SQU Journal for Science, 2020, 25(2), 130-137 DOI:10.24200/squjs.vol25iss2pp130-137 Sultan Qaboos University 130 A Statistical Analysis on Forecasting Prices of Some Important Food Commodities in Bangladesh Mohammad Abdullah Al-Mamun1, Sheikh Mohammad Sayem1, Khondaker Mohammad Mostafizur Rahman 1 and Mohammad Zakir Hossain 2* 1 Department of Agricultural Statistics, Bangladesh Agricultural University; Mymensingh, Bangladesh. 2 Department of Operations Management and Business Statistics, College of Economics and Political Science, Sultan Qaboos University, P.O. Box 20, Al-Khod, PC 123, Muscat, Sultanate of Oman. *Email: mzhossain@squ.edu.om. ABSTRACT: This paper investigates the best possible forecasting price models for three important agricultural products in Bangladesh namely potato, onion and garlic using time-series and secondary data from January 2000 to December 2014. The main objective of this paper is to find out the appropriate time series models using some of the latest selection criteria that could describe the best price patterns of the above mentioned three crops. To forecast the prices of the crops, the ARIMA models were used, based on model selection criteria and error statistics among the competing models. The overall findings of the study indicate that the fitted models are satisfactory for the respective commodities. The study observed increasing trends in forecasted prices of all three commodities. In particular, the increase in the price of garlic has been observed to be very high compared to that of potato and onion. The study also found that the best fitted SARIMA model for potato is SARIMA (1,0,0) (0,1,2)12, for onion SARIMA (2,0,0) (0,1,1)12, and for garlic SARIMA (2,1,3) (0,1,3)12. Keywords: Box-Jenkins methodology; Forecasting price; Autocorrelation function; ARIMA model; SARIMA model; Estimation; Identification; Akaike information criterion; Bayesian information criterion. بعض المنتجات الغذائية الهامة في بنغالديش للتنبؤ بأسعارتحليل إحصائي مصطفى الرحمن حمدوخندكر م اكر حسينزعبد هللا المأمون وشيخ محمد صايم، ومحمد حمدم والثوم، ممكنة للتنبؤ حول ثالثة منتجات زراعية مهمة في بنغالديش وهي البطاطا والبصل درست هذه الورقة البحثية أفضل نماذج أسعار :صلخمال . تهدف هذه الدراسة إلى ايجاد نماذج متسلسالت زمنية مناسبة، 0202إلى ديسمبر 0222الت زمنية وبيانات ثانوية اعتبارا من يناير لسباستخدام متس ه. من ألل التنبؤ بسسعار باستخدام بعض المعايير المختارة حديثا والتي يمكن أن توصف أفضل أنماط األسعار للمحاصيل الزراعية الثالثة المذكورة أعال استناداً على معايير نماذج مختارة وإحصاء األخطاء بين النماذج المتنافسة. وقد أشارت النتائج (ARIMAالمحاصيل، فقد تم استخدام نماذج أريما ) ة في األسعار المتوقعة لجميع السلع الثالث. وعلى اإللمالية للدراسة إلى أن النماذج المركبة مناسبة للسلع المعنية. والحظت الدراسة أن االتجاهات متزايد مناسب وله الخصوص مادة الثوم، حيث لوحظ أن الزيادة في سعرها مرتفعة للغاية مقارنة بالبطاطا والبصل. وولدت الدراسة أيًضا أن أفضل نموذج وللثوم هو SARIMA(2,0,0)(0,1,1)12صل هو وللب SARIMA(1,0,0)(0,1,2)12للبطاطا هو ( يتم الداللة عليه كما يلي:SARIMAلساريما ) .SARIMA(2,1,3)(0,1,3)12 ، نموذج ساريما (ARIMAأريما ) (، سعر التنبؤ، وظيفة االرتباط التلقائي، نموذجBox-Jenkinsمنهجية صندوق لينكنز ) :مفتاحيةالكلمات ال (SARIMAالتقدير، التعريف، معيار معلوما ،)( ت أكيكAkaike) معيار معلومات ،( بايزBayesian.) 1. Introduction angladesh is predominantly an agriculture-based country where more than 80% of the population are engaged in agriculture, of which 70% are purely in the labor force. The agriculture sector is one of the main important sectors for the economic development of the country which contributes 16.33% of the total gross domestic product (GDP) in Bangladesh (Bangladesh Bureau of Statistics (BBS)) [1]. Potato is one of the most important vegetables in B A STATISTICAL ANALYSIS ON FORECASTING PRICES 131 Bangladesh. It contributes 55% of the country’s total vegetable production [2]. The market price of potatoes decreases from Tk. 1164/100 kg to Tk. 1077/100 kg (Department of Agriculture Marketing (DAM)) [3]. The onion is also a very important spice crop for the people of Bangladesh. The market price of onion increases from Tk. 1403/100 kg to Tk. 2806/100 kg [3]. Garlic is another important spice crop of the country, which is used for medicinal purposes as well. The market price of garlic decreases from Tk. 3730/100 kg to Tk. 2806/100 kg [3]. Forecasting prices of the heavily consumed major commodities is very essential for the businessmen, planners and policymakers of a developing country like Bangladesh where approximately 40% of the people are living below the poverty line. For many developing countries, primary commodities remain an important source of export earnings, and commodity price movements have a major impact on overall macroeconomic performance. Hence, commodity-price forecasting is a key input to macroeconomic policy planning and formulation [4]. The fluctuation of the prices of vegetable and spice crops always makes the government anxious and it has great impact on the millions of the country’s producers and consumers. Early forecasting of the probable prices of vegetable and spice crops could help the policy makers to predict the probable fluctuations in their prices [5]. Forecasting prices of commodities is very important in decision making at all levels and sectors of the economy. This is particularly true in the agriculture sector where policy decisions are characterized by risks and uncertainty, largely due to uncertain yields and relatively low price elasticity of demand for most agricultural commodities in order to make good decisions and policies [6]. The farmers are emotionally and financially affected by the fluctuation in prices of agricultural commodities and its adverse effect on the GDP of a country. Prediction of the prices may help the agriculture supply chain in making necessary decisions in minimizing and managing the risk of price fluctuations [7]. Future prices are also used by crop insurance programs to decide their first-stage and harvest prices [8]. Price forecast therefore, is vital to facilitate efficient decisions and it will play a major role in coordinating the supply and demand of farm products. Hence, forecasting cereal prices will be useful to producers, consumers, processors, rural development planners and other stake holders and agencies/institutions involved in the market [9]. The main objective of this paper is to forecast the monthly prices of the selected three most useful agricultural commodities namely, potato, onion and garlic. Four different models on time series data, namely autoregressive (AR) model, moving average (MA) model, autoregressive integrated moving average (ARIMA) model and seasonal autoregressive integrated moving average (SARIMA) model (popularly known as Box-Jenkins methodology) [10] are extensively used in this study. 2. Data and Methodology Data collection: The monthly secondary data of the prices of the three commodities namely potato, onion and garlic were collected from DAM, BBS and the Food Planning and Monitoring Unit (FPMU) under the Ministry of Food and Disaster Management of Bangladesh. In order to find out the best possible models for forecasting prices of the three items, we used tabular and graphical approaches under descriptive statistics and Box- Jenkins methodology. Time series analysis: Time series analysis was chosen to analyze the data because this particular analysis requires absolute values of forecast, and it usually produces a better result. The ARIMA process is a mathematical model generally used for forecasting. Under this process, the forecasts are based on linear functions of the sample observations in order to find the simplest models that provide an adequate description of the observed data. The time series process, when differenced, follows both AR and MA models and is known as the autoregressive integrated moving averages (ARIMA) model. The model is often abbreviated as ARIMA (p, d, q) where ‘p’ stands for AR part, ‘d’ for integrated part and ‘q’ for MA part. The ARIMA model as used in this study required a sufficiently large data set and involves four steps within the framework of Box-Jenkins methodology, these being identification, estimation, diagnostic checking and forecasting[11]. Identification: T h e m a i n t o o l s i n i d e n t i f i c a t i o n a r e t h e autocorrelation f u n c t i o n ( A C F ) , t h e p a r t i a l autocorrelation function (PACF), and the resulting correlations, which are simply plots of ACFs and PACFs against the lag length. The ACF and PACF are estimated from the sample data. This estimated ACF and PACF are used as a guide to choose appropriate models. The decision regarding transformation is necessary to stabilize the variance of the series through a time plot and it shows data scattered horizontally around a constant mean; the ACF and PACF drop to, or near to, zero quickly which indicates that the data are stationary. If the time plot is not horizontal, or the ACF and PACF do not drop to zero, then non- stationary is to be implied. Estimation: At this stage precise estimates of the coefficients, the AR and MA parameters, seasonal and non-seasonal, of the tentative model chosen at the identification stage have to be determined. In other words, having selected appropriate values for non-seasonal (p, d, q) and seasonal (P, D, Q) of the model, parameters are estimated, typically using simple least squares. Diagnostic Checking: After choosing a particular ARIMA model and having estimated its parameters, the next step is to see whether the chosen model fits the data reasonably well by performing some diagnostic tests. At the diagnostic test one sample test of the chosen model is to see if the residuals estimated from the model are white noise. In this regard, the Portmanteau test can be applied to the residuals as an additional test of fit. The Box-Pierce Q test and Ljung-Box Q tests are the popular portmanteau tests for testing the statistical significance of autocorrelation coefficients. If the portmanteau test is found to be significant, then the model will be inadequate. In such a case, MOHAMMAD ZAKIR HOSSAIN ET AL 132 another ARIMA model needs to be considered. Besides the pattern of significant spikes in the ACF and PACF of the residuals, we cannot improve the model. For example, the significant spikes at the small lags suggest increasing the non-seasonal AR or MA component of the model. Similarly, significant spikes at the seasonal lags suggest adding a seasonal component to the chosen model. Forecasting: If the residuals of the selected model are white noise then the model can be used for forecasting purposes. The reason for the general acceptability of the ARIMA model is its wide successes in forecasting. In this study, we have used the autoregressive (AR) model, moving average (MA) model, autoregressive integrated moving average (ARIMA) model and seasonal autoregressive integrated moving average (SARIMA) in order to find the best suited model and to increase the accuracy of the forecast. 3. The Model The first-order autoregressive disturbance or AR (l) process is of the form [12] 𝑌𝑡 = 𝜌𝑌𝑡−1 + 𝑢𝑡 where 𝑢𝑡 ~ N (0, 1). Similarly, the second-order AR (2) process is of the form 𝑌𝑡 = 𝜌1𝑌𝑡−1+ 𝜌2𝑌𝑡−2 + 𝑢𝑡 . In general, the pth-order AR (p) process is considered as 𝑌𝑡 = 𝜌1𝑌𝑡−1+ 𝜌2𝑌𝑡−2+ ∙∙∙ + 𝜌𝑝𝑌𝑡−𝑝 + 𝑢𝑡 . (1) The first order moving average or MA (1) process is expressed as 𝑌𝑡 = 𝑢𝑡 + 𝜃𝑢𝑡−1. The second- order moving average MA (2) process is of the form 𝑌𝑡 = 𝑢𝑡 + 𝜃1𝑢𝑡−1+ 𝜃2𝑢𝑡−2. The general form of moving average or MA (q) process is considered as 𝑌𝑡 = 𝑢𝑡 + 𝜃1𝑢𝑡−1+ 𝜃2𝑢𝑡−2 + ∙∙∙ + 𝜃𝑞 𝑢𝑡−𝑞 . (2) For time series analysis, the general autoregressive integrated moving average (ARIMA) (P, d, q) model in terms of backward shift operator B is as follows: (1-𝜑1B− ∙∙∙ − 𝜑𝑝𝐵 𝑝)𝑊𝑡 = (1−𝜃1B− ∙∙∙ −𝜃𝑞 𝐵 𝑞 )𝐴𝑡 (3) where, 𝑤𝑡 = (1 − 𝐵) 𝑑 𝑒𝑡 is the first difference of the original time series 𝑌𝑡 and 𝐴𝑡 is the random shock which forms a white noise process with mean zero [13]. Similarly, the seasonal ARIMA, often called SARIMA (p, d, q) (P, D, Q)s in terms of the backward shift operator B can be expressed as (1−𝜑𝑠 𝐵 𝑠 −∙∙∙ −𝜑𝑠𝑝𝐵 𝑠𝑝 ) 𝑊𝑡 = (1−𝜃𝑠 𝐵 𝑠 −∙∙∙ −𝜃𝑠𝑄 𝐵 𝑠𝑄 ) 𝐴𝑡 (4) where, 𝑤𝑡 = (1 − 𝐵 𝑠 )𝑑 𝑌𝑡 , s = 12 for monthly data and s = 4 for quarterly data. Contrary to (1.3), here the random shocks 𝐴𝑡 do not form a white noise process. Combining (1.3) and (1.4) we obtain (1-𝜑1B− ∙∙∙ − 𝜑𝑝𝐵 𝑝) (1−𝜑𝑠 𝐵 𝑠 −∙∙∙ −𝜑𝑠𝑝𝐵 𝑠𝑝 )𝑊𝑡 =(1−𝜃1B− ∙∙∙ −𝜃𝑞 𝐵 𝑞 ) (1−𝜃𝑠 𝐵 𝑠 −∙∙∙ −𝜃𝑠𝑄 𝐵 𝑠𝑄 ) 𝐴𝑡 . (5) As a Final generalization, a constant term 𝜃0 needs to be added to the model (1.5) in order to accommodate the possibility that the variables 𝑤𝑡 may have a non-zero mean. Thus the resulting model can be written as (1-𝜑1B− ∙∙∙ − 𝜑𝑝𝐵 𝑝) (1−𝜑𝑠 𝐵 𝑠 −∙∙∙ −𝜑𝑠𝑝𝐵 𝑠𝑝 )𝑊𝑡 = 𝜃𝑜+(1−𝜃1B− ∙∙∙ −𝜃𝑞 𝐵 𝑞 ) (1−𝜃𝑠 𝐵 𝑠 −∙∙∙ −𝜃𝑠𝑄 𝐵 𝑠𝑄 ) 𝐴𝑡 (6) where, 𝑤𝑡 =(1 − 𝐵) 𝑑 (1 − 𝐵𝑠 )𝐷 𝑌𝑡 . Therefore, the equation (1.6) stands as (1-𝜑1B− ∙∙∙ − 𝜑𝑝𝐵 𝑝)(1−𝜑12𝐵 12 −∙∙∙ −𝜑12𝑝 𝐵 12𝑝 )𝑊𝑡 = 𝜃𝑜+(1−𝜃1B− ∙∙∙ −𝜃𝑞 𝐵 𝑞 ) (1−𝜃12𝐵 12 −∙∙∙ −𝜃12𝑄 𝐵 12𝑄 ) 𝐴𝑡 (7) where, 𝑤𝑡 =(1 − 𝐵) 𝑑 (1 − 𝐵12)𝐷 𝑌𝑡 . This is the multiplicative model of order (p, d, q) (𝑃, 𝐷, 𝑄)12. Here the term (1-𝜑1B− ∙∙∙ − 𝜑𝑝 𝐵 𝑝) is known as the regular autoregressive operator of order p, the term (1−𝜑12𝐵 12 −∙∙ ∙ −𝜑12𝑝𝐵 12𝑝) is known as seasonal autoregressive operator of order p, the term (1−𝜃12𝐵 12 −∙∙∙ −𝜃12𝑄 𝐵 12𝑄 ) is the seasonal moving average operator of order Q. The multiplicative model (1.7) represents a common form of most of the seasonal time series models considered in practice. The best model is obtained with the following diagnostics, by lowest values of Akaike’s information criteria (AIC) and Schwartz Bayesian criteria (SBC or BIC). To check the adequacy for the residuals, Q statistic is used. A modified Q statistic is the Box-Ljung Q statistic as given below: Q'= n (n+2)∑ 𝑟𝑘² (𝑛−𝑘) 𝑝 𝑘=1 . The Q statistic is compared to the critical value of chi-square distribution. If the p-value associated with the Q statistic is small, the model is considered as adequate. Forecasting the future periods using the parameters for the tentative model has been selected. Trend fitting: For evaluating the adequacy of AR, MA, ARIMA and SARIMA processes, various reliability statistics like R 2 , Root Mean Square Error (RMSE), Mean Absolute Percent Error (MAPE), Mean Absolute Error (MAE) and BIC were used. The smaller the various reliability statistics, the better the efficiency of the model in predicting the future production. A STATISTICAL ANALYSIS ON FORECASTING PRICES 133 4. Results and Discussion Accuracy of forecasting depends on the time series data which must be stationary. Apart from the graphical method of using ACF and PACF for determining whether the time series is stationary, a very popular method of determining this is the Augmented Dickey Fuller (ADF) test. In the present study the ADF tests of the three different agricultural commodities such as potato, onion and garlic prices were conducted using EViews software. The most suitable models were selected based on their ability for reliable prediction. Lower values of RMSE and MAPE were preferable whereas for Normalized BIC, higher values were preferable. Furthermore, the Ljung-Box test (portmanteau test) was conducted to see if the residual ACF at different lag times was significantly different from zero, where not being different from zero was expected. After the best model was identified, forecasts for future values from January 2000 to December 2014 were made. The best fitting model was determined for the three different commodities based on secondary data from January 2000 to December 2014 by using the statistical software SPSS 20. From the following tables, it can be seen that the best SARIMA model for forecasting the wholesale price of potato is SARIMA (1,0,0)(0,1,2)12, for onion it is SARIMA (2,0,0)(0,1,1)12 , and for garlic it is SARIMA (2, 1, 3)(0,1,3)12. In the three best models, the ACF and PACF of residuals have no significant spikes and the residuals are found to be white noise. After fitting the best selected models, the prices of three selected commodities were forecasted for January 2015 to December 2015 based on the collected secondary data from January 2000 to December 2014. Overall, the forecast prices of the selected commodities were found to be consistent with some few upturns and downturns of the observed series. Table 1. Model Selection Criteria for Tentatively Selected SARIMA Models for Onion. Model R 2 RMSE BIC MAE MAPE Ljung-Box(Q- statistics) P-value SARIMA(2,0,0)(0,1,1)12 .853 478.97 12.49 329.02 16.21 19.88 .176 SARIMA(2,0,1)(0,1,2)12 .854 480.24 12.56 327.56 16.15 18.74 .131 SARIMA(2,0,1)(0,1,3)12 .853 483.14 12.60 328.94 16.23 16.87 .155 SARIMA(2,0,1)(0,1,4)12 .855 480.84 12.62 325.26 16.03 16.22 .133 SARIMA(3,0,1)(0,1,2)12 .857 476.40 12.57 327.78 16.26 13.87 .309 SARIMA(3,0,1)(0,1,3)12 .858 475.81 12.60 319.63 15.78 13.11 .286 Table 2. Model Parameters of SARIMA (2, 0, 0) (0,1,1)12 for Onion. Type Coefficient Standard error P- value AR(1) 1.1364 .0724 0.00 AR(2) -0.3923 .0729 0.00 SMA(12) 0.8587 .0615 0.00 Constant 39.319 6.785 0.00 Table 3. Model Selection Criteria for Tentatively Selected SARIMA Models for Garlic. Model R 2 RMSE BIC MAE MAPE Ljung- Box(Q- statistics) P- value SARIMA(1,1,4)(1,1,5)12 .904 973.622 14.16 519.00 10.86 10.00 .188 SARIMA(2,1,5)(1,1,5)12 .906 971.621 14.21 511.59 10.61 8.11 .150 SARIMA(2,1,3)(0,1,3)12 .906 953.208 14.02 513.93 10.64 10.54 .394 SARIMA(2,1,4)(0,1,3)12 .905 964.886 14.08 519.80 10.84 7.81 .553 SARIMA(2,1,4)(0,1,4)12 .905 968.313 14.11 518.50 10.81 7.77 .456 SARIMA(2,1,4)(0,1,5)12 .907 959.466 14.13 506.96 10.48 7.24 .404 Table 4. Model Parameters of SARIMA (2, 1, 3) (0,1,3)12 for Garlic. Type Coefficient Standard error P-value AR(1) 0.3127 0.0730 0.00 AR(2) -0.8822 0.0631 0.00 MA(1) 0.2857 0.1000 0.00 MA(2) -O.8862 0.0629 0.00 MA(3) -0.1551 0.0856 0.07 SMA(12) 0.9894 0.0822 0.00 SMA(24) 0.1812 0.1276 0.15 SMA(36) -0.3005 0.1105 0.00 Constant 16.87 -0.18 0.85 MOHAMMAD ZAKIR HOSSAIN ET AL 134 Table 5. Model Selection Criteria for Tentatively Selected SARIMA Models for Potato. Model R 2 RMSE BIC MAE MAPE Ljung-Box(Q- statistics) P- value SARIMA(0,1,1)(0,1,1)12 .829 217.159 10.884 138.89 13.70 16 .453 SARIMA(0,1,2)(0,1,2)12 .844 208.651 10.865 132.20 13.12 7.78 .900 SARIMA(0,1,3)(0,1,2)12 .844 209.210 10.901 132.41 13.14 7.69 .863 SARIMA(0,1,4)(0,1,3)12 .846 209.102 10.961 131.46 13.21 7.40 .765 SARIMA(0,1,5)(0,1,3)12 .846 209.658 10.997 131.35 13.20 7.46 .681 SARIMA(1,0,0)(0,1,2)12 .848 205.002 10.779 134.58 13.22 10.84 .764 Table 6. Model Parameters of SARIMA (1, 0, 0) (0,1,2)12 for Potato. Type Coefficient Standard error P- value Constant 6.080 1.704 0.00 AR(1) 0.893 0.035 0.00 SMA(12) 1.268 0.086 0.00 SMA(24) -0.3608 0.091 0.00 Figure 1. Residual of ACF and PACF for the best SARIMA models for Potato, Garlic and Onion. The actual and fitted prices using the best fitted model of the respective commodities are presented below for one year (March 2009 – February 2010) to check the validity of the models employed in our study. Table 7. Comparison of Actual and Fitted price of the Commodities per 100 kg. Potato Onion Garlic Month Actual price Fitted price Actual price Fitted price Actual price Fitted price Jan-14 632 1068 3430 5703 7898 7767 Feb-14 1514 548 1938 1896 6951 7695 Mar-14 784 1579 2166 1535 4658 4516 Apr-14 1124 983 2156 2208 4650 4550 May-14 1264 1329 1928 2624 4522 5683 Jun-14 1511 1406 2825 2071 5795 5034 Jul-14 1713 1619 3251 3496 6279 6107 Aug-14 1756 1768 3380 3581 5812 6370 Sep-14 1760 1808 3472 3584 6853 5824 Oct-14 1847 1830 3252 3900 7566 7367 Nov-14 1850 1893 3299 3554 7218 7731 Dec-14 1952 1818 2752 3003 6970 7083 A STATISTICAL ANALYSIS ON FORECASTING PRICES 135 Figure 2. Plot of Actual and Predicted Prices of Potato, Onion and Garlic. The forecasted prices of the selected commodities are given below Table 8. 0 1000 2000 3000 4000 5000 6000 O N IO N P R IC E MONTH Actual Price Predicted price 0 2000 4000 6000 8000 10000 G A R LI C P R IC E MONTH Actual Price Predicted price 0 500 1000 1500 2000 2500 P O T A T O P R IC E MONTH Actual Price Predicted price MOHAMMAD ZAKIR HOSSAIN ET AL 136 Table 8. Forecasted Prices per 100 kg of Potato, Garlic and Onion. Forecasted price per 100kg Month Potato Onion Garlic Jan-15 1700 2167 7034 Feb-15 1170 2360 6306 Mar-15 1370 2579 4339 Apr-15 1454 2546 4607 May-15 1600 2896 5159 Jun-15 1664 3099 5364 Jul-15 1738 3453 5793 Aug-15 1780 3730 6231 Sep-15 1799 3932 6589 Oct-15 1822 4280 6905 Nov-15 1940 4554 6858 Dec-15 1898 4038 6655 4. Conclusion During the last few decades, a huge amount of work has been done by using time series data on the major crops of Bangladesh such as rice, wheat, tea, jute, lentil, etc. See for example, [14] and [15]. We used three important food crops, i.e. potato, onion and garlic, for our analysis. We observed from Table 8 that the forecasted price for potato rose from Tk.1700 per 100 kg in January 2015 to Tk.1898 by the end of the year 2015. For onion, the forecasted price of onion rose from Tk. 2167 per 100 kg in January 2015 to Tk. 4038 by the end of the year 2015. The table reveals that the price fluctuations had erratic trends in nature. The forecasted price of garlic decreased greatly from Tk. 7034 per 100 kg in January 2015 to Tk. 5793 in July 2015. After this, the price slightly increased from Tk. 5793 per 100 kg in July 2015 to Tk. 6655 in December 2015. Based on the above numerical figures, we may conclude that the overall prices for the selected three important food commodities are expected to increase in the next one year. This could be very helpful information to businessmen, policy makers and planners in order to make future economic decisions regarding these types of agricultural food commodities. The seasonal autoregressive integrated moving average model traces out the seasonal effect of the desired variable. The current research identified SARIMA (1,0,0) (0,1,2)12 for potato, SARIMA (2,0,0) (0,1,1)12 for onion and SARIMA (2,1,3) (0,1,3)12 for garlic have been proved to be the best possible models for forecasting purposes on the basis of the latest model selection criteria. The forecasting performances of the chosen models were found to be satisfactory, as shown by Figure 2. As we know, more reliable results on forecasting accuracy mainly depend on accuracy of data on the selected variables. Thus, we recommend that data banks in Bangladesh should be better organized and of better quality in order to obtain the best possible outcomes through forecasting models. Conflict of Interest The authors declare no conflict of interest. Acknowledgment MZH thanks the College of Economics and Political Science for infrastructural support. The authors would like to thank the anonymous reviewers for their constructive suggestions which helped to improve the quality of the paper. References 1. Bangladesh Bureau of Statistics. Yearbook of Agricultural Statistics of Bangladesh, Ministry of Planning, Government of the People’s Republic of Bangladesh, Dhaka, 2014. 2. Bangladesh Bureau of Statistics. Yearbook of Agricultural Statistics of Bangladesh, Ministry of Planning, Government of the People’s Republic of Bangladesh, Dhaka, 2009. 3. Department of Agricultural Marketing. The monthly data collection of the selected commodity (potato, onion, garlic) prices in Bangladesh from January 2000 to December 2014, Dhaka, 2014. 4. Chakriya, B. and Husain, A. Forecasting commodity prices: futures versus judgment”, IMF Working Paper 04/41, 2004. 5. Hassan, M.F., Islam, M.A., Imam, M.F. and Sayem, S.M. Forecasting wholesale price of coarse rice in Bangladesh: a seasonal autoregressive integrated moving average approach. A working paper, 2013. 6. Ali, M.Z., Hossain, M.Z. and Samad, Q.A. Historical, Ex-Post and Ex-Ante forecasts for three kinds of pulse prices in Bangladesh: masur, gram and khesari. Jahangirnagar University Journal of Science, 2002, 25, 255-265. A STATISTICAL ANALYSIS ON FORECASTING PRICES 137 7. Kiron, M. Sabu, T.K. and Manoj, K. Predictive analytics in agriculture: forecasting prices of arecanuts in Kerala, Procedia Computer Science, 2020, 171, 699-708. 8. Hongbing, O., Xiaolu, W. and Qiufeng, W. Agricultural commodity future prices prediction via long- and short- term time series network, Journal of Applied Economics, 2019, 22(1), 468-483. 9. Jadhav, V., Chinnappa, R.B.V. and Gaddi, G.M. Application of ARIMA model for forecasting agricultural prices, Journal of Agricultural Science and Technology, 2017, 19(5), 981-992. 10. Box, G.E.P. and Jenkins, G.M. Time Series Analysis: Forecasting and Control, Holden-Day, San Francisco, CA, 1976. 11. Schwarz, G. Estimating the dimension of a model. Annals of Statistics, 1978, 6(2), 461-464. 12. Makridakis, S., Wheelwright, S. and Hyndman, R.J. Forecasting Methods and Applications. 3rd edition, John Willey and Sons, New York, 1998. 13. Pankratz, A. Forecasting with Univariate Box-Jenkins Models: Concepts and Cases, John Wiley, New York, 1984. 14. Hossain, M.Z., Samad, Q.A. and Ali, M.Z. ARIMA model for forecasting the prices of field pea, blackgram and mungbean in Bangladesh. Bangladesh Journal of Agricultural Research, 2004, 29(2), 289-303. 15. Samad, Q.A. An econometric model for the world market of tea. Ph.D. thesis, University of Belefield, Germany. Received 20 February 2019 Accepted 1 st July 2020