02 - 106080
TRANSCRIPT
13th Agricultura
lResearch Symposiu
m2014
Department of Agribusiness ManagementFaculty of Agriculture and Plantation Management
Wayamba University of Sri Lanka
Forecasting Paddy Production of Batticaloa District in Sri Lanka:
Linear Time Series Models
L.H.A.M.Silva106080
Content
• Introduction• Objectives• Methodology• Results and Discussion• Conclusion• Acknowledgements• References
IntroductionPaddy– The staple food of Sri Lanka– Contributed 1.6% to the Gross Domestic Production
in year 2013
• Highest paddy producing districts
District Production (t)Ampara 297,229Polonnaruwa 261,263Kurunegala 189,281Anuradhapura 161,406Hambanthota 112,623Batticaloa 111,943
• Batticaloa district – 6th position
Forecasting– Process of making statements about future events
which actual outcomes have not yet been observed
– Forecasting is important to• Government• Policy makers• Intermediaries• Consumers
– For• Planning• Decision making
• Forecasting paddy yield is a challenge
– As the yield depends on• External factors• Internal factors
• Climate has a close relationship with yield
• Climate factors such as– Rainfall– Day length– Relative humidity– Temperature
• Climate factors fluctuates rapidly• Lack of continuous & accurate climate data
• One of the best approaches - Time series models
Objectives
Study was conducted toIdentify accurate linear time series models to forecast paddy production of Batticaloa district
MethodologyData collection
• Seasonal time series data• From year 1980 to 2013 on paddy production• From official website of Department of Census
and Statisticshttp://www.statistics.gov.lk/
Analysis
• Time series plot
• Trend models– Linear– Quadratic– Exponential Growth– Pearl-Reed Logistic (S-Curve)
• Time Series Models– Single Exponential Smoothing– Double Exponential Smoothing– Winters’ Method– ARIMA Models
• Minitab version 15 was used
Model selection and validation• Model Selection Criteria
Mean Absolute Percentage Error (MAPE)
Where,
PEt = Percentage error at t timeYt = Observed value at t timeFt = Forecasted value at t time
• Model validationResidual Analysis
–Autocorrelation function of the residual
–Anderson – Darling test
–Run chart
Results and Discussion
1980/1981
1981/1982
1982/1983
1983/1984
1984/1985
1985/1986
1986/1987
1987/1988
1988/1989
1989/1990
1990/1991
1991/1992
1992/1993
1993/1994
1994/1995
1995/1996
1996/1997
1997/1998
1998/1999
1999/2000
2000/2001
2001/2002
2002/2003
2003/2004
2004/2005
2005/2006
2006/2007
2007/2008
2008/2009
2009/2010
2010/2011
2011/2012
2012/20130
50000
100000
150000
200000
250000
Year
Prod
uctio
n (t
)
Maha season• Maximum (2009/2010) – 193,274 t• Minimum (1987/1988) – 17,105 t
• Average – 84,638 t• Standard deviation – 40,220 t
19801982
19841986
19881990
19921994
19961998
20002002
20042006
20082010
20120
20000
40000
60000
80000
100000
120000
Year
Prod
uctio
n (t
)Yala season• Maximum (2013) – 111,943 t• Minimum (2007) – 7,000 t
• Average – 42,920 t• Standard deviation – 21,164 t
• A gradual increment after year 2008– May be due to recovery from civil war
Trend analysis
Fitted model MAPE Value
Linear 70Exponential Growth 50Quadratic 69Pearl-Reed logistic (S curve) 49
• Best trend model with lowest MAPE
Pearl-Reed logistic (S curve)
For both Yala and Maha seasons
𝑌 𝑡=106 /(26.3162−(0.592015×1.04193 𝑡))
For Yala Season
Fitted model MAPE Value
Linear 47Exponential Growth 42Quadratic 44Pearl-Reed logistic (S curve) 43
• Best trend model with lowest MAPE
Exponential growth trend model
For Maha Season
Fitted model MAPE Value
Linear 50Exponential Growth 48Quadratic 51Pearl-Reed logistic (S curve) 46
• Best trend model with lowest MAPE
Pearl-Reed logistic (S curve)
• Single exponential smoothing was better than double exponential smoothing
• Single exponential smoothing method can only forecast one period ahead
Exponential smoothing models
Single exponentialsmoothing
Double exponentialsmoothing
α=0.101MAPE = 64
α=0.1, ɤ=0.01MAPE = 69
Time Series Models
Holt-Winter’s Trend and Seasonality Model (Winters’ method)
• Best fitted Winters’ model– Seasonal length-12–Multiplicative model–α (Level) = 0.8–ɤ (Trend) =0.01–Δ (Seasonal) =0.01
• MAPE = 35• Identification of outliers
– Standardized residual
• Adjusting outliers–3 month simple moving average method–Reduced MAPE from 46 to 35
• Residual analysis– Run chart – P value for clustering – 0.806
– Anderson-Darling test – P value obtained 0.262
– Autocorrelation function
ARIMA models• Best fitted ARIMA model
–ARIMA 111• MAPE = 43.9• Identification of outliers
–Standardized residual• Adjusting outliers
–3 month simple moving average method–Reduced MAPE from 68.8 to 43.9
• Residual analysis– Run chart – P value for clustering – 0.228
– Anderson-Darling test – P value – 0.110
– Autocorrelation function
ConclusionModel MAPE
Single exponential smoothing 64
Double exponential smoothing 69
Winters’ method 35
ARIMA 43.9
Year-Season Obs. Fitted / forecast values
2011-Yala 77,196 61,2612011/12-Maha 171,715 152,7552012-Yala 83,599 79,3062012/13-Maha 115,630 141,0282013-Yala 111,943 72,6272013/14-Maha 158,6952014-Yala 105,4812014/15-Maha 213,9642015-Yala 91,8342015/16-Maha 266,901
Fitted and forecast values using the Winter’s modelForecast
If the prevailing conditions remain – An increment in paddy production
Acknowledgments
All the academic and non-academic staff members of Faculty of Agriculture and
Plantation Management
References• Agrawal, R., Jain, R. C., Jha, M. P. and Singh, D. (1980) Forecasting
of rice yield using climate variables. In Indian Journal of Agricultural Science, 50(9), 680-684.
• Anon. (2013). Annual Report (2013). Central Bank of Sri Lanka available from: www.cbsl.gov.lk. (Accessed 28th April 2013).
• De Datta, S.K. (1981). Principles and Practices of Rice Production. John Wiley and Sons, Inc.
• Raghavender, M. (2010). Forecasting paddy production in Andhra Pradesh with ARIMA model. In: International Journal of Agricultural and Statistics Sciences, 6(1), 251-258.
• Rahman, N.M.F. (2010). Forecasting of bro rice production in Bangladesh: An ARIMA approach. In: Journal of Bangladesh Agricultural University. Available from: http://www.banglajol.info/index.php/jbau/article/download/6406/4901(Accessed 20th March 2014).
• Sivapathasundaram, V., and Bogahawatte, C. (2012). Forecasting of paddy production in Sri Lanka: A time series analysis using ARIMA model. In: Tropical Agricultural Research, 24 (1), 21-30.
• Thattil, R.O., and Walisinghe, W.M.P.K. (2000). Forecasting Paddy Yields. Available from: http://www.goviya.lk/agrilearning/Paddy/Paddy_Research/Paddy_pdf/SE2.pdf. (Accessed 20th June 2014)
• Wheelwright, S. C. and Hydman, R. J. (1998) Forecasting methods and applications. eds. Makridakis, S. European Institute of business administration (INSEAD), 3rd ed. 146 – 169.