MODELING THE IMPACT OF CLIMATE VARIABILITY ON LOCAL FOOD PRICES - Food Security, Food Prices and Climate Variability - page 141

20

10

0

-10

-20

2008
m1

2009
m1

2010
m1

2011
m1

2012
m1

Markov
NDVI anomaly effect

Seasonal

FIGURE 7.4 Seasonal and Markov price components and NDVI anomaly for Nairobi, Kenya.

components used in this study. The Markov trend and seasonal components are linked with
the NDVI anomaly information to capture the information in the weather shock with the
analysis.
By using satellite-derived vegetation data to account for seasonal variations as well as esti-
mates of drought on changes in price levels (Deaton and Laroque, 1992; Brown et al ., 2008),
we estimate local price dynamics by incorporating seasonality (using monthly dummies), an
autoregressive process and a trend (where applicable). We also incorporate explanatory vari-
ables as shocks to this system (Harvey, 1989). The autoregressive process has a time varying
error. We estimate the impact of two explanatory variables - world prices for the relevant
commodity and normalized difference vegetation index anomalies. These anomalies are cal-
culated only for those months for which the ten-year monthly average is at least half as large
as the average of the highest month. This allows us to restrict possible impacts to anomalies in
months that are likely to influence local food prices. We use three month lags for each and
the total impact for each factor is evaluated by summing up the coefficients.
By explicitly measuring weather anomalies and world price changes as shocks to a
dynamic system, we reduce the possibility that price changes coincidentally move with
external shocks. For example, by explicitly modeling seasonality, we are eliminating the
possibility that estimated impact of a world price increase merely reflects coincidental
seasonal movements ( Figure 7.5 ).
A weakness we share with other approaches is our inability to measure transport costs. We
could either assume stationary and proportional transport costs or ignore them. We chose the
latter because it is clear that transport costs are not stationary for the countries that we exam-
ined. In future work, we would like to measure transport costs using additional data on dis-
tance and local fuel prices, and incorporate transport costs and the attendant non-linearity into
the estimation methodology.