Experienced R Programmer to Create Seasonal Forecast Script
Budget: $250 – $750 USD
I am looking for an experienced R programmer to create a script to generate seasonal time series forecasts.
The programming requirements for this project are simple and basic. All forecast data, including historical periods, will be contained in the data tables and all that will be required is to select the appropriate field. I will provide detailed instructions as to how to create the data tables needed to perform all calculations and forecasts.
This script will combine a source data file that includes the time series values with a forecast season matrix file that contains all of the seasonal divisions to create quarterly seasonal forecasts with a series of seasonal forecast models. It will generate standard ARIMA, ESM, and HOLT forecasts for the data, combine these standard forecasts with the seasonal forecasts, and finally quantify the aggregate accuracy of each forecast model using Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE).
The seasonal forecasts are based on a moving average of the Seasonal Mean Value (SMV) for each instance of a season. The Mean Value (MV) for each instance of a season (for example, the mean value for January 2002, the mean value for February 2002, etc.) is divided by the Annual Mean Value (AMV) for the calendar year to create the SMV for that instance of the season.
The forecasted value for a future season will be the Last Quarter Mean (LQM) — the mean value of the last 5 to 10 values from the quarter immediately prior to the quarter being forecasted — multiplied by either the A1, A2, or A3 forecast metric for that season. A1 is the SMV of the season 1 year prior (similar to a Seasonal Naïve forecast); A2 is the mean SMV of the two prior years; A3 is the mean SMV of the three prior years.
The seasonal forecasts will also be combined with each of the standard forecasts (ARIMA, ESM, HOLT) to create hybrid seasonal forecasts, both as a 1:1 forecast (each forecast value is the mean of the standard and the seasonal values) and as a 3:1 forecast (the mean of 3x the standard forecast plus 1x of the seasonal forecast).
I expect this project to take between 5 and 10 hours for an experienced R programmer. It is much simpler than it sounds, and once the coding is completed for the first seasonal model, it is merely repeated for the other six models.
I have an older version of the script that performs many of the required functions, but I’m not sure how useful it will be. Still, it’s available for reference.
Please submit a proposal including your relevant experience. I will provide qualified candidates with a detailed scope of work for the project, and an Excel Workbook that illustrates the tables and all of the required calculations.
The programming requirements for this project are simple and basic. All forecast data, including historical periods, will be contained in the data tables and all that will be required is to select the appropriate field. I will provide detailed instructions as to how to create the data tables needed to perform all calculations and forecasts.
This script will combine a source data file that includes the time series values with a forecast season matrix file that contains all of the seasonal divisions to create quarterly seasonal forecasts with a series of seasonal forecast models. It will generate standard ARIMA, ESM, and HOLT forecasts for the data, combine these standard forecasts with the seasonal forecasts, and finally quantify the aggregate accuracy of each forecast model using Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE).
The seasonal forecasts are based on a moving average of the Seasonal Mean Value (SMV) for each instance of a season. The Mean Value (MV) for each instance of a season (for example, the mean value for January 2002, the mean value for February 2002, etc.) is divided by the Annual Mean Value (AMV) for the calendar year to create the SMV for that instance of the season.
The forecasted value for a future season will be the Last Quarter Mean (LQM) — the mean value of the last 5 to 10 values from the quarter immediately prior to the quarter being forecasted — multiplied by either the A1, A2, or A3 forecast metric for that season. A1 is the SMV of the season 1 year prior (similar to a Seasonal Naïve forecast); A2 is the mean SMV of the two prior years; A3 is the mean SMV of the three prior years.
The seasonal forecasts will also be combined with each of the standard forecasts (ARIMA, ESM, HOLT) to create hybrid seasonal forecasts, both as a 1:1 forecast (each forecast value is the mean of the standard and the seasonal values) and as a 3:1 forecast (the mean of 3x the standard forecast plus 1x of the seasonal forecast).
I expect this project to take between 5 and 10 hours for an experienced R programmer. It is much simpler than it sounds, and once the coding is completed for the first seasonal model, it is merely repeated for the other six models.
I have an older version of the script that performs many of the required functions, but I’m not sure how useful it will be. Still, it’s available for reference.
Please submit a proposal including your relevant experience. I will provide qualified candidates with a detailed scope of work for the project, and an Excel Workbook that illustrates the tables and all of the required calculations.
Related categories:
Scripting