Livestock Research for Rural Development 32 (5) 2020  LRRD Search  LRRD Misssion  Guide for preparation of papers  LRRD Newsletter  Citation of this paper 
Bangladesh dairy has been emerging as a transformation from livelihoodoriented dairy toward enterprisedriven dairy while it can still be seen as a key source of income and livelihoods for millions of people. There is an increasing emphasis by the government for achieving selfsufficiency. However, there have been challenges as the lack in authentic milk production data availability which counteract in estimating the selfsufficiency. There is a growing demand for milk along with the increasing consumer’s preferences on highquality milk. The objective of this study was to forecast Bangladesh milk production to analyze the possible timeframe for achieving the selfsufficiency. This study applies Autoregressive Integrated Moving Average (ARIMA) model and compares with the classical Trend Analysis Model (TAM) with three scenarios i) Exponential growth, ii) Compounding Average Growth Rate (CAGR), iii) Simple Average Growth Rate(SAGR), built in MS Excel using time series milk data (2006 to 2019) obtained from the Department of Livestock Services (DLS), Integrated Dairy Research Network (IDRN) Sector database and International Farm Comparison Network (IFCN) Sector database.
The results revealed the superiority of ARIMA model over the trend analysis. The maximum milk production of Bangladesh in 2025 and 2030 will be 10.24 and 13.65 million tons, respectively (according to IDRN data) while that will be 16.32 and 17.13 million ton, respectively according to DLS. Considering the only population as demand factor (deterministic approach), taking the results from the ARIMA model, Bangladesh will be selfsufficient in 2029 according to DLS data and according to IDRN data, Bangladesh will achieve 71% selfsufficient in 2029. The forecasting was calculated only with normal year with some mild shock in the milk production. However, the impact of severe shocklike Coronavirus in 2020 has not been possible to integrate due to anecdotal in data, which implies that including such shock in future this forecasting might lead to the different pace of the selfsufficiency achievement. The results of ARIMA model using DLS data was substantially different than IDRN data which might need to take into account to identify why those data were different and might revise the data backwards to increase the degree of precision for estimating the future milk production data
From this study, it is evident that under normal condition, Bangladesh could reach to the selfsufficiency in 2029 but the year like 2020 (with pandemic coronavirus) need to be observed in the future and might need strategic decision to continue the current growth rate. The precise documentation of the milk production data would lead to the right path toward selfsufficiency. The future study might focus on the stochastic demand and supply factor for selfsufficiency analysis.
Keywords: ARIMA, coronavirus, IDRN, IFCN, milk production, selfsufficiency, and Trend analysis
Dairying in Bangladesh has been transforming from traditional subsistence to more marketoriented and enterprisedriven approach in the dairy production system (Uddin et al., 2020) which would open the opportunity for dairy farmers to exploit the rising demand on milk and milk products at national as well as global dairy markets. Bangladesh has milk production of 9.92 million tons in 2018 (DLS, 2019). In contrast, the Integrated Dairy Research Network (IDRN) estimated for 2018 as 7.98 million tons positing 23rd in the global milk production while the topmost milk producing in the world is India with estimated 201.22 million ton of milk production (IFCN, 2019). The second milk producing country in the world is the United States of America with the estimated milk production of 95.31 million ton. The third and fourth in milk production were Pakistan and Germany corresponding to 48.36 and 33.18 million ton, respectively.
The milk availability per capita is reported as 165.07 ml/day/capita against the minimal requirement of 250 ml/day/capita (DLS, 2019). The selfsufficiency according to DLS is 64% while the same is 53% according to IDRN. Bangladesh needs a substantial increase in milk production to be selfsufficient while India and Pakistan are selfsufficient with an average consumption of 430 and 670 ml/day/capita, respectively.
To increase the availability of milk at the consumers level might depend on the supply of adequate quality of milk with affordable price which can be done ensuring the responsible actions by the topmost processors, dairy farmers and other marketing agencies. Concerning the processing profile in Bangladesh, only 9% of the total milk production is delivered to the processors and the remaining 91% is traded as informally (IFCN, 2019). The major market players in the country are Milk vita, Pran Dairy Ltd, BRAC Dairy and Food (Arong) and Akij Dairy Ltd (Farm fresh) corresponding to only 5% share to the total milk production in the country.
Along the supply chain of milk production to consumption, milk price is one of the major factors for dairy sector competitiveness (Roland et al., 2016) because of its huge influence on the future development of the sector. Creating updated knowledge on the current situation and trend back to the past on milk production is the utmost importance in making the reliable forecasting of the future growth and dynamics of the dairy sector development. In order to predict the future, the past and the present have to be carefully analyzed since the future is the further development of the present (Roland et al., 2016).
Forecasting for the future (e.g. making outlook) is important to make a strategic plan by any government in one hand, is complex and need a strong database and statistical knowledge on the other hand. Since the data is scarce in Bangladesh, utilization of alternative data source is justified to show how the database is supporting the country’s dairy sector development. As Bangladesh is currently deficient in liquid milk production and a net importing country for the powder milk, the government has been implementing a megaproject called Livestock and Dairy Development Project (LDDP) in cooperation with World Bank. Among several objectives of this project, one of the major objectives is to increase milk production to a level that leads to becoming selfsufficient in the near future. Given such background and data scarcity in mind, this study aimed at forecasting the milk production based on the two data sources in order to estimate the possible timeframe for achieving selfsufficiency in Bangladesh.
The nature of this study inquires a multiple forecasting methodology as forecasting of milk production is a complex procedure and forecasting works under several conditional assumptions. Even with such limitation, the straightforward forecasting of milk production using different methodology would provide an output which can be compared with the previous period (that has already occurred) and simultaneously can be extended for prediction for the desired year. The predicted forecasting will provide a proxy to the real situation but still is an important tool for policy application and strategic decision process for enhancing dairy development. This study has applied multiple forecasting tools for making a valid conclusion on the forecasted milk production along with the number of years required to be selfsufficient in Bangladesh.
For modelling any forecasting model, sources of data, data consistency and longterm timeseries data are extremely important. In this study, we use two sources of data sets: i) milk production and dairy cattle population reported by Department of Livestock Services (DLS) as annual data; ii) milk production database developed by Integrated Dairy Research Network (IDRN) with the collaboration of International Farm Comparison Network (IFCN), Germany (Hemme et al 2014 and www.idrndairy.org). The DLS data is available from 2006 to 2018 and for IDRN data, it is from 1996 to 2019 (also available monthly data from 2006). For making standardization dataset, we take the data from 2006 to 2019 (estimated for DLS data for 2019).
There are several methods available for forecasting (Gooijer and Hyndman 2006) which might vary from the sector to sector and local need and context. In our study, four different forecasting tools were applied: Autoregressive Integrated Moving Average (ARIMA); Exponential Growth Rate; Compounded Average Growth Rate (CAGR), and Simple Average Growth Rate (SAGR).
ARIMA model is also known as BoxJenkins method (Box and Jenkins 1970) who developed a coherent, versatile threestage iterative cycle for time series identification, estimation, and verification which is also known as the BoxJenkins approach. This method has an enormous impact on the theory and practice of modern time series analysis and forecasting (Gooijer and Hyndman 2006). This helps to analyze both probabilistic and stochastic properties of single time series data. Unlike regression model, ARIMA model allows being explained by the movement of a variable by its past or lagged values. It produces predictions based on the synthesis of historical data. ARIMA can be run for single and multiple variables, however, for this study, we run this as a single variable which is called “Univariate ARIMA”. ARIMA model is most widely used elsewhere, however, this would be the first time (so far author’s knowledge goes) for the case of milk production forecasting in Bangladesh. The exception is the study done by Hossain and Hassan 2013 who forecasted milk production using multiple regression techniques (linear, compound, exponential, quadratic, and cubic). The ARIMA model consists of:
ARIMA (p,d,q) takes the times series data where p denotes the number of autoregression need to be conducted (AR), d is the number of differentiation need to be done for the time series data in case the data is not found to be stationary (I) and q the number of moving average terms (q). The detailed mathematical notion of ARIMA model as explained by Deshmukh and Paramasivam 2016 is stated below:
AutoRegressive Process of order (p) is,
Yt = μ + Ø_{1} Y_{t1} + Ø_{2} Y_{t2} + ….. + Ø_{p}Y_{tp} + ε_{t} ;
And the general form of ARIMA model of order (p, d, q) is
Yt = μ + Ø_{1} Y_{t1} + Ø_{2} Y_{t2} + ….. + Ø_{p}Y_{tp} + ε_{t} + μ + θ_{1} ε_{t1} + θ_{2} ε_{t2} + …… + θ_{q} ε_{tq} + ε_{t};
Where,
Y_{t} is milk production, ε_{t} are independently and normally distributed with zero mean and constant variance for t = 1, 2, …., n; and Ø_{p} and θ_{q} are also estimated
The data were tested using graphical representation and Augmented DickeyFuller test for both IDRNBAU and DLS data. The graphical representation shows that milk production data in the time series is upward trending with substantial fluctuation. This implies that time series of milk production data cannot have the constant mean and variance leading to the decision on the lack of stationary data. To confirm this Augmented DickeyFuller test was performed. For timeseries data to be stationary, the Z(t) should have a large negative number, the pvalue should be significant at least on 5% level. In our data set (both IDRNBAU and DLS) that were tested, neither of these conditions were met. Therefore, the null hypothesis i.e. timeseries data of milk production is nonstationary, cannot be rejected. And since the time series milk production is nonstationary, further analysis cannot be performed on. A solution on this to be still used by ARIMA model is to estimate the first differencing of the time series and the newly created variable for milk production needs to be again tested by graphical and Augmented DickeyFuller test. This was done for both data sets and after doing the 1^{st} Order differences, the null hypothesis was rejected, indicating that both data fulfil the stationary property which can then proceed for running ARIMA model.
This method originated with the work of Brown (1959) which is, thereafter, used in statistical framework by Box and Jenkins (1970) and Abraham and Ledolter (1986)
This is used for a specific case where the growth occurs instantaneous rate of change (that is derivative) of quantity concerning time is proportional to the quantity itself. In this case, the time is considered as an exponent (considering the constant for other types of growth, such as quadratic). In contrast, if the quantity decreases over time, this is called exponential decay. In our study, we consider only the exponential growth which is estimated as below:
Y=b x m^{x}
Where,
X is the number of years in the calculation for predicted milk production (independent variable),
Y is the final milk production value (dependent variable),
m is the constant base for the x value,
b is a constant which is the value of Y when x = 0.
For the case of the multiple of ranges of x values, the estimation becomes different which is not used in this study as our datasets did not permit to do that.
This estimate compounding growth over multiple years. The limitation of the SAGR where the growth rate is assumed to be constant over two subsequent years, which is not in reality, CAGR is used to solve this problem. CAGR takes into account geometric progression ratio that provides a constant rate of growth over the period. However, this does not include the effect of the volatility of periodic milk production. This is, however, useful to compare the growth rates from various sets of data, which is the case for this study. The CAGR is calculated as below following the model of Anson et al (2010):
End value = milk production in the last year to be estimated,
Start value = Milk production in the base year to be estimated,
No. of year count = Total number of years from the base year to the last estimated year.
Simple Average Growth Rate (SAGR): this shows simply the growth from the previous years and uses the equal growth rate for the forecasted years (Hayes 2019). Using the procedure, changes over time in the base from which the growth is estimated include the increase in milk production from yeartoyear growth. This is simple and easy to estimate using excel and less time consuming, especially where there is limited data, this is the valid procedure to make the outlook for production. The mathematical notion for this
End value = Milk production in the last year to be estimated,
Start value = Milk production in the base year to be estimated,
No. of year count = Total number of years form the base year to the last estimated year.
The data were forecasted from 2020 to 2030 although there might be a caution to forecast for so long time by ARIMA. But in this study, this was done purposively in order to estimate the required time to be selfsufficient.
All of the models (ARIMA and different types of Trend Analysis model) were applied for both IDRNBAU datasets and DLS data set in order to predict which datasets have consistency property. The data were analyzed using the STATA software version 12.0 and Microsoft Excel 2016.
Milk production is the main output variable in dairy farms which drives the profitability for a particular dairy farm. To meet the demand of the growing population, the Bangladesh government has been investing significant resources to increase milk production. In order to boost future production, understanding the behaviour of historic milk production is necessary. To keep this in purview, the milk production (million ton) and its development (% growth change) from 2006 to 2019 have been depicted in Figure 1a and 1b, respectively.
Figure 1a. Milk production (mton/year)  Figure 1b. Milk production growth (%) 
The milk production in Bangladesh as revealed from both datasets has been increasing at a rate faster during the period from 2006 to 2019 but with a strong fluctuation in milk production. However, except at the beginning of the development (e.g. 2006 and 2007) when it was found almost a similar amount of milk production. The milk production in 2006 and 2007, was close to each other and slight variation was observed. Interestingly to note that milk production has been reported by DLS was lower than IDRNBAU until 2011. From 2012 onward, the milk production data reported by DLS was significantly higher than IDRNBAU. The reasons were not explained and there is not found a methodological explanation why such an increase was reported. A difference of milk production at the year 2006 was found that DLS has lower milk production (0.23 million ton) which was turned into a difference of the DLS from IDRNBAU was 2.33 million ton in 2019.
Looking into the annual growth rate change (expressed as %), the average growth rate for IDRNBAU and DLS is found to be 8.8 and 11.5%, respectively (Figure 1b). The development of milk production for the DLS data fluctuated with the highest growth rate of 47% in 2012 while the lowest was 14% in 2008. In contrast, the IDRNBAU milk production growth was the highest in 2008 which was 20% and the lowest was found as 5% in 2012. The contradictory results for 2008 and 2012 were due to the fact that IDRNBAU data incorporated the natural calamity and other shocks that was happened over the time which might not include in the DLS data or there could be any other reason which is not published by DLS. The IDRNBAU data sets considered the Tsunami (Locally called “Sidor”) in the southwest region in 2007, the world economic recession in 2009 and the melamine crisis in 2008, the prevalence of Anthrax in the main dairy region in 2011. The data published by DLS lacks this explanation of how do they address those shocks, especially milk production in 2012 has increased by 47%.
Since Bangladesh dairy sector is progressing faster, the accurate estimation of milk production and related other factors that drives the milk production variation like to play a vital role in helping to make the right intervention. Due to the shortage of the database, the lack of appropriate methodological tools for the DLS data, and the lack of institutional endorsement of the IDRN (since IDRNData is still considered as researchoriented database), it is highly relevant for the DLS (as a regulatory body of the government) to rethink on the possible validation process to revise the historic data and to take a holistic approach to estimate the future outlook data for milk production. Milk production is the key aspect of understanding the country’s dairy sector development. The DLS and IDRNBAU can work together while the DLS has institutional mandate along with countrywide network while the IDRNBAU is strong on methodology, modelling, data generation and data validation expertise tools. Synchronization of the working of both might solve the inconsistency in the dataset in the near future.
Based on the historic development of milk production (2006 – 2019) which was found to be inconsistent, the forecasting of milk production was done for both datasets in order to compare the milk production growth potential in the coming year along with the selfsufficiency achievement for Bangladesh.
The data from two sources (IDRN and DLS) were checked for its stationarity (e.g. having equal mean and variance over the year). The data were diagnosed as nonstationary as explained in the methodology section. The problem of nonstationarity was solved by performing 1st differencing order of the historic data which was then plotted with the help of AC and PAC graphical representation and depicted in Figure 2a and 2b.
Figure 2a. Graphical representation of correlogram (ac)  Figure 2b. Graphical representation of partial correlogram (pac) 
Figure 2a and 2b shows for which array the ARIMA should take into account and this was helpful to construct and select the ARIMA model which is depicted in Table 1 where it is found that ARIMA model should run from 1 to 12.
Table 1. ARIMA model array 

SL No. 
AR 
I 
MA 
ARIMA 
1 
1 
1 
1 
(1,1,1) 
2 
1 
1 
2 
(1,1,2) 
3 
1 
1 
3 
(1,1,3) 
4 
1 
1 
4 
(1,1,4) 
5 
1 
1 
5 
(1,1,5) 
6 
1 
1 
6 
(1,1,6) 
7 
1 
1 
7 
(1,1,7) 
8 
1 
1 
8 
(1,1,8) 
9 
1 
1 
9 
(1,1,9) 
10 
1 
1 
10 
(1,1,10) 
11 
1 
1 
11 
(1,1,11) 
12 
1 
1 
12 
(1,1,12) 
Table 2 shows the values of LogLikelihood, Pvalue and AIC and BIC values which were used for the selection of the best fitting ARIMA model to run the forecasting. Taking the values from table 2, ARIMA (1,1,1) was selected for forecasting the milk production both for DLS and IDRN data.
The ARIMA model results for forecasting from 2020 to 2030 and Trend analysis results are depicted in Table 3. The table shows that ARIMA model and CAGR results seem plausible while other two Exponential and SAGR is ended up with extremely high values for the DLS and even it is significantly lower compared to the DLS but substantially higher in the context of the real situation of Bangladesh dairy sector.
Table 2. Selection of the model based on LogLikelihood, AIC, BIC and Pvalue 

ARIMA 
Log 
Coefficient 
AIC 
BIC 
P 
Level of 
(1,1,1) 
871.849 
0.84 
1749.70 
1759.09 
0.000 
** 
(1,1,2) 
861.40 
0.75 
1732.81 
1748.46 
0.000 
** 
(1,1,3) 
857.89 
0.69 
1727.79 
1746.57 
0.000 
** 
(1,1,4) 
853.70 
0.58 
1721.39 
1743.30 
0.000 
** 
(1,1,5) 
857.41 
0.86 
1730.82 
1755.86 
0.000 
** 
(1,1,6) 
854.98 
1.01 
1727.96 
1756.13 
0.000 
** 
(1,1,7) 
847.60 
0.08 
1715.19 
1746.49 
0.82 
NS 
(1,1,8) 
842.55 
0.65 
1707.10 
1741.53 
0.000 
** 
(1,1,9) 
840.46 
0.44 
1704.92 
1742.48 
0.058 
* 
(1,1,10) 
838.82 
0.68 
1703.63 
1744.32 
0.002 
* 
(1,1,11) 
838.81 
0.68 
1705.62 
1749.44 
0.047 
* 
(1,1,12) 
836.29 
0.13 
1702.57 
1749.52 
0.726 
NS 
The ARIMA model shows that the maximum milk production in 2030 can be achieved is 13.6 million ton (as per the IDRNBAU data) and 18.1 million ton (as per the DLS data) which is significantly different from each other.
Table 3. Comparison of forecasted milk production between IDRN and DLS data 

Milk 
Forecasting Method 
2020 
2021 
2022 
2023 
2024 
2025 
2026 
2027 
2028 
2029 
2030 

IDRN 
ARIMA 
8.7 
8.4 
8.9 
9.3 
9.8 
10.2 
10.7 
10.2 
11.6 
12.1 
13.6 

Trend 
Exponential 
9.6 
10.6 
11.7 
12.9 
14.3 
15.7 
17.4 
19.2 
21.2 
23.4 
25.8 

CAGR 
8.9 
9.6 
10.5 
11.4 
12.4 
13.5 
14.7 
15.9 
17.3 
18.9 
20.5 

SAGR 
9.4 
11.0 
12.7 
14.7 
17.1 
19.9 
23.0 
26.7 
31.0 
36.0 
41.7 

DLS 
ARIMA 
11.2 
11.9 
12.6 
13.2 
13.9 
14.6 
15.3 
16.0 
16.7 
17.4 
18.1 

Trend 
Exponential 
14.0 
16.1 
18.5 
21.2 
24.4 
28.0 
32.2 
37.1 
42.6 
49.0 
56.3 

CAGR 
11.7 
13.0 
14.5 
16.2 
18.1 
20.1 
22.4 
25.0 
27.9 
31.1 
34.7 

SAGR 
13.2 
16.5 
20.8 
26.1 
32.8 
41.3 
51.8 
65.2 
81.9 
102.9 
129.3 

Also, it was found significant differences in milk production using CAGR. Among the datasets, DLS data has a very high variation (too extreme in upper and lower bound) which influences the forecasting results. Due to such high variation, it is again argued that it might difficult to make realistic inferences based on the data forecasting. However, the advantages of this forecasting are that it identified the inconsistency as well as it guides how shall the future data collection be.
The demand for milk in Bangladesh is calculated by DLS using the deterministic approach, e.g. taking the World Health Organization (WHO) recommended average daily intake per capita of 250 g per day over the year. The demand is also changed with the changes of per capita income, rate of urbanization and changes in the consumers' preference which were not taken into account in this demand calculation. The forecasted milk production from table 3 is used to estimate the selfsufficiency until 2030 which are shown in Figure 3.
Figure 3. Forecasted milk demand and selfsufficiency (2020 – 2030) 
Figure 3 shows that milk demand will increase to 17.22 million ton in 2030 which could be achieved according to DLS data in 2029 while the IDRNBAU data shows 80% selfsufficiency in 2030. The diverse of the selfsufficiency forecasting results might be because both databases have a different dimension of development path with different peak and lean growth over the year which has impacted the forecasting results. The rate of increase in selfsufficiency is, somewhat, similar pace with both databases. The IDRNBAU data shows an increase of 22% (initial 56% in 2020 and 80% in 2030) while that for DLS dataset shows 34% (initial 72% in 2020 and 106% in 2030). However, this does not take into account the other dynamic drivers in the demand calculation as well as the demand of the WHO might be changed in future. This enquires that the future selfsufficiency will, highly, depend on the strategic decision, how the government would define the future nutritional security and dietary recommendation for the overall population.
The data of dairy sector is relatively scarce in Bangladesh compared with other agricultural commodities even substantial efforts being made by the Department of Livestock Services (DLS) under the Ministry of Fisheries and Livestock (MoFL). Concerning dairy farms and farming system, the DLS update three key data on milk production, total cattle and total buffalo number while it estimates the number of variables such as demand of milk, milk availability and milk surplus/deficiency.
At the same pace, the Integrated Dairy Research Network (IDRN) which is a proposed network under the Department of Animal Nutrition of Bangladesh Agricultural University has been constantly working to update the latest data by utilizing the methods and models developed by the International Farm Comparison Network (IFCN), Germany (Hemme 2000) which has been further refined and extended to address the dynamics of the global changes in a dairy farm and dairy sector. The IDRN has Dairy Sector Database consisting of 29 variables and dairy farm database consisting of 647 variables.
The milk production data reported by IDRN and DLS has been diagnosed in order to establish a logical backcalculation procedure. This calculation will help the validation of the data on which the decision can be made more precisely.
Table 4. Herd structure data used in the milk production calculation 

Parameter 
Herd 
Source 
No. of cattle 
24.5 
DLS, 2019 
No. of buffalo 
1.49 

No. of Male cattle 
12.5 
IDRN farm survey of 1823 in 2019 
No. female cattle 
12.0 

No. of lactating cow 
3.7 

No. of dry cow 
3.2 

No. of heifer 
1.9 

No. of female calf 
3.1 

No. of buffalo 
1.49 

No. of lactating buffalo 
0.30 

Herd structure data is estimated based on the coefficients (different proportion of the cattle found in the farm level using the IDRN farm survey data) 
Herd structure data is estimated based on the coefficients (different proportion of the cattle found in the farm level using the IDRN farm survey data).
The herd structure used in this calculation is depicted in Table 4. Using this data, the milk production from both IDRN and DLS data is diagnosed. The detailed diagnosis of the milk production is depicted in Table 5.
Table 5. Diagnosis of the milk production data reported by IDRN and DLS, 2019 

Source 
IDRN 
DLS 
Total milk production (m.ton) 
8.14 
10.47* 
Milk production: cow/day in kg 
6.0 
7.48 
To produce the total milk, dairy farm density: No/km^{2 **} 
279 
359 
* estimated from DLS 2018 data 
Table 5 clearly shows that milk production data in both sources has been overestimated as the real dairy farming situation might not reflect. As the country dairy farming has milk production mostly from local cows representing 67% of the total cows with an average production of only 2.80 while for remaining is cross bred 33% and produce 6.39 kg/day (Huque 2014). However, over the time and entrepreneur initiative, the ratio might have been changed which is not yet explored. In case of IDRN data, the coefficient that was taken using the farm survey data but those data tended to be biased toward mostly from the major dairy region in the country which might be different if the data is generated from all 64 districts. On the other hand, the DLS data is estimated institutional office record and less used the statistical tool and survey tools which also might tending toward upper bound bias. If we estimate the dairy farm density to produce the total milk, it is found that for producing the DLS reported milk production, the density is 359 farms/km^{2 } and for that of IDRN milk production, it is 279 dairy farm/km^{2}. This density data seems to be overestimated as well and is divergent from reality. However, the mega project of Livestock and Dairy Development Project (LDDP) could lead to the refining of the current milk production data and also need to include more variables to make the estimate that reflects the reality.
Similar to milk production data, the estimation of demand also needs to be analyzed further. The DLS estimates the demand for utilizing the deterministic approach. Under the deterministic approach, the constant amount of milk demand is used in the calculation, while for the stochastic approach, the demand is taken as a function of stochastic variables such as an increase in income, population and urbanization and change in consumer preferences.
Table 6. Comparison of Selfsufficiency between the deterministic and stochastic approach 

Unit 
2025 
2030 

Forecasted Milk DemandDeterministic 
Million ton 
16.35 
17.22 
Forecasted Milk DemandStochastic 
Million ton 
21.25 
27.55 
IDRN forecasted milk production 
Million ton 
10.24 
13.65 
DLSforecasted milk production 
Million ton 
14.63 
18.10 
IDRNForecasted SelfSufficiency Deterministic 
% 
63 
79 
DLSForecasted SelfSufficiency Deterministic 
% 
89 
105 
IDRNForecasted SelfSufficiency–Stochastic 
% 
48 
50 
DLSForecasted SelfSufficiency Stochastic 
% 
69 
66 
Deterministic approach: use of constant demand of 250 g/day/capita while Stochastic approach: use of the varying amount of demand, for 2025, 325 g and 2030, 400 g/day/capita 
If we take the stochastic nature of the demand, the demand might change over time. Taking both deterministic and stochastic approach in demand calculation, the selfsufficiency estimation has been depicted in Table 6. The forecasted milk production and demand is taken from ARIMA model. Bangladesh will achieve selfsufficiency in next 10 years if the deterministic approach is applied but as the country has been undergoing tremendous economic progress and becoming a middleincome country in 2024 (World Bank, 2019), the demand might increase which will lead to supporting the stochastic approach. However, the data used for selfsufficiency by IDRN and DLS has to be taken into account as both data shows two different periods to achieve selfsufficiency. At the similar pace, the impact of Coronavirus (COVID19) which has occurred only in the year 2020, could not be taken into account during forecasting modelling due to anecdotal in data which could lead the future decrease in milk production.
As policy implication perspectives, the analysis used in this study is not enough to judge the superiority of the data one over other, however, it could be speculated that both databases are overestimated. The government might focus on the revision of the historic development of milk production data for developing DLS database. At the same pace, IDRN as a researchbased database network might work together with DLS to support the methods and models to estimate the milk production which is closer to reality. However, it is also noteworthy to mention that dairy data is very much inconsistent and proper care must be taken during any research or database update. This implies that any data related finding might be tested and validated with the national panel of experts before publishing where the Livestock and Dairy Development Project (LDDP) can take initiatives. The precise estimation of the milk production data is linked with selfsufficiency defining the country’s position as surplus or deficit which might be addressed by the national dairy development policy in Bangladesh.
The authors sincerely acknowledge the Department of Livestock Services (DLS), the Department of Animal Nutrition, all of the expert members, database members and Team members of the IDRNBAU research network. The authors also acknowledge the International Farm Comparison Network (IFCN), Germany for providing the models and database access. The Authors also acknowledge the Ministry of Education (BANBEIS) project (Project reference 490/2018) for financial support. The authors also sincerely acknowledge Dr. Khan Shahidul Huque, former Director General of Bangladesh Livestock Research Institute (BLRI) for his excellent feedback and comments.
Abraham B and Ledolter J 1986 Forecast functions implied by autoregressive integrated moving average models and other related forecast procedures, International Statistical Review, 54 (1), 5166. https://www.jstor.org/stable/1403258?seq=1
Anson M J P, Fabozzi F J and Jones F J 2010 The Handbook of Traditional and Alternative Investment Vehicles: Investment Characteristics and Strategies, 489. https://www.wiley.com/enge/The+Handbook+of+Traditional+and+Alternative+Investment+Vehicles:+Investment+Characteristics+and+Strategiesp9780470609736
Box G E P and Jenkins G M 1970 Time series analysis: Forecasting and Control. San Francisco: Holden Day (revised ed. 1976).
Brown R G 1959 Statistical forecasting for inventory control, New York, McGrawHill. https://www.worldcat.org/title/statisticalforecastingforinventorycontrol/oclc/573848798
Deshmuk S S and Paramasivam R 2016 Forecasting of milk production in India with ARIMA and VAR time series models. https://arccjournals.com/uploads/articles/3DR954.pdf
DLS 2019 Livestock Economy at a glance, Livestock Economic Division, Department of Livestock Services, Dhaka. http://www.dls.gov.bd/site/page/22b1143b932344f8bfd8647087828c9b/LivestockEconomy
Douphrate D I, Hagevoort G R, Nonnenmann M W, LunnerKolstrup C, Reynolds S J, Jakob M and Kinsel M 2013 The dairy industry: a brief description of production practices, trends, and farm characteristics around the world. Journal of agromedicine, 18(3), 187197. https://www.ncbi.nlm.nih.gov/pubmed/23844787
Gooijer J G and Hyndman R J 2006 25 years of time series of forecasting. International Journal of Forecasting, 22: 443473. https://www.sciencedirect.com/science/article/abs/pii/S0169207006000021
Hayes A 2019 Average Annual Growth Rate. https://www.investopedia.com/terms/a/aagr.asp
Hemme T 2000 Ein Konzept zur international vergleichnden Analyse von Politikund Technikfolgen in der Landwirtschaft. LandbauforshungVölkernode, Sonderheft 215 (2000).
Hemme T, Uddin M M and Ndambi O A 2014 Benchmarking cost of milk production in 46 countries, Journal of Reviews on Global Economics, 3: 254270. https://www.lifescienceglobal.com/media/zj_fileseller/files/JRGEV3A20Ndambi.pdf
Hossain M J and Hassan M F 2013 Forecasting of Milk, Meat and Egg Production in Bangladesh, Research Journal of Animal, Veterinary and Fisheries Sciences, 1 (9), October 2013. https://www.researchgate.net/publication/303814705_Forecasting_of_Milk_Meat_and_Egg_Production_in_Bangladesh
Huque K S 2014 A performance profile of dairying in Bangladesh program, policies and way forward, Bangladesh Journal of Animal Science, 43 (2), 81103. https://www.banglajol.info/index.php/BJAS/article/view/20662
IDRN 2020 Integrated Dairy Research Network – Monthly dairy sector update, Bangladesh Agricultural university, Bangladesh. Available at: www.idrndairy.org
IFCN 2019 Dairy Report for better understanding of milk production worldwide. IFCNthe Dairy Research Network, University of Kiel, Germany. www.ifcndairy.org
Perdijk O, van Splunter M, Savelkoul H F J, Brugman S, and van Neerven R J 2018 Cow’s milk and immune function in the respiratory tract: potential mechanism. Frontier in Immunology, 9, 143. https://www.frontiersin.org/articles/10.3389/fimmu.2018.00143/full
Roland S, Berix V and Reneta GP 2016 Possible methods for price forecasting. MultiScienceXX. MicroCAD International Multidisciplinary Scientific Conference, University of Miskolc, Hungary, 2122 April, 2016. https://www.unimiskolc.hu/~microcad/publikaciok/2016/E_feliratozva/E_3_Csorgits_Lajos.pdf
Uddin M M, Khaleduzzaman A B M and Akter A 2020 Sustainable dairying and selfsufficiency in quality milk production in Bangladesh: the role of policy and a way forward. Bangladesh Journal of Academy of Science (Submitted).
World Bank 2019 The world Bank in Bangladesh. https://www.worldbank.org/en/country/bangladesh/overview .
Received 3 April 2020; Accepted 7 April 2020; Published 1 May 2020