I am currently using an auto regression model for time series data. What other models can perform better. I am basically trying to model the vibration of an accelerometer data.
You need to start with ARIMA first by following the Box-Jenkins procedure, then move on to determine if additional modifications are required according to the diagnostic checking on normality, autocorrelation, and heteroskedasticity. Most importantly, stationarity check by the ADF test is a must for ARIMA models and alike. For additional guidance, please read my working paper on RG.
The answer depends on whether you want to explain the model, or whether you just want to use it for forecasting? If you only want to forecast, then the univariate time series models generally outperform the more sophisticated models. On the other hand, if you have covariates that you want (or need) to include in the model, then you will need to go to more complex modelling approaches, such as ARIMA, or complex regression models.
I am adding a link to a paper I wrote a long time ago on TSA that should give you some direction. At the very least, it will provide you with good references to pursue.
Article Evaluating Disease Management Program Effectiveness: An Intr...
If your data is a univariate one, then i will recommend the Box Jenkins Approach for model buidling in Time series for example ARIMA (i.e. if you difference your unstationary series), ARMA, SARIMA, SARMA etc for forecasting. If it is a multivariate one, then you need to apply either Vector Autoregression Model (VAR) or Vector Error Correction model (VEC) that is, whether the sets of the multivariate variables are co-integrated or not. You can find attached a paper i wrote in applying SARIMA to a univariate data.
Article Using SARIMA to Forecast Monthly Mean Surface Air Temperatur...
You need to start with ARIMA first by following the Box-Jenkins procedure, then move on to determine if additional modifications are required according to the diagnostic checking on normality, autocorrelation, and heteroskedasticity. Most importantly, stationarity check by the ADF test is a must for ARIMA models and alike. For additional guidance, please read my working paper on RG.
Basically from Time Series data one can expect;
1. Trend and tendencies present in the data set.
2. Forecasting.
In order to study 1st objective, we basically go for smoothing techniques such as MA, exponential smoothing (depend on situation we use particular method), etc.
And for the 2nd we use different models such as ARIMA, SARIMA, Holt-Winter model etc.
Hello everyone,
I am using Eviews.
I have performed unit root test on my multivariate model. however, i have the variables of interest stationary at level(no tend no constant).
what should be my next step ?
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression