This document discusses reducing basis risk in index-based crop insurance for a semi-arid agricultural region in Kazakhstan. It explores using quantile regression to model the relationship between crop yields and weather indices like precipitation and temperature. Quantile regression may better capture the impact of extreme weather events on yields compared to ordinary least squares. The document also examines choosing appropriate time periods and weather-based indices to minimize the difference between actual losses and insurance payouts.
Artificial intelligence in the post-deep learning era
Sarah CONRADT "Reducing meteorological basis risk in a semi-arid agricultural region"
1. Reducing meteorological basis risk in a
semi-arid agricultural region
THU 2.3: DLDD and climate change (ID 344)
Presented by: Sarah Conradt R. Bokusheva
3. Crop insurances
Indemnity-based
Damage / insured yield
Index-based
(weather) index
e.g.
Indemnity
payment
Rainfall [mm]
4/12/2013 DUSYS/IED/AFEE 3
4. Basis risk
Which indices are suitable
for (semi-) arid regions?
How?
Precipitation
Ordinary Least Squares
Temperature
Quantile Regression
Soil moisture
Time period?
…
4/12/2013 DUSYS/IED/AFEE 4
12. Appendix: Weather indices & Optimization
Time period constant for all years
Variable for different farms and indices
More ‘advanced’ model? (external weather conditions)
Freitag, 12. April 2013 DUSYS/IED/AFEE 12
13. Appendix: Study region
Table 1 Summary statistics of the 47 farm data
Number Average yield Min. Max. Average Min. Max.
CV sown
Rayon of 1980-2010 yield [0.1 yield [0.1 CV yield sown area sown area sown area
area
Farms [0.1 t /ha] t /ha] t /ha] [ha] [ha] [ha]
R1 12 8.9 0.2 24.0 0.44 13’599 805 24’700 0.43
R2 11 8.8 0.8 21.0 0.43 16’900 800 34’073 0.41
R3 7 8.3 1.2 19.3 0.42 15’316 500 30’750 0.49
R4 10 10.7 0.9 25.6 0.47 14’720 1155 10’940 0.44
R5 7 9.2 0.3 22.1 0.43 19’666 2000 82’850 0.65
CV: coefficient of variation. Source: Data from the regional statistical offices of Kazakhstan.
Table 1 Summary statistics: weather indices (1980-2010)
Average Average Number of years
Rayon Min. CP Max. CP CV CP Min. SI Max. SI CV SI
CP SI where SI < 0.7
R1 140.9 94 215 0.26 0.72 0.26 1.38 0.38 14
R2 132.7 33 234 0.34 0.71 0.14 1.57 0.47 19
R3 126.5 46 267 0.38 0.65 0.22 1.74 0.55 20
R4 163.5 83 269 0.35 0.87 0.30 1.93 0.50 13
R5 147.7 70 297 0.39 0.75 0.22 1.82 0.48 16
CV: coefficient of variation, CP: cumulative precipitation [mm], SI: Selyaninov index. Source: Data from the
National Hydro-Meteorological Agency of Kazakhstan.
4/12/2013 DUSYS/IED/AFEE 13
Hinweis der Redaktion
Also a warm welcome from my side. I will talk today Agriculture production in a semi-arid regionMy short talk today deals with agriculture in a arid region: namellyKazakshstan.
K. represents one the most important wheat producers in the world and regularly faces extreme droughts that lead to substantial yield losses. (semi-) arid regions, systemic weatherevents, such as droughts, dry windsHow to stabilize income of farmers?
2 basic groups: Damage-based indemnity insurance is crop insurance in which the insurance claim is calculated by measuring the percentage damage in the field soon after the damage occurs. With index insurance products, payments are based on an independent measure highly correlated with farm-level yieldrealizations of a specific weather parameter Here the indemnity is based on realizations of a specific weather parameter measured over a prespecified period of time at a particular weather station. Weather index based: explain: indemnification payments are triggered by a specific pattern of an index and not by actual yield or in field assessment. Strong dependency b/W yield and index is necessary. Certain advantages: Asym. Info reduced: moral hazard (The term defines a situation where behaviour of one party change in a detremental way after buying insurance) and adverse selection. Lower transaction costs, covariate risk exposure.
Basis risk: is the mismatch between the index realization actual yield realization. Thus you get a payment even if you had a good year with high yields or the inverse where you had very low yields but you get no indemnity payment. inefficientCopulas: describedependence b/w random variables; marginal distributions and dependency by copulaNon linear dependency structures can be modelled, Copulas: regression analysis based on or assumes linear correlation (multivariate normal distribution fine); tail dependency downside risk; marginal distributions to form joint distribution.
90%confintSkewed yield distribution (median)Calculate the standard errors or conf. intervalls, I used a bootstrap approach (Monte Carlo method where size n samples are taken from observed data with replacement)Different: - Conditional mean function (conditional mean of a response variable!) vscondquantilefctDifferent assumptions on error termsLRM: error identically independently and normally distributed with mean zero, unknown variance sigma2: Homoscedasticity: conditional Var(Y|x) is a constant sigma2 for all values of covariatesConditional mean of y given x E[y|x], i.e. average of y values corresponding to a fixed value of covariate x (how location of conditional distr. behaves by utilizing mean of distrib. to represent central tendency)HERE: average yield given weather conditionsQR: minimize average weighted distance, with weighting depending on wheather points are above/below qMonotone equivariance: (Q(h(y)|x) = h Q[y|x] not the case for LRM
Critical periods in that region for that crop
Both: - continuous response variable, response variable is linear in unknown parametersDifferent: - Conditional mean function (conditional mean of a response variable!) vscondquantilefctDifferent assumptions on error termsLRM: error identically independently and normally distributed with mean zero, unknown variance sigma2: Homoscedasticity: conditional Var(Y|x) is a constant sigma2 for all values of covariatesConditional mean of y given x E[y|x], i.e. average of y values corresponding to a fixed value of covariate x (how location of conditional distr. behaves by utilizing mean of distrib. to represent central tendency)HERE: average yield given weather conditionsQR: minimize average weighted distance, with weighting depending on wheather points are above/below qMonotone equivariance: (Q(h(y)|x) = h Q[y|x] not the case for LRM
Weather index based: explain: indemnification payments are triggered by a specific pattern of an index and not by actual yield or in field assessment. Strong dependency b/W yield and index is necessary. Certain advantages. Asym. Info reduced: moral hazard (The term defines a situation where behaviour of one party change in a detremental way after buying insurance) and adverse selectionK. represents one the most important wheat producers in the world and regularly faces extreme droughts that lead to substantial yield losses. (semi-) arid regions, systemic weatherevents, such as droughts, dry windsDetrending remove deterministic component; comparable; remove component which influenced yield level over timeCopulas: regression analysis based on or assumes linear correlation (multivariate normal distribution fine); tail dependency downside risk; marginal distributions to form joint distribution.Priors: posterior distribution (prior info used, combine with observed data)