Automatically fit a model to a time series:

The model selected for the data:

Create a forecast using some monthly temperature data:

The monthly temperature data and a 24-month forecast:

Forecast the average temperature for June 1, 2014:

A 95% confidence interval about the forecast:

Fit a particular model to a time series:

Fit a "GARCH"(1,1) model to the data:

The model and parameter estimates:

Fit a time series model to a 100-business-day time series:

Simulate the next 30 business days 50 times:

Use the original data to simulate future observations with TimeSeriesModel:

Fit a model:

Use the best-fit process to simulate 10 future observations:

No information about time stamps or initial values is passed to RandomFunction:

Simulate using TimeSeriesModel to use information given by the original data:

Automatically select a model for a time series:

Equivalently, specify Automatic:

Find a model in a particular family:

Search the autoregressive family of models:

Select the best model from an explicit set of candidates:

Search the ARMA family with orders and from 0 to 2:

Fit a model with fixed orders:

Fit an "ARIMA"(2,1,3) model to the data:

Fix the integration order and search for the autoregressive and moving average orders:

Parameterizations for the ARIMA family of models:

Automatically detect the orders , , and :

Alternatively, use Automatic for all parameter orders:

Equivalent ways to fix the integration order while leaving and Automatic:

Equivalent ways to fix and while leaving the integration order Automatic:

Parameterizations for the SARIMA family of models:

Automatically detect , , , , , , and :

Alternatively, use Automatic for all parameter orders:

Equivalent ways to fix the seasonality parameter while leaving all others Automatic:

Equivalent ways to fix the seasonality and integration orders and :

Obtain a list of available properties:

Extract the best-fitting process:

Equivalently, use Normal:

Show some fit criteria for the selected model:

The criterion that was used to choose the best model:

Report the top models that were fitted to the data:

The entries from the table:

Obtain a list of candidate models fitted to the data:

Compare the models by AIC:

Check whether fit residuals are autocorrelated:

There is a significant spike at lag 3:

Adding a moving-average component removes the autocorrelation at lag 3:

Check whether the partial correlation function for the fit residuals is zero:

There is a significant spike at lag 3:

Adding a moving-average component removes the partial correlation at lag 3:

Test model residuals for autocorrelation:

There are significant spikes at lags 3 and 4:

Adding a moving-average component removes the significant autocorrelations:

Obtain values used in residual whiteness diagnostic plots:

The first 5 autocorrelation values and 95% confidence region used in the "ACFPlot":

The first 5 partial autocorrelation values and 95% confidence region used in the "PACFPlot":

The Ljung–Box -values for lags 1 to 12 and the critical value used in the plot:

Report the values and significance of model parameters:

The value entries from the table:

Obtain parameter standard errors from the parameter covariance matrix:

Standard errors can be obtained directly:

Set the confidence level to use for prediction limits:

Use a 90% confidence level:

The confidence level can be changed after fitting:

The ConfidenceLevel setting affects thresholds for feature detection:

Require that a constant be estimated for the model:

Set the constant to zero:

By default, parameter orders are selected using a stepwise search:

Use a grid search instead:

Perform a hybrid search:

Use different selection criteria to pick the best model:

By default, "AIC" is used:

Use "BIC":

Set the maximum number of candidate models to keep for reporting:

Retain a maximum of 15 models:

Keep all models that were fitted:

By default, a maximum of 10 models are kept:

Set the maximum allowable autoregressive and moving-average orders:

The maximum orders limit the number of models that can be fitted:

The stepwise method fits models within "MaxStepSize" orders of the current best model:

By default, the maximum order step is set to 1:

Increasing the step size can improve quality but may increase search time:

Compare to a grid search:

Use WorkingPrecision to obtain higher-precision estimates:

Forecast the average temperature for the next month at your location:

Local average monthly temperature for the past 10 years:

Fit a time series model:

Next month's forecast:

Plot a forecast for the next year with 95% confidence bands:

The following data contains the monthly total number of US international airline passengers (in thousands) from January 1949 to December 1960:

Split the data at January 1958:

Fit a time series model to the airline passenger data before 1958:

Compare the forecasted and actual number of airline passengers for June of 1959:

The forecast from 1958 to 1961 compared to the actual data:

Simulate airline passenger counts seven years into the future and compare to a forecast:

Fit a time series model to the data:

Create the seven-year forecast and prediction limits:

A plot of the forecast:

Simulate 100 paths seven years into the future:

Show the 95% prediction limit with the simulated paths:

Fit a time series model to the differences between actual and estimated prime numbers:

Fit a model to the differences:

Show the forecast of the next 100 differences with 95% prediction limits:

Use the model to improve the estimate of the 300^th prime:

The original and improved estimators compared to the actual primes:

The number of accidental deaths in the United States was recorded monthly for six years:

Fit an ARProcess to the data:

The model misses the seasonality in the data:

The diagnostic plots suggest a model with seasonality of 12 is needed:

The seasonal model seems to generally outperform the non-seasonal one:

Obtain an empirical estimate of the 90% forecast prediction limits two months into the future:

Compare to the asymptotic estimates:

Number of mink furs sold yearly by the Hudson's Bay Company, Canada, from 1850 to 1911:

Fit a model to the data:

Simulate the values of the model for the next 10 years:

The simulated values may be negative, but the mean function provides a positive forecast:

Plot the data with the forecast given by the simulation:

Retail monthly sales in United States:

Create TimeSeries from the selection:

Plot the sales with grid lines at December peaks:

Fit a seasonal model:

The process parameters:

Find the forecast for the next 7 years:

Calculate 95% confidence bands for the forecast:

There is an upper and a lower band:

Plot the forecast within the 95% confidence region:

The following time series are bi-quarterly sales and advertising of US patent-medicine proprietor Lydia Pinkham. Pinkham began making Vegetable Compound as a home remedy, sharing it with her neighbors. Her compound, a blend of ground herbs, was 18% alcohol by content. In 1875 the Pinkham family decided to go into business selling the medicine, which Pinkham claimed could cure any female complaint from nervous prostration to a prolapsed uterus. It quickly gained acceptance, and the business was soon grossing close to $300,000 a year. Not until the 1920s, when federal regulation of drugs and advertising increased, did the Lydia E. Pinkham Medicine Co. reduce both the alcohol content of the medicine and the claims for its efficacy:

Find a model for the data:

Forecast the values for the next 5 years:

The model predicts steep decline of customer trust in the efficacy of the Pinkham remedy: