Dr N Balakumar

Once a researcher finalises his/her project topic and sample size and completes data collection, the next prime issue faced by him/her is the selection of apt research methodology.

If you do not choose the most appropriate methodology, the all your efforts will become waste, since your statistical results depend on the methodology selected. Further, your findings and interpretations are a function of results. Therefore, choosing the correct methodology is very crucial for any research work, especially that are empirical in nature.

Let us view this issue with examples.

Assume that a student is working on a forecasting related work, say, forecasting in an ago-based industry or forecasting of share prices.

An immediate methodology selection by most of the researchers (and academicians) will be to use the traditional regression models.

Is this the right choice? Continue reading to explore more.

The range of corporate forecasting exercises is wide, reflecting the many aspects of business forecasting and decision making that can be studied quantitatively.

Forecasting in an agro-based industry depends on multiple factors, since agriculture itself depends on several factors. At the industry or firm level a principal influence flows from the business cycle situation of the macro economy, which dominates the variable that is being estimated over the short and medium term.

Key raw materials for an agro-based industry are derived from agricultural commodities. Thus, the availability and the price of raw materials is a function of several dependent factors, such as, monsoon, geographical advantages/disadvantages, cost of transportation, government policies on pricing of agricultural commodities and subsidies on power & fertiliser, to name a few.

Therefore, agricultural forecasting is considered as the frontier of business forecasting – they represent its greatest challenge and potential.

The methodology to arrive at the appropriate statistical forecasting model is highly correlated to the nature of data or the independent variable that determine the dependent variable.

Several forecasting models, like, Moving Average, Auto-regressive, ARIMA, Least-square Method, Link-relative Method, Stochastic, etc. are followed by the practitioners of agricultural forecasting.

The selection of methodology or the forecasting model is based on the nature of independent variables. The values of independent variables could be linear, non-linear (exponential, polynomial, logarithmic, geometric, etc.) The forecasting models selected would vary according to the behavioural pattern of the variables. Thus, prior to estimating the required dependent variable, identification of the restrictions of the independent variables is a must to arrive at the right forecasting model. Therefore, the selection of the appropriate prediction model could be arrived only after a careful study of the variables that will affect the dependent variable.

The business modeling differs from traditional econometrics. In actual practice, the estimated values are used for a firm’s business decision making, which urges the need for a forecasting model that is close to reality.

Therefore, to proceed further in the direction of making an attempt to select the most appropriate forecasting methodology, it becomes imperative to understand the nature of independent variables.

Similar problem will be faced by a candidate who tries to use the regression models for share price forecasting.

Thus, statistically, even if you have a statistically significant regression model, it may be difficult or impossible to explain the movement of dependent variable (share price, for example) through the use of structural model. The reasons are:

(a) Data are not available of all those explanatory variables which affect share prices (dependent variable); and

(b) To get the forecasting value for the dependent variable from a regression equation, those explanatory variables that are not lagged must themselves be forecasted. This is even more difficult to forecast than estimating the dependent variable itself.

Thus, there could be seen a strong case to seek for other means of obtaining a forecast for the dependent variable. Such an alternative approach is modern time series analysis.

It may not be that very easy to determine methodology by a fresher of research. It requires academic experience and more than that research expertise. Thus, you must consult your faculty and your connections with reality on methodology selection that will lead to a superior project report.