** –5.29*** –5.67*** –5.27*** –7.52*** –6.47*** –4.20***
–5.50*** –5.66*** –5.39*** –4.80*** –3.65*** –3.24** –6.25*** –6.52*** –6.69*** –5.09*** –3.59*** –5.68***
–2.59 –2.56 –2.55 –1.70 –2.45 –2.16 –2.50 –2.91 –3.50** –2.89 –1.58 –2.41
Source: Author's calculations. Note: 1%(***), 5%(**), and 10%(*) critical values are –3.75, –3.00, and –2.63; and –4.38, –3.60, and –3.24 with trend.
After individual regressions, the estimated short-run own-price elasticity was found to be statistically significant at 5 percent in three out of 12 cases: coffee in Tanzania (0.33), cotton in China (0.39), and tobacco in Tanzania (0.59). The adjustment coefficient was statistically significant at 5 percent only in one case, coffee in Tanzania. For this case, the adjustment coefficient was greater than 1, implying a larger supply response to price changes in the short run than in the long run. A linear time trend was found to explain the behavior of tea production in Kenya and coffee production in Vietnam. That only one of the 12 estimated adjustment coefficients is statistically significant is a puzzling finding. In partial adjustment models such as the Nerlove’s, estimated long-run behavior is usually more robust than short-run relationships. In an attempt to make more robust inferences, panel estimation was used. After pooling the data, the panel was first tested for stationarity, since it is of the long form. A few alternatives are available to test panel data for stationarity, but almost all of them require a balanced dataset. One exception is the Fisher’s test suggested in Maddala and Wu (1999). The test basically integrates p-values from individual ADF tests to produce a chi-squared statistic. Table 3A.5 lists p-values from the individual ADF tests. N
∑ ln π , where π is
The chi-squared test statistic is defined as λ = −2
i
i =1
i
the p-value from the ADF for the ith country/commodity pair, with 2N degrees of freedom. N is the number of country/commodity pairs. For