search

TylerRayRogers / ECMT-674-Project

Develop forecasting models of County level Sales Tax collections in the State of Texas
2 stars 0 forks source link

Question #2 #2

Open TylerRayRogers opened 3 years ago

TylerRayRogers commented 3 years ago

a. Present your preferred ETS and ARIMA forecasting models for the Sales Tax variable based on a six-month “hold-out” sample where you forecast into the last six months of the data (that is, define your “training” set as ending in August 2020). i. Present a basic model description for your preferred ETS and ARIMA models along with residual diagnostics. ii. Present a six-month ahead forecast for September, 2020 – February, 2021. How do your forecasts compare to the actual values for the Sales Tax variable? iii. Present a series of 6 one-step-ahead forecasts September, 2020 – February, 2021. How do your forecasts compare to the actual values for the Sales Tax variable? iv. Based on your analysis in i-iii, what is your preferred forecasting model, ETS or ARIMA? b. Using the full dataset and the model you developed in 2.a., present a 12 month ahead path forecast for March, 2021-Februrary, 2022.

Connorhull commented 3 years ago

2a. Present your preferred ETS and ARIMA forecasting models for the Sales Tax variable based on a six-month “hold-out” sample where you forecast into the last six months of the data (that is, define your “training” set as ending in August 2020).

i. Present a basic model description for your preferred ETS and ARIMA models along with residual diagnostics.

[Need assistance answering this question - Having trouble understanding question]

Connorhull commented 3 years ago

ii. Present a six-month ahead forecast for September, 2020 – February, 2021. How do your forecasts compare to the actual values for the Sales Tax variable?

ETS and Arima Comparison

Both the ETS forecast and the ARIMA forecast are able to predict six months ahead in a fairly accurate manner when compared to the actual data. Although visually very similar, the ARIMA forecast is more accurate when you compare accuracy measures. More specifically, the MPE of the Arima forecast was 7.07% compared to the MPE of the ETS forecast at 7.7%.

ETS Accuracy Measure image

ARIMA Accuracy Measure image

Connorhull commented 3 years ago

iii. Present a series of 6 one-step-ahead forecasts September, 2020 – February, 2021. How do your forecasts compare to the actual values for the Sales Tax variable?

Onestep Comparisons

ETS Onestep Accuracy Measure image

ARIMA Onestep Accuracy Measure image

How do your forecasts compare to the actual values for the Sales Tax variable? Visually, it seems the Arima onestep forecast is more accurate compared to the ETS onestep forecast when using the actual data set as a base. In addition, across the accuracy measurements the Arima onestep forecast beats out the ETS onestep forecast in almost every measure.

Connorhull commented 3 years ago

iv. Based on your analysis in i-iii, what is your preferred forecasting model, ETS or ARIMA?

Overall, in both onestep and basic forecasting models it seems the ARIMA forecasting model is superior. Specifically, the ARIMA forecasting model is preferred over the ETS model because this model preforms superiorly to ETS in Mean Percentage Error (MPE) and Root Mean Square Error (RMSE). In my opinion, these two accuracy metrics are the most important because both include actual rather than absolute values of the forecast, positive and negative forecast errors can offset each other; as a result the formula can be used as a measure of the bias in the forecasts. Please see actual accuracy measure tables for more details.

Connorhull commented 3 years ago

2b. Using the full dataset and the model you developed in 2.a., present a 12 month ahead path forecast for March, 2021-February, 2022.

12 Month ARIMA Forecast (2b)