About Us

1. Electric load forecasting using artificial neural network and deficit management

The neural network was constructed with one input layer of five nodes, one hidden layer, and one output layer with one node. The inputs were selected on the basis of availability of data in the historical database. The forecast was 24-hour-ahead forecast with inputs of prior days, including total load, temperature, and forecast of temperature. The single output forecast was total daily load.

Deficit management module was modeled for excess load and energy shortage scenarios.

A total of 168 ANNs was used, one for each day type and hour of the day. Following is a table giving actual load vs. predicted load:

2. Individual Credit Rating

Developed an artificial neural network (ANN) model for a finance company providing loans to individuals for consumer goods. Based on the completed records of a large number of individuals, the ANN model provided an excellent fit to the data and also had a very high (86%) success rate in making "out of sample" predictions

3. Commodity Markets

For textile mills, cotton accounts for a very large percentage of cost of production. For one cotton variety, we used historical data to make forecasts. Using statistical and ANN models, the short term predictions were within 1 to 2 percent of the actual prices. The long term trends were good only with the ANN model. The model provided optimal strategy for purchase of cotton for the textile mill.

4. Macroeconomic Analysis

In order to study inflation in India, a macroeconomic multivariate model was developed using ANN methodology. The data consisted of twenty-one years of average monthly values of five significant variables. These five variables were: real effective exchange rate (REER), money supply (M3), index of industrial production (IIP), food arrivals (F), and the wholesale price index (WPI). It was found that the usual statistical approach was reasonable only when making a univariate time series study of M3 and WPI. All other variables had nonlinear features. The multivariate ANN model was the best in making very accurate (mean error 1-2%) short-term forecasts and long-term trends (mean error up to 4%). The multivariate model also helped generate alternative scenarios. For example, it was possible to forecast the change in inflation rate and industrial activity by giving impulses to REER and M3.

The crucial policy implication is that within the objective of price stability there exists a case for intervention by the Central Bank up to a limit set by equilibrium rate of real exchange rate.

5. Marketing Models

Carried out a variety of marketing studies for a major multinational company in India, manufacturing consumer products of daily use.
  • Completed an assignment to find out the correlation between the buying behavior and the demographic profile of a large number of households in metro cities in India. The objective was to generate quantitative information to devise optimal sales and promotion strategies. This is being implemented.

  • A model has been constructed for long term planning. The basic inputs were the portfolio of products manufactured by the company and a given fixed target of overall growth. Within the model an optimal workable strategy was found. It is being examined currently.