Both primary and secondary data were used for the study. Secondary data were collected from several issues of Bangladesh Bureau of Statistics (BBS) and Department of Agricultural Marketing (DAM). For the collection of primary data, three districts namely Bogra, Serajgong and Gaibandha were selected purposively depending upon the concentration of production and commercially marketing of winter chilli. The study was considered for the time period of 1985 to 2014 for time series analysis. Purposive and simple random sampling techniques were used to pick a primary sample. A total of 92 farmers and traders were selected randomly from the three areas. Primary data were collected by face-to-face interview through pretested structured interview schedules. The collected data and information were reduced to tabular form which included classification of tables into meaningful results. Except this the following analytical techniques were used for the study. Yt = α + β0Xt + β1Xt-1 + β2Xt-2 +…+ βkXt-k + ut .......................(2) Where, Yt is chilli production in period t, Xt is chilli price in period t and Xt-1 is chilli production in one period earlier than t. In a distributed lag model, Schwarz information criterion (SIC) was used to determine the lag length Dikmen, 2005). Schwarz proposes reduction to the lowest: SIC = ln σ2 + m ln n...................(3) Here, σ2 is the highest probability estimate of σ2 = (RSS/n), m is length of the lag, n is the number of observations and RSS is the residual sum of square. In summary, a regression model is used along with some lag values (=m), and m value that makes the value of Schwarz criterion the lowest is selected (Gujarati, 2005). At this stage, without making any limitation to the form of the distributed lag, a very large k value length of the lag is used at the start. Then, when the duration of lag is shortened, whether the model goes wrong is checked (Davidson and Mackinnon, 1993). Yt = α (1- λ) + β0Xt + λYt-1 + vt .........................(4) Theoretical framework: Distributed lag models has a special place in literature of economics in that they can allow the analyzing the behaviors of economical units (consumer, producers, etc.) based on appropriate dynamic models. Studied and used for the first time by Irving Fisher (Isyar, 1999), distributed lag models takes into account not only the present year value but also the previous year values of defining variable. If how far back will be gone for defining variable is not described, this is called an “infinite lag model” and shown as follows: Yt = α + β0 Xt + β1 Xt-1 + β 2 Xt-2 + …+ ut ............(5) On the other hand, if the number of years to go back is defined as k for defining variable, it is called “finite distributed lag model” and has been defined as: Yt = α + β0Xt + β1Xt-1 + β2Xt-2 +…+ βkXt-k + ut .................(6) In this model, dependent variable Y (Yt … Yt-k) is not only influenced by the present day value (Xt) but also by the past values (Xt-1 ……. Xt-k) of defining variable. Most often, Y responds to X after some time, and the time to respond is called “lag period” (Dikmen, 2005). Unknown parameters in distributed lag models (α, α0, …, α k) can be estimated using the classical least squares method (Alt, 1942; Tinbergen, 1949). Model-specific estimates in distributed lag models have certain drawbacks (Gujarati, 2005). One of them is the lack of a pre-information in the model about how long the lag period will be. Another is that when a data set that can estimate the lag period is not set up, degree of freedom is increasingly decreased in statistical significance tests of parameters. Yet another, but the most significant, drawback is that variables decided as defining variables are in a multiple linear relationships. In order to overcome the above mentioned drawbacks, a method was developed by, and named after Koyck (1954). Leendert Marinus Koyck (1918- 1962) was a Dutch economist who studied and worked at the Netherlands School of Economics, which is now called the Erasmus University Rotterdam (Franses and Oest, 2004). Based on the assumption that lags in independent variable affect the dependent variable to some extent and the weight of these lags decrease geometrically, model is reduced and thus made to estimate the regression equation (Dikmen, 2005). In order to obtain the reduced model, Koyck assumed that in an infinitely distributed lag model all β’s had the same signs and geometrically decrease as shown below: βk = β0 λk k = 0,1,...... (7) In the Koyck model, a sensible range for λ would be the interval 0;1). One possibility, involving the entire distribution over λ would be to consider the class of “sup test statistics”, which corresponds to the highest value of the original test statistic within the range for α. This approach is advocated by Davies (1987); see also Hansen (1996) and Carrasco (2002). βk is the lag coefficient. Lag coefficient βk varies by λ as well as by β0. The closer λ to 1, the less the decrease in βk. On the other hand, the closer λ to zero, the greater the decrease in βk (Gujarati, 2005). In other words, λ values close to 1 mean that values of defining variables in remote past have a significant effect on dependent variable, and λ values close to zero mean that values of the defining variable in the remote past rapidly lose their effects of dependent variable. Mean lag number is the weighted average of all lags and is calculated for Koyck model as shown in Equation (4). Mean lag number shows the time period necessary for a one unit change in X defining variable to have a detectable effect on dependent variable Y (Dikmen, 2005). In view of these explanations, infinite lag model is formed using OLS method as shown in Equation (8). Yt = α + β0Xt + β0λXt-1 + β0λ2Xt-2 + ….+ ut ............(9) Linear regression solution cannot be applied to regression Equation (8) since it has infinite lag and λ coefficients are not linear. In order to solve this problem, the model has been taken one period back and the following regression model has been developed: Yt-1 = α + β0Xt-1 + β0λXt-2 + β0λ2Xt-3 + ….+ ut-1 ....................(10) When the equation (9) is multiplied by λ, the Equation (10) is obtained; λYt-1 = λa + λβ0Xt-1 + λ2β0Xt-2 + λ3β0Xt-3 +...+ λut-1.................... (11) When the Equation (10), whose lag is taken one period back, is subtracted from Equation (8), whose lag is infinite, the following Equation is reached: Yt - λYt-1 = α (1- λ ) + β0Xt + (ut - ut-1 ) ..............(12) If the Equation (12) is reorganized, Equation (13) is obtained; Yt = α (1- λ) + β0Xt + λYt-1 + vt .................(13) In Equation (13) vt = ( ut - λut-1 ) and, it is the moving average mean of ut and ut-1. The procedure explained above is known as Koyck transformation. Equation (13) is described as Koyck model. In Koyck model, variables that consist of lag values of defining variables are not defined. Thus, multiple relationship problems are in a sense solved. On the other hand, while in infinitely distributed lag model it is necessary to predict infinite number of β using α, in Koyck model distributed lag model can be resolved only through estimating α, β0 and λ.