Package 'bayesforecast'

Description

Fit univariate time series models using 'Stan' for full Bayesian inference. A wide range of distributions and models are supported, allowing users to fit Seasonal ARIMA, ARIMAX, Dynamic Harmonic Regression, GARCH, t-student innovation GARCH models, asymmetric GARCH, Random Walks, and stochastic volatility models. Prior specifications are flexible and explicitly encourage users to apply prior distributions that actually reflect their beliefs. Model fit can easily be assessed and compared with typical visualization methods, information criteria such as loglik, AIC, BIC WAIC, Bayes factor and leave-one-out cross-validation methods.

References

Carpenter, B. and Gelman, A. and Hoffman, D. and Lee, D. and Goodrich, B. and Betancourt, M. and Brubaker, and Guo, L. and Riddell. 2017. Stan: A probabilistic programming language. Journal of Statistical Software 76(1). doi: 10.18637/jss.v076.i01.

Stan Development Team. (2018). Stan Modeling Language Users Guide and Reference Manual, Version 2.18.0. url: https://mc-stan.org.

Hyndman, R. & Khandakar, Y. (2008). Automatic time series forecasting: the forecast package for R. Journal of Statistical Software. 26(3), 1-22.doi: 10.18637/jss.v027.i03.

Tsay, R (2010). Analysis of Financial Time Series. Wiley-Interscience. 978-0470414354, second edition.

Shumway, R.H. and Stoffer, D.S. (2010).Time Series Analysis and Its Applications: With R Examples. Springer Texts in Statistics. isbn: 9781441978646. First edition.

Description

Convenience function for computing the pointwise Akaike Information Criteria method from a varstan object.

Usage

aic(x)

Arguments

Value

A numeric array of size R, containing the posterior samples of the AICc for a varstan object, where R is the number of iterations. If multiple chains are fitted, then the array is of length M*R, where M is the number of chains

Author(s)

Asael Alonzo Matamoros

Examples

library(astsa)
 model = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
 fit1 = varstan(model,iter = 500,chains = 1)

 aic1 = aic(fit1)
 mean(aic1)

Description

Convenience function for computing the pointwise corrected Akaike Information Criteria method from a varstan object.

Usage

AICc(x)

Arguments

Value

A numeric array of size R, containing the posterior samples of the AICc for a varstan object, where R is the number of iterations. If multiple chains are fitted, then the array is of length M*R, where m is the number of chains

Author(s)

Asael Alonzo Matamoros

Examples

library(astsa)
 model = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
 fit1 = varstan(model,iter = 500,chains = 1)

 aic1 = AICc(fit1)
 mean(aic1)

Description

Total annual air passengers (in millions) including domestic and international aircraft passengers of air carriers registered in Australia. 1970-2016.

Usage

air

Format

the format is: Annual Time-Series (1:27) from 1990 to 2016:

Source

fpp2

References

Hyndman, R. & Khandakar, Y. (2008). Automatic time series forecasting: the forecast package for R. Journal of Statistical Software. 26(3), 1-22.doi: 10.18637/jss.v027.i03

Description

Convert a varstan object to a stanfit object of the rstan package.

Usage

as.stan(object)

Arguments

Value

a stanfit object.

Author(s)

Asael Alonzo Matamoros

Examples

# Fitting a GARCH(1,1) model
 dat = garch(ipc,order = c(1,1,0))
 fit1 = varstan(dat,iter = 500,chains = 1)

 # Converting to a Stanfit object
 stanfit1 = as.stan(fit1)

Description

Quarterly visitor nights (in millions) spent by international tourists to Australia. 1999-2015

Usage

aust

Format

the format is: Quarterly Time-Series (1:44) from 1999 to 2015:

Source

fpp2

References

Hyndman, R. & Khandakar, Y. (2008). Automatic time series forecasting: the forecast package for R. Journal of Statistical Software. 26(3), 1-22.doi: 10.18637/jss.v027.i03

Description

Returns the best seasonal ARIMA model using a bic value, this function theauto.arima function of the forecast package to select the seasonal ARIMA model and estimates the model using a HMC sampler.

Usage

auto.sarima(ts,seasonal = TRUE,xreg = NULL,chains=4,iter=4000,warmup=floor(iter/2),
                adapt.delta = 0.9,tree.depth =10,stepwise = TRUE, series.name = NULL,
                prior_mu0 = NULL,prior_sigma0 = NULL,prior_ar = NULL, prior_ma = NULL,
                prior_sar = NULL,prior_sma = NULL, prior_breg = NULL,...)

Arguments

Details

Automatic ARIMA model fitting implemented by Rob Hyndman, this function finds the best Seasonal ARIMA model using bic, and then proceeds to fit the model using varstan function and the default priors of a Sarima model constructor.

This function provides an initial model fit for beginning the Bayesian analysis of the univariate time series. For better fit and model selection try different models and other model selection criteria such as loo or bayes_factor.

The default arguments are designed for rapid estimation of models for many time series. If you are analyzing just one time series, and can afford to take some more time, it is recommended that you set stepwise=FALSE and reduce the number of iterations per chain (iter).

For more information look at auto.arima() function of forecast package.

Value

A varstan object with the "best" fitted ARIMA model to the data

Author(s)

Asael Alonzo Matamoros

References

Hyndman, R. & Khandakar, Y. (2008). Automatic time series forecasting: the forecast package for R. Journal of Statistical Software. 26(3), 1-22.doi: 10.18637/jss.v027.i03.

Box, G. E. P. and Jenkins, G.M. (1978). Time series analysis: Forecasting and control. San Francisco: Holden-Day. Biometrika, 60(2), 297-303. doi:10.1093/biomet/65.2.297.

Kennedy, P. (1992). Forecasting with dynamic regression models: Alan Pankratz, 1991. International Journal of Forecasting. 8(4), 647-648. url: https://EconPapers.repec.org/RePEc:eee:intfor:v:8:y:1992:i:4:p:647-648.

See Also

Sarima varstan.

Examples

# Automatic Sarima model for the birth data
 auto.sarima(birth,iter = 500,chains = 1)

 # Dynamic Harmonic regression
 auto.sarima(birth,xreg = fourier(birth,K= 6),iter = 500,chains = 1)

Description

autoplot takes an object of type ts or mts and creates a ggplot object suitable for usage with stat_forecast.

Usage

## S3 method for class 'ts'
autoplot(
  object,
  series = NULL,
  xlab = "Time",
  ylab = deparse(substitute(object)),
  main = NULL,
  facets = FALSE,
  colour = TRUE,
  ...
)

## S3 method for class 'ts'
fortify(model, data, ...)

Arguments

Details

fortify.ts takes a ts object and converts it into a data frame (for usage with ggplot2).

Value

None. Function produces a ggplot2 graph.

Author(s)

Mitchell O'Hara-Wild.

See Also

plot.ts, fortify

Examples

library(ggplot2)
autoplot(USAccDeaths)

lungDeaths <- cbind(mdeaths, fdeaths)
autoplot(lungDeaths)
autoplot(lungDeaths, facets=TRUE)

Description

Preliminary autoplot methods for varstan models only valid for univariate time series models. The function prints the fitted values time series, the trace and density plots for the sampled model parameters, or the residuals' posterior mean time series.

Usage

## S3 method for class 'varstan'
autoplot(object, prob = 0.95, ...)

Arguments

Value

None. Function produces a ggplot2 graph.

Examples

library(astsa)
 sf1 = auto.sarima(ts = birth,iter = 500,chains = 1)
 # fitted model
 autoplot(sf1)

Description

Compute Bayes factors from marginal likelihoods.

Usage

## S3 method for class 'varstan'
bayes_factor(x1, x2, log = FALSE, ...)

Arguments

Details

The computation of marginal likelihoods based on bridge sampling requires a lot more posterior samples than usual. A good conservative rule of thump is perhaps 10-fold more samples (read: the default of 4000 samples may not be enough in many cases). If not enough posterior samples are provided, the bridge sampling algorithm tends to be unstable leading to considerably different results each time it is run. We thus recommend running bridge_sampler multiple times to check the stability of the results.

For more details check the bridgesampling package.

Value

The bayes factors of two models.

Examples

library(astsa)
 # Fitting a seasonal arima model
 mod1 = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
 fit1 = varstan(mod1,iter = 500,chains = 1)

 # Fitting a Dynamic harmonic regression
 mod2  = Sarima(birth,order = c(0,1,2),xreg = fourier(birth,K=6))
 fit2 = varstan(mod2,iter = 500,chains = 1)

 # compute the Bayes factor
 bayes_factor(fit1, fit2)

Description

beta(shape1,shape2)

Usage

beta(shape1 = 2, shape2 = 2)

Arguments

Details

Define a beta prior distribution using the hyper parameters shape1 and shape2, by default a beta(2,2) distribution is return.

Value

a numerical vector interpreted as a prior in Stan

Description

Convenience function for computing the pointwise Bayesian Information Criteria method from a varstan object.

Usage

bic(x)

Arguments

Value

A numeric array of size R, containing the posterior samples of the aic for a varstan object, where R is the number of iterations. If multiple chains are fitted, then the array is of length M*R, where M is the number of chains

Author(s)

Asael Alonzo Matamoros

Examples

library(astsa)
 model = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
 fit1 = varstan(model,iter = 500,chains = 1)

 bic1 = bic(fit1)
 mean(bic1)

Description

Monthly live births (adjusted) in thousands for the United States, 1948-1979.

Usage

birth

Format

the format is: Time-Series (1:373) from 1948 to 1979:

Source

astsa

References

http://www.stat.pitt.edu/stoffer/tsa4/ http://www.stat.pitt.edu/stoffer/tsda/

Description

Computes log marginal likelihood via bridge sampling, which can be used in the computation of Bayes factors and posterior model probabilities.

Usage

## S3 method for class 'varstan'
bridge_sampler(samples, ...)

Arguments

Details

The varstan class is just a thin wrapper that contains the stanfit objects.

Computing the marginal likelihood via the bridgesampler package for stanfit objects.

The computation of marginal likelihoods based on bridge sampling requires a lot more posterior samples than usual. A good conservative rule of thump is perhaps 10-fold more samples (read: the default of 4000 samples may not be enough in many cases). If not enough posterior samples are provided, the bridge sampling algorithm tends to be unstable leading to considerably different results each time it is run. We thus recommend running bridge_sampler multiple times to check the stability of the results.

For more details check the bridgesampling package.

Value

the model's marginals likelihood from the bridge_sampler package.

Examples

library(astsa)
# Fitting a seasonal ARIMA model
mod1 = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
fit1 = varstan(mod1,iter = 500,chains = 1)

fit1
bridge_sampler(fit1)

# Fitting a Dynamic harmonic regression
mod2  = Sarima(birth,order = c(0,1,2),xreg = fourier(birth,K=6))
fit2 = varstan(mod2,iter = 500,chains = 1)

fit2
bridge_sampler(fit2)

Description

cauchy(mu,sd)

Usage

cauchy(mu = 0, sd = 1)

Arguments

Details

Define a Cauchy prior distribution using the hyper parameters mu and sigma, by default a standard Cauchy(0,1) distribution is return.

Value

a numerical vector interpreted as a prior in Stan

Description

Performs a visual check of residuals in time series models, this method is inspired in the check.residuals function provided by the forecast package.

Usage

check_residuals(object, ...)

Arguments

Value

None. Function produces a ggplot2 graph.

Author(s)

Asael Alonzo Matamoros.

Examples

library(astsa)
 sf1 = auto.sarima(ts = birth,iter = 500,chains = 1)
 # fitted model
 check_residuals(sf1)

Description

chisq(df)

Usage

chisq(df = 7)

Arguments

Details

Define a gamma prior distribution using the degree freedom df hyper parameter, by default an chisq(7) distribution is return.

Value

a numerical vector interpreted as a prior in Stan

Description

The vector dem2gbp contains daily observations of the Deutschmark vs British Pound foreign exchange rate log-returns. This data set has been promoted as an informal benchmark for GARCH time-series software validation. See McCullough and Renfro (1999), and Brooks, Burke, and Persand (2001) for details. The nominal returns are expressed in percent as in Bollerslev and Ghysels (1996). The sample period is from January 3, 1984, to December 31, 1991, for a total of 1974 observations.

Usage

demgbp

Format

the format is: Time-Series (1:350) from 1984 to 1985:

Source

bayesGARCH

References

Engle, R. (1982). Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United Kingdom Inflation. Econometrica, 50(4), 987-1007. url: http://www.jstor.org/stable/1912773.

Bollerslev, T. (1986). Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics. 31(3), 307-327. doi: https://doi.org/10.1016/0304-4076(86)90063-1.

Ardia, D. and Hoogerheide, L. (2010). Bayesian Estimation of the GARCH(1,1) Model with Student-t Innovations. The R Journal. 2(7), 41-47. doi: 10.32614/RJ-2010-014.

Description

exponential(rate)

Usage

exponential(rate = 1)

Arguments

Details

Define a gamma prior distribution using the rate hyper parameter, by default an exponential(1) distribution is return.

Value

a numerical vector interpreted as a prior in Stan

Description

Extract chains of an stanfit object implemented in rstan package

Usage

extract_stan(object,pars,permuted = TRUE,inc_warmup = FALSE,include = TRUE, ...)

Arguments

Value

a list with the posterior samples of the provided parameters.

Author(s)

Asael Alonzo Matamoros

Examples

library(astsa)
 # Fitting a GARCH(1,1) model
 dat = garch(ipc,order = c(1,1,0))
 fit2 = varstan(dat,iter = 500,chains = 1)

 # Extracting the mean parameter
 mu0 = extract_stan(fit2,pars = "mu0")

Description

The function returns the posterior estimate of the fitted values of a varstan model, similar to the fit_values functions of other packages.

Usage

## S3 method for class 'varstan'
fitted(object, robust = FALSE, ...)

Arguments

Details

This function returns a time series of the predicted mean response values.

Value

A time series (ts) of predicted mean response values.

Author(s)

Asael Alonzo Matamoros

See Also

posterior_predict.varstan

Description

forecast is a generic function for forecasting from time series or varstan models. The function invokes particular methods which depend on the class of the first argument.

Usage

## S3 method for class 'varstan'
forecast(
  object,
  h = 10,
  probs = c(0.8, 0.9),
  xreg = NULL,
  robust = FALSE,
  draws = 1000,
  seed = NULL,
  ...
)

Arguments

Details

If model=NULL,the function forecast.ts makes forecasts using ets models (if the data are non-seasonal or the seasonal period is 12 or less) or stlf (if the seasonal period is 13 or more).

If model is not NULL, forecast.ts will apply the model to the object time series, and then generate forecasts accordingly.

Value

An object of class "forecast".

The function summary is used to obtain and print a summary of the results, while the function plot produces a plot of the forecasts and prediction intervals.

The generic accessors functions fitted.values and residuals extract various useful features of the value returned by forecast$model.

An object of class "forecast" is a list usually containing at least the following elements:

Author(s)

Asael Alonzo Matamoros.

See Also

The "forecast" methods of the forecast package.

Examples

fit = auto.sarima(ts = birth,iter = 500,chains = 1)
 fc = forecast(fit,h = 12)

Description

fourier returns a matrix containing terms from a Fourier series, up to order K, suitable for use in Sarima or auto.sarima.

Usage

fourier(x, K, h = NULL)

Arguments

Details

The period of the Fourier terms is determined from the time series characteristics of x. When h is missing, the length of x also determines the number of rows for the matrix returned by fourier. Otherwise, the value of h determines the number of rows for the matrix returned by fourier, typically used for forecasting. The values within x are not used.

Typical use would omit h when generating Fourier terms fitting a model and include h when generating Fourier terms for forecasting.

When x is a ts object, the value of K should be an integer and specifies the number of sine and cosine terms to return. Thus, the matrix returned has 2*K columns.

When x is a msts object, then K should be a vector of integers specifying the number of sine and cosine terms for each of the seasonal periods. Then the matrix returned will have 2*sum(K) columns.

Value

Numerical matrix.

Author(s)

Rob J Hyndman

See Also

seasonaldummy

Examples

library(astsa)
 # Dynaimc Harmonic regression
 sf1 = auto.sarima(birth,xreg = fourier(birth,K= 6),iter = 500,chains = 1)

Description

gamma(shape,rate)

Usage

gamma(shape = 2, rate = 1)

Arguments

Details

Define a gamma prior distribution using the hyper parameters shape and rate, by default an gamma(2,1) distribution is return.

Value

a numerical vector interpreted as a prior in Stan

Description

Constructor of the GARCH(s,k,h) object for Bayesian estimation in Stan.

Usage

garch(ts,order = c(1,1,0),arma = c(0,0),xreg = NULL,
             genT = FALSE,asym = "none",series.name = NULL)

Arguments

Details

The function returns a list with the data for running stan() function of rstan package.

By default the garch() function generates a GARCH(1,1) model, when genT option is TRUE a t-student innovations GARCH model (see Ardia (2010)) is generated, and for Asymmetric GARCH models use the option asym for specify the asymmetric function, see Fonseca, et. al (2019) for more details.

The default priors used in a GARCH(s,k,h) model are:

• ar ~ normal(0,0.5)

• ma ~ normal(0,0.5)

• mu0 ~ t-student(0,2.5,6)

• sigma0 ~ t-student(0,1,7)

• arch ~ normal(0,0.5)

• garch ~ normal(0,0.5)

• mgarch ~ normal(0,0.5)

• dfv ~ gamma(2,0.1)

• breg ~ t-student(0,2.5,6)

For changing the default prior use the function set_prior().

Value

The function returns a list with the data for running stan() function of rstan package.

Author(s)

Asael Alonzo Matamoros.

References

Engle, R. (1982). Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United Kingdom Inflation. Econometrica, 50(4), 987-1007. url: http://www.jstor.org/stable/1912773.

Bollerslev, T. (1986). Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics. 31(3), 307-327. doi: https://doi.org/10.1016/0304-4076(86)90063-1.

Fonseca, T. and Cequeira, V. and Migon, H. and Torres, C. (2019). The effects of degrees of freedom estimation in the Asymmetric GARCH model with Student-t Innovations. arXiv doi: arXiv: 1910.01398.

Ardia, D. and Hoogerheide, L. (2010). Bayesian Estimation of the GARCH(1,1) Model with Student-t Innovations. The R Journal. 2(7), 41-47. doi: 10.32614/RJ-2010-014.

See Also

Sarima auto.arima set_prior

Examples

# Declaring a garch(1,1) model for the ipc data.
dat = garch(ipc,order = c(1,1,0))
dat

# Declaring a t-student M-GARCH(2,3,1)-ARMA(1,1) process for the ipc data.
dat = garch(ipc,order = c(2,3,1),arma = c(1,1),genT = TRUE)
dat

# Declaring a logistic Asymmetric GARCH(1,1) process.
dat = garch(ipc,order = c(1,1,0),asym = "logit")
dat

Description

Get the sampled parameters of a varstan object.

Usage

get_parameters(object)

Arguments

Value

a vector with the sampled parameters

Author(s)

Asael Alonzo Matamoros

Examples

library(astsa)
 sf1 = auto.sarima(birth,iter = 500,chains = 1)
 get_parameters(sf1)

Description

The functions gets the defined distribution of a defined model parameter

Usage

get_prior(model,par,lag = 0)

Arguments

Value

None. Prints the prior distribution of a desired parameter.

Author(s)

Asael Alonzo Matamoros

Examples

library(astsa)
# get all the ar parameters
dat = Sarima(birth,order = c(2,1,2))
get_prior(model = dat,par = "ar")

# change the mean constant parameter
dat = set_prior(model = dat,par = "mu0",dist = student(0,2.5,7))
get_prior(dat,par = "mu0")

# change and print only the second ma parameter
dat = set_prior(model = dat,par = "ma",dist = beta(2,2),lag = 2)
get_prior(dat,par = "ma")

Description

Plot of the auto-correlation function for a univariate time series.

Usage

ggacf(y, title = NULL)

Arguments

Value

None. Function produces a ggplot2 graph.

Author(s)

Asael Alonzo Matamoros

Examples

x = rnorm(100)
ggacf(x)

Description

Plots a histogram and density estimates using ggplot.

Usage

gghist(y, title = NULL, xlab = NULL, ylab = "counts", bins, add.normal = TRUE)

Arguments

Value

None. Function produces a ggplot2 graph.

Author(s)

Rob J Hyndman

Examples

x = rnorm(100)
gghist(x,add.normal = TRUE)

Description

Plot the quantile-quantile plot and quantile-quantile line using ggplot.

Usage

ggnorm(y, title = NULL, add.normal = TRUE)

Arguments

Value

None. Function produces a ggplot2 graph.

Author(s)

Asael Alonzo Matamoros

Examples

x = rnorm(100)
ggnorm(x)

Description

Plot of the partial autocorrelation function for a univariate time series.

Usage

ggpacf(y, title = NULL)

Arguments

Value

None.

Author(s)

Mitchell O'Hara-Wild and Asael Alonzo Matamoros

Examples

x = rnorm(100)
ggpacf(x)

Description

Constructor of the ets("A","A","Z") object for Bayesian estimation in Stan.

Usage

Holt(ts,damped = FALSE,xreg = NULL,genT = FALSE,series.name = NULL)

Arguments

Details

The function returns a list with the data for running stan() function of rstan package.

When genT option is TRUE a t-student innovations ssm model (see Ardia (2010)) is generated see Fonseca, et. al (2019) for more details.

The default priors used in a ssm( ) model are:

• level ~ normal(0,0.5)

• trend ~ normal(0,0.5)

• damped~ normal(0,0.5)

• sigma0 ~ t-student(0,1,7)

• level1 ~ normal(0,1)

• trend1 ~ normal(0,1)

• dfv ~ gamma(2,0.1)

• breg ~ t-student(0,2.5,6)

For changing the default prior use the function set_prior().

Value

The function returns a list with the data for running stan() function of rstan package.

Author(s)

Asael Alonzo Matamoros.

References

Fonseca, T. and Cequeira, V. and Migon, H. and Torres, C. (2019). The effects of degrees of freedom estimation in the Asymmetric GARCH model with Student-t Innovations. arXiv doi: arXiv: 1910.01398.

See Also

Sarima auto.arima set_prior garch

Examples

mod1 = Holt(ipc)

# Declaring a Holt damped trend model for the ipc data.
mod2 = Holt(ipc,damped = TRUE)

Description

Constructor of the ets("A","A","A") object for Bayesian estimation in Stan.

Usage

Hw(ts,damped = FALSE,xreg = NULL,period = 0,genT = FALSE,series.name = NULL)

Arguments

Details

The function returns a list with the data for running stan() function of rstan package.

When genT option is TRUE a t-student innovations ssm model (see Ardia (2010)) is generated see Fonseca, et. al (2019) for more details.

The default priors used in a ssm( ) model are:

• level ~ normal(0,0.5)

• Trend ~ normal(0,0.5)

• damped~ normal(0,0.5)

• Seasonal ~ normal(0,0.5)

• sigma0 ~ t-student(0,1,7)

• level1 ~ normal(0,1)

• trend1 ~ normal(0,1)

• seasonal1 ~ normal(0,1)

• dfv ~ gamma(2,0.1)

• breg ~ t-student(0,2.5,6)

For changing the default prior use the function set_prior().

Value

The function returns a list with the data for running stan() function of rstan package.

Author(s)

Asael Alonzo Matamoros.

References

Fonseca, T. and Cequeira, V. and Migon, H. and Torres, C. (2019). The effects of degrees of freedom estimation in the Asymmetric GARCH model with Student-t Innovations. arXiv doi: arXiv: 1910.01398.

See Also

Sarima auto.arima set_prior garch

Examples

mod1 = Hw(ipc)

# Declaring a Holt Winters damped trend model for the ipc data.
mod2 = Hw(ipc,damped = TRUE)

# Declaring an additive Holt-Winters model for the birth data
mod3 = Hw(birth,damped = FALSE)

Description

inverse.chisq(df)

Usage

inverse.chisq(df = 7)

Arguments

Details

Define a inverse chi square prior distribution using the hyper parameter df, by default an inverse.chisq(df = 2) distribution is return.

If sigma has a chi square distribution then 1/sigma has n inverse chi square distribution.

Value

a numerical vector interpreted as a prior in Stan

Description

inverse.gamma(shape,rate)

Usage

inverse.gamma(shape = 2, rate = 1)

Arguments

Details

Define a inverse.gamma prior distribution using the hyper parameters shape and rate, by default an inverse.gamma(2,1) distribution is return.

If sigma has a gamma distribution then 1/sigma has n inverse gamma distribution. The rate parameter is the inverse of an scale parameter.

Value

a numerical vector interpreted as a prior in Stan

Description

Monthly return coefficients for the inflation. An economic indicator of a country's economy.

Usage

ipc

Format

A time series of monthly data from 1980 to 2018.

Source

https://www.bch.hn/series_estadisticas.php

Description

jeffey.df()

Usage

jeffrey()

Details

Define a non informative Jeffrey's prior distribution, by default an jeffrey.df( ) distribution is return.

This prior can only be used in garch models with t-student innovations, or Bekk models with generalized t-student distribution.

Value

a numerical vector interpreted as a prior in Stan

Description

laplace(mu,sd)

Usage

laplace(mu = 0, sd = 1)

Arguments

Details

Define a Laplace prior distribution using the hyper parameters mu and sigma, by default a standard Laplace distribution is return.

The laplace distribution is exactly the same as the double exponential distribution

Value

a numerical vector interpreted as a prior in Stan

Description

LKJ(df)

Usage

LKJ(df = 2)

Arguments

Details

Define a Lewandowski Kurowicka and Joe (LKJ) matrix correlation prior distribution using the degree freedom df hyper parameter,by default a LKJ(2) distribution is return.

Value

a numerical vector interpreted as a prior in Stan

References

Lewandowski D, Kurowicka D, Joe H (2009). "Generating random correlation matrices based on vines and extended onion method." Journal of Multivariate Analysis, 100(9), 1989 2001. ISSN 0047-259X. doi:https://doi.org/10.1016/j.jmva.2009.04.008. URL: http://www.sciencedirect.com/science/article/pii/S0047259X09000876.

Description

Constructor of the ets("A","N","N") object for Bayesian estimation in Stan.

Usage

LocalLevel(ts,xreg = NULL,genT = FALSE,series.name = NULL)

Arguments

Details

The function returns a list with the data for running stan() function of rstan package.

By default the ssm() function generates a local level model (or a ets("A","N","N") or exponential smoothing model from the forecast package). If trend is set TRUE, then a local trend ssm model is defined (a equivalent ets("A","A","N") or Holt model from the forecast package). For damped trend models set damped to TRUE. If seasonal is set to TRUE a seasonal local level model is defined (a equivalent ets("A","N","A") model from the forecast package). For a Holt-Winters method (ets("A","A","A")) set Trend and seasonal to TRUE.

When genT option is TRUE a t-student innovations ssm model (see Ardia (2010)) is generated see Fonseca, et. al (2019) for more details.

The default priors used in a ssm( ) model are:

• level ~ normal(0,0.5)

• sigma0 ~ t-student(0,1,7)

• level1 ~ normal(0,1)

• dfv ~ gamma(2,0.1)

• breg ~ t-student(0,2.5,6)

For changing the default prior use the function set_prior().

Value

The function returns a list with the data for running stan() function of rstan package.

Author(s)

Asael Alonzo Matamoros.

References

Fonseca, T. and Cequeira, V. and Migon, H. and Torres, C. (2019). The effects of degrees of freedom estimation in the Asymmetric GARCH model with Student-t Innovations. arXiv doi: arXiv: 1910.01398.

See Also

Sarima auto.arima set_prior garch

Examples

mod1 = LocalLevel(ipc)

Description

Convenience function for extracting the pointwise log-likelihood matrix or array from a fitted Stan model.

Usage

## S3 method for class 'varstan'
log_lik(object, permuted = TRUE, ...)

Arguments

Value

Usually, an S x N matrix containing the pointwise log-likelihood samples, where S is the number of samples and N is the number of observations in the data. If permuted is FALSE, an S x N x R array is returned, where R is the number of fitted chains.

References

Vehtari, A., Gelman, A., & Gabry J. (2016). Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. In Statistics and Computing, doi:10.1007/s11222-016-9696-4.

Gelman, A., Hwang, J., & Vehtari, A. (2014). Understanding predictive information criteria for Bayesian models. Statistics and Computing. 24, 997-1016.

Watanabe, S. (2010). Asymptotic equivalence of Bayes cross validation and widely applicable information criterion in singular learning theory. The Journal of Machine Learning Research. 11, 3571-3594.

Examples

library(astsa)
 model = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
 fit1 = varstan(model,iter = 500,chains = 1)

 log1 = log_lik(fit1)
 log1

Description

Convenience function for extracting the posterior sample of the accumulated log-likelihood array from a fitted varstan object.

Usage

loglik(object, permuted = TRUE)

Arguments

Value

A real value with the accumulated log likelihood.

References

Vehtari, A., Gelman, A., & Gabry J. (2016). Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. In Statistics and Computing, doi:10.1007/s11222-016-9696-4.

Gelman, A., Hwang, J., & Vehtari, A. (2014). Understanding predictive information criteria for Bayesian models. Statistics and Computing. 24, 997-1016.

Watanabe, S. (2010). Asymptotic equivalence of Bayes cross validation and widely applicable information criterion in singular learning theory. The Journal of Machine Learning Research. 11, 3571-3594.

Examples

library(astsa)
 model = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
 fit1 = varstan(model,iter = 500,chains = 1)

 log1 = loglik(fit1)
 log1

Description

The loo method for varstan objects. Computes approximate leave-one-out cross-validation using Pareto smoothed importance sampling (PSIS-LOO CV).

Usage

## S3 method for class 'varstan'
loo(x, ...)

Arguments

Value

an object from the loo class with the results of the Pareto-Smooth Importance Sampling, leave one out cross validation for model selection.

References

Vehtari, A., Gelman, A., & Gabry J. (2016). Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. In Statistics and Computing, doi:10.1007/s11222-016-9696-4.

Gelman, A., Hwang, J., & Vehtari, A. (2014). Understanding predictive information criteria for Bayesian models. Statistics and Computing. 24, 997-1016.

Watanabe, S. (2010). Asymptotic equivalence of Bayes cross validation and widely applicable information criterion in singular learning theory. The Journal of Machine Learning Research. 11, 3571-3594.

See Also

• The loo package vignettes for demonstrations.

• psis() for the underlying Pareto Smoothed Importance Sampling (PSIS) procedure used in the LOO-CV approximation.

• pareto-k-diagnostic for convenience functions for looking at diagnostics.

• loo_compare() for model comparison.

Examples

library(astsa)
model = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
fit1 = varstan(model,iter = 500,chains = 1)

loo1 = loo(fit1)
loo1

Description

Convenient way to call MCMC plotting functions implemented in the bayesplot package.

Usage

## S3 method for class 'varstan'
mcmc_plot(
  object,
  pars = NULL,
  combo = c("dens", "trace"),
  fixed = FALSE,
  exact_match = FALSE,
  ...
)

mcmc_plot(object, ...)

Arguments

Value

A ggplot object that can be further customized using the ggplot2 package.

Examples

## Not run: 
sf1 = stan_ssm(ipc,iter = 500,chains = 1)

# plot posterior intervals
mcmc_plot(sf1)

# only show population-level effects in the plots
mcmc_plot(sf1, pars = "level")

## End(Not run)

Description

The function returns a string with the users defined model for the given time series data.

Usage

model(object,...)

Arguments

Details

if object is a varstan object the function will print the information of the defined model inside of the object. If object is one of the model classes (like Sarima, garch, SVM or varma), then it will print the model information as well.

For full information of the model with the used priors use the function report or just print the object.

Value

a string with the defined time series model.

Author(s)

Asael Alonzo Matamoros.

See Also

report print

Examples

library(astsa)
model1 = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
model(model1)

Description

naive is the model constructor for a random walk model applied to y. This is equivalent to an ARIMA(0,1,0) model. naive() is simply a wrapper to maintain forecast package similitude. seasonal returns the model constructor for a seasonal random walk equivalent to an ARIMA(0,0,0)(0,1,0)m model where m is the seasonal period.

Usage

naive(ts,seasonal = FALSE,m = 0)

Arguments

Details

The random walk with drift model is

$Y_t = mu_0 + Y_{t-1} + epsilon_t$

where $epsilon_t$ is a normal iid error.

The seasonal naive model is

$Y_t = mu_0 + Y_{t-m} + epsilon_t$

where $epsilon_t$ is a normal iid error.

Value

The function returns a list with the data for running stan() function of rstan package.

Author(s)

Asael Alonzo Matamoros

References

Hyndman, R. & Khandakar, Y. (2008). Automatic time series forecasting: the forecast package for R. Journal of Statistical Software. 26(3), 1-22.doi: 10.18637/jss.v027.i03.

Box, G. E. P. and Jenkins, G.M. (1978). Time series analysis: Forecasting and control. San Francisco: Holden-Day. Biometrika, 60(2), 297-303. doi:10.1093/biomet/65.2.297.

Kennedy, P. (1992). Forecasting with dynamic regression models: Alan Pankratz, 1991. International Journal of Forecasting. 8(4), 647-648. url: https://EconPapers.repec.org/RePEc:eee:intfor:v:8:y:1992:i:4:p:647-648.

See Also

Sarima

Examples

library(astsa)
# A seasonal Random-walk model.
model = naive(birth,seasonal = TRUE)
model

Description

normal(mu,sd)

Usage

normal(mu = 0, sd = 1)

Arguments

Details

Define a normal prior distribution using the hyper parameters mu and sigma, by default a standard normal distribution is return.

Value

a numerical vector interpreted as a prior in Stan

Description

Annual oil production (millions of tonnes), Saudi Arabia, 1965-2013.

Usage

oildata

Format

the format is: Annual Time-Series (1:18) from 1996 to 2013:

Source

fpp2

References

Hyndman, R. & Khandakar, Y. (2008). Automatic time series forecasting: the forecast package for R. Journal of Statistical Software. 26(3), 1-22.doi: 10.18637/jss.v027.i03

Description

Preliminary plot methods for varstan models only valid for univariate time series models. The function prints the fitted values time series, the trace and density plots for the sampled model parameters, or the residuals' posterior mean time series.

Usage

## S3 method for class 'varstan'
plot(x, prob = 0.95, ...)

Arguments

Value

None. Function produces a ggplot2 graph.

Examples

library(astsa)
 sf1 = auto.sarima(ts = birth,iter = 500,chains = 1)
 # fitted model
 plot(sf1)

Description

Compute posterior samples of the expected value/mean of the posterior predictive distribution. Can be performed for the data used to fit the model (posterior predictive checks) or for new data. By definition, these predictions have smaller variance than the posterior predictions performed by the posterior_predict.varstan method. This is because only the uncertainty in the mean is incorporated in the samples computed by posterior_epred while any residual error is ignored. However, the estimated means of both methods averaged across samples should be very similar.

Usage

## S3 method for class 'varstan'
posterior_epred(
  object,
  h = 0,
  xreg = NULL,
  robust = FALSE,
  draws = 1000,
  seed = NULL,
  ...
)

Arguments

Value

An array of predicted mean response values. For categorical and ordinal models, the output is an S x N x C array. Otherwise, the output is an S x N matrix, where S is the number of posterior samples, N is the number of observations, and C is the number of categories. In multivariate models, an additional dimension is added to the output which indexes along the different response variables.

Description

The posterior_interval function computes Bayesian posterior uncertainty intervals. These intervals are often referred to as credible intervals, for more details see rstanarm

Usage

posterior_interval(mat, prob = 0.9, ...)

Arguments

Value

A matrix with two columns and as many rows as model parameters (or the subset of parameters specified by pars and/or regex_pars). For a given value of prob, $p$, the columns correspond to the lower and upper 100*p% interval limits and have the names $100\alpha/2$ and $100(1 - \alpha/2)$%, where $\alpha = 1-p$. For example, if prob=0.9 is specified (a 90% interval), then the column names will be "5%" and "95%", respectively.

Author(s)

Asael Alonzo Matamoros

Description

The posterior predictive distribution is the distribution of the outcome implied by the model after using the observed data to update our beliefs about the unknown parameters in the model. Simulating data from the posterior predictive distribution using the observed predictors is useful for checking the fit of the model. Drawing from the posterior predictive distribution at interesting values of the predictors also lets us visualize how a manipulation of a predictor affects (a function of) the outcome(s). With new observations of predictor variables we can use the posterior predictive distribution to generate predicted outcomes.

Usage

## S3 method for class 'varstan'
posterior_predict(
  object,
  h = 0,
  xreg = NULL,
  robust = FALSE,
  draws = 1000,
  seed = NULL,
  ...
)

Arguments

Value

A draws by h data.frame of simulations from the posterior predictive distribution. Each row of the data.frame is a vector of predictions generated using a single draw of the model parameters from the posterior distribution.

Author(s)

Asael Alonzo Matamoros

Description

This is a convenience function for computing $y - y_{h}$ The method for stanreg objects calls posterior_predict internally, where as the method accepts the data.frame returned by posterior_predict as input and can be used to avoid multiple calls to posterior_predict.

Usage

## S3 method for class 'varstan'
predictive_error(
  object,
  newdata = NULL,
  xreg = NULL,
  draws = 1000,
  seed = NULL,
  ...
)

Arguments

Value

A draws by nrow(newdata) data.frame.

Note

If object is a varstan object of a varma model then newdata has to be a matrix with number of cols as the dimension of the time series and number of rows as the number new elements.

If object is a posterior_predict data.frame, then the length of newdata has to be equal to the ncol of object.

If object is a posterior_predict data.frame, for a varma model, then the dimension product of newdata matrix has to be equal to the ncol of object.

See Also

posterior_predict function from rstanarm package, to draw from the posterior predictive distribution without computing predictive errors.

Description

Print a garch model

Usage

## S3 method for class 'garch'
print(x, ...)

Arguments

Value

None. prints the object

Description

Print a Holt model

Usage

## S3 method for class 'Holt'
print(x, ...)

Arguments

Value

None. prints the object.

Description

Print a Holt-Winter model

Usage

## S3 method for class 'Hw'
print(x, ...)

Arguments

Value

None. prints the object.

Description

Print a Local Level model

Usage

## S3 method for class 'LocalLevel'
print(x, ...)

Arguments

Value

None. prints the object.

Description

Print a naive model

Usage

## S3 method for class 'naive'
print(x, ...)

Arguments

Value

None. prints the object

Description

Print a Sarima model

Usage

## S3 method for class 'Sarima'
print(x, ...)

Arguments

Value

None. prints the object

Description

Print a state-space model

Usage

## S3 method for class 'ssm'
print(x, ...)

Arguments

Value

None. prints the object.

Description

Print a Stochastic Volatility model

Usage

## S3 method for class 'SVM'
print(x, ...)

Arguments

Value

None. prints the object

Description

Print a varstan object

Usage

## S3 method for class 'varstan'
print(x, ...)

Arguments

Value

None. prints the object.

Description

The function returns a report with the users defined model for the given time series data and all the current defined priors of the model.

Usage

## S3 method for class 'varstan'
prior_summary(object, ...)

Arguments

Details

if object is a varstan object the function will print the information of the defined model inside of the object. If object is one of the model classes (like Sarima or garch) then it will print the report information as well.

Value

none. prints a string with the defined time series model report

Author(s)

Asael Alonzo Matamoros

Examples

library(astsa)
dat2 = garch(birth,order = c(1,1,0))
prior_summary(dat2)

Description

The function returns a report with the users defined model for the given time series data and all the current defined priors of the model.

Usage

report(object,...)

Arguments

Details

if object is a varstan object the function will print the information of the defined model inside of the object. If object is one of the model classes (like Sarima or garch) then it will print the report information as well.

Value

none. prints a string with the defined time series model report

Author(s)

Asael Alonzo Matamoros

Examples

library(astsa)
dat2 = garch(birth,order = c(1,1,0))
report(dat2)

Description

The function returns the posterior estimate of the residuals of a varstan model, similar to the residual functions of other packages.

Usage

## S3 method for class 'varstan'
residuals(object, robust = FALSE, ...)

Arguments

Details

This function only extracts the point-wise estimate of the time series residuals for extracting all the data use extract_stan() or posterior_intervals function

Value

An array with the posterior estimate of the residuals of the time series model,

Author(s)

Asael Alonzo Matamoros

Description

Constructor of the SARIMA model for Bayesian estimation in Stan.

Usage

Sarima(ts,order = c(1,0,0),seasonal = c(0,0,0),xreg = NULL,period = 0,series.name = NULL)

Arguments

Details

The function returns a list with the data for running stan() function of rstan package

If xreg option is used, the model by default will cancel the seasonal differences adjusted (D = 0). If a value d > 0 is used, all the regressor variables in xreg will be difference as well.

The default priors used in Sarima are:

• ar ~ normal(0,0.5)

• ma ~ normal(0,0.5)

• mu0 ~ t-student(0,2.5,6)

• sigma0 ~ t-student(0,1,7)

• sar ~ normal(0,0.5)

• sma ~ normal(0,0.5)

• breg ~ t-student(0,2.5,6)

For changing the default prior use the function set_prior

Value

The function returns a list with the data for running stan() function of rstan package.

Author(s)

Asael Alonzo Matamoros

References

Box, G. E. P. and Jenkins, G.M. (1978). Time series analysis: Forecasting and control. San Francisco: Holden-Day. Biometrika, 60(2), 297-303. doi:10.1093/biomet/65.2.297.

Kennedy, P. (1992). Forecasting with dynamic regression models: Alan Pankratz, 1991. International Journal of Forecasting. 8(4), 647-648. url: https://EconPapers.repec.org/RePEc:eee:intfor:v:8:y:1992:i:4:p:647-648.

Hyndman, R. & Khandakar, Y. (2008). Automatic time series forecasting: the forecast package for R. Journal of Statistical Software. 26(3), 1-22.doi: 10.18637/jss.v027.i03

See Also

garch set_prior

Examples

# Declare a multiplicative seasonal ARIMA model for the birth data.

library(astsa)
model = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
model

#Declare an Dynamic Harmonic Regression model for the birth data.
model = Sarima(birth,order = c(1,0,1),xreg = fourier(birth,K = 2))
model

Description

setting a prior distribution to an specify model parameter.

Usage

set_prior(model, par = "ar", dist = normal(), lag = 0)

Arguments

Details

varstan provides its own functions to manipulate the parameter prior, this functions return a prior_dist class, the dist argument only accepts this objects.

lag parameter is an optional value to change the prior distribution of one parameter in particular, this argument is only valid for: "ar","ma", "arch", "garch", "mgarch", or "breg" par arguments. lag has to be a integer lower than the total amount of lagged parameters of the model. For example, to ONLY change the prior of the second "arch" parameter in a garch(3,1) model, a lag = 2 values must be specified.

For varma and Bekk models the covariance matrix Sigma is factorized as follows:

              Sigma = D' Omega D

Where Omega is the correlation matrix that accepts an LKJ prior distribution D is a diagonal matrix with the inverse std deviations

For changing the degree freedom in a LKJ distribution for omega use par = "LKJ" and dist = LKJ(df), where df are the desired degree freedom.

For changing the the priors in the diagonal D use par = "sigma0" and select one of the available prior distributions.

For ar, ma garch, arch parameters in varma and Bekk models only normal distributions priors with different mu and sd are accepted. Even if get_prior accepts its change, Stan will change it to a normal(0,1) prior.

Value

a time series model class specified in varstan with the changed prior.

Author(s)

Asael Alonzo Matamoros

Examples

library(astsa)
dat = Sarima(birth,order = c(1,1,2))
dat = set_prior(model = dat,par = "ar",dist = normal(0,2))
dat

dat = set_prior(model = dat,par = "mu0",dist = student(mu=0,sd = 2.5,df = 7))
dat

dat = set_prior(model = dat,par = "ma",dist= beta(shape1 = 2,shape2 = 2),lag = 2)
dat

Description

Constructor of the ets("Z","Z","Z") object for Bayesian estimation in Stan.

Usage

ssm(ts,trend = FALSE,damped = FALSE,seasonal = FALSE,xreg = NULL,
           period = 0,genT = FALSE,series.name = NULL)

Arguments

Details

The function returns a list with the data for running stan() function of rstan package.

By default the ssm() function generates a local level model (or a ets("A","N","N") or exponential smoothing model from the forecast package). If trend is set TRUE, then a local trend ssm model is defined (a equivalent ets("A","A","N") or Holt model from the forecast package). For damped trend models set damped to TRUE. If seasonal is set to TRUE a seasonal local level model is defined (a equivalent ets("A","N","A") model from the forecast package). For a Holt-Winters method (ets("A","A","A")) set Trend and seasonal to TRUE.

When genT option is TRUE a t-student innovations ssm model (see Ardia (2010)) is generated see Fonseca, et. al (2019) for more details.

The default priors used in a ssm( ) model are:

• level ~ normal(0,0.5)

• Trend ~ normal(0,0.5)

• damped~ normal(0,0.5)

• Seasonal ~ normal(0,0.5)

• sigma0 ~ t-student(0,1,7)

• level1 ~ normal(0,1)

• trend1 ~ normal(0,1)

• seasonal1 ~ normal(0,1)

• dfv ~ gamma(2,0.1)

• breg ~ t-student(0,2.5,6)

For changing the default prior use the function set_prior().

Value

The function returns a list with the data for running stan() function of rstan package.

Author(s)

Asael Alonzo Matamoros.

References

Fonseca, T. and Cequeira, V. and Migon, H. and Torres, C. (2019). The effects of degrees of freedom estimation in the Asymmetric GARCH model with Student-t Innovations. arXiv doi: arXiv: 1910.01398.

See Also

Sarima auto.arima set_prior garch

Examples

mod1 = ssm(ipc)

# Declaring a Holt model for the ipc data.
mod2 = ssm(ipc,trend = TRUE,damped = TRUE)

# Declaring an additive Holt-Winters model for the birth data
mod3 = ssm(birth,trend = TRUE,damped = TRUE,seasonal = TRUE)

Description

Fitting a GARCH(s,k,h) model in Stan.

Usage

stan_garch(
  ts,
  order = c(1, 1, 0),
  arma = c(0, 0),
  xreg = NULL,
  genT = FALSE,
  asym = "none",
  chains = 4,
  iter = 2000,
  warmup = floor(iter/2),
  adapt.delta = 0.9,
  tree.depth = 10,
  stepwise = TRUE,
  prior_mu0 = NULL,
  prior_sigma0 = NULL,
  prior_ar = NULL,
  prior_ma = NULL,
  prior_mgarch = NULL,
  prior_arch = NULL,
  prior_garch = NULL,
  prior_breg = NULL,
  prior_gamma = NULL,
  prior_df = NULL,
  series.name = NULL,
  ...
)

Arguments

Details

The function returns a varstan object with the fitted model.

By default the garch() function generates a GARCH(1,1) model, when genT option is TRUE a t-student innovations GARCH model (see Ardia (2010)) is generated, and for Asymmetric GARCH models use the option asym for specify the asymmetric function, see Fonseca, et. al (2019) for more details.

The default priors used in a GARCH(s,k,h) model are:

• ar ~ normal(0,0.5)

• ma ~ normal(0,0.5)

• mu0 ~ t-student(0,2.5,6)

• sigma0 ~ t-student(0,1,7)

• arch ~ normal(0,0.5)

• garch ~ normal(0,0.5)

• mgarch ~ normal(0,0.5)

• dfv ~ gamma(2,0.1)

• breg ~ t-student(0,2.5,6)

For changing the default prior use the function set_prior().

Value

A varstan object with the fitted GARCH model.

Author(s)

Asael Alonzo Matamoros.

References

Engle, R. (1982). Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United Kingdom Inflation. Econometrica, 50(4), 987-1007. url: http://www.jstor.org/stable/1912773.

Bollerslev, T. (1986). Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics. 31(3), 307-327. doi: https://doi.org/10.1016/0304-4076(86)90063-1.

Fonseca, T. and Cequeira, V. and Migon, H. and Torres, C. (2019). The effects of degrees of freedom estimation in the Asymmetric GARCH model with Student-t Innovations. arXiv doi: arXiv: 1910.01398.

Ardia, D. and Hoogerheide, L. (2010). Bayesian Estimation of the GARCH(1,1) Model with Student-t Innovations. The R Journal. 2(7), 41-47. doi: 10.32614/RJ-2010-014.

See Also

Sarima auto.arima set_prior

Examples

# Declaring a garch(1,1) model for the ipc data.
 sf1 = stan_garch(ipc,order = c(1,1,0),iter = 500,chains = 1)

 # Declaring a t-student M-GARCH(2,3,1)-ARMA(1,1) process for the ipc data.
 sf2 = stan_garch(ipc,order = c(2,3,1),arma = c(1,1),genT = TRUE,iter = 500,chains = 1)

Description

Fitting an Holt state-space model in Stan.

Usage

stan_Holt(
  ts,
  damped = FALSE,
  xreg = NULL,
  genT = FALSE,
  chains = 4,
  iter = 2000,
  warmup = floor(iter/2),
  adapt.delta = 0.9,
  tree.depth = 10,
  stepwise = TRUE,
  prior_sigma0 = NULL,
  prior_level = NULL,
  prior_level1 = NULL,
  prior_trend = NULL,
  prior_trend1 = NULL,
  prior_damped = NULL,
  prior_breg = NULL,
  prior_df = NULL,
  series.name = NULL,
  ...
)

Arguments

Details

The function returns a varstan object with the fitted model.

When genT option is TRUE a t-student innovations ssm model (see Ardia (2010)) is generated see Fonseca, et. al (2019) for more details.

The default priors used in a ssm( ) model are:

• level ~ normal(0,0.5)

• Trend ~ normal(0,0.5)

• damped~ normal(0,0.5)

• sigma0 ~ t-student(0,1,7)

• level1 ~ normal(0,1)

• trend1 ~ normal(0,1)

• dfv ~ gamma(2,0.1)

• breg ~ t-student(0,2.5,6)

For changing the default prior use the function set_prior().

Value

A varstan object with the fitted SSM model.

Author(s)

Asael Alonzo Matamoros.

References

Fonseca, T. and Cequeira, V. and Migon, H. and Torres, C. (2019). The effects of degrees of freedom estimation in the Asymmetric GARCH model with Student-t Innovations. arXiv doi: arXiv: 1910.01398.

See Also

Sarima auto.arima set_prior garch

Examples

# Declaring a Holt model for the ipc data.
 sf1 = stan_Holt(ipc,iter = 500,chains = 1)

 # Declaring a Holt damped trend model for the ipc data.
 sf2 = stan_Holt(ipc,damped = TRUE,iter = 500,chains = 1)

Description

Fitting a Holt-Winters state-space model in Stan.

Usage

stan_Hw(
  ts,
  damped = FALSE,
  xreg = NULL,
  period = 0,
  genT = FALSE,
  chains = 4,
  iter = 2000,
  warmup = floor(iter/2),
  adapt.delta = 0.9,
  tree.depth = 10,
  stepwise = TRUE,
  prior_sigma0 = NULL,
  prior_level = NULL,
  prior_level1 = NULL,
  prior_trend = NULL,
  prior_trend1 = NULL,
  prior_damped = NULL,
  prior_seasonal = NULL,
  prior_seasonal1 = NULL,
  prior_breg = NULL,
  prior_df = NULL,
  series.name = NULL,
  ...
)

Arguments

Details

The function returns a varstan object with the fitted model.

When genT option is TRUE a t-student innovations ssm model (see Ardia (2010)) is generated see Fonseca, et. al (2019) for more details.

The default priors used in a ssm( ) model are:

• level ~ normal(0,0.5)

• Trend ~ normal(0,0.5)

• damped~ normal(0,0.5)

• Seasonal ~ normal(0,0.5)

• sigma0 ~ t-student(0,1,7)

• level1 ~ normal(0,1)

• trend1 ~ normal(0,1)

• seasonal1 ~ normal(0,1)

• dfv ~ gamma(2,0.1)

• breg ~ t-student(0,2.5,6)

For changing the default prior use the function set_prior().

Value

A varstan object with the fitted SSM model.

Author(s)

Asael Alonzo Matamoros.

References

Fonseca, T. and Cequeira, V. and Migon, H. and Torres, C. (2019). The effects of degrees of freedom estimation in the Asymmetric GARCH model with Student-t Innovations. arXiv doi: arXiv: 1910.01398.

See Also

Sarima auto.arima set_prior garch

Examples

# Declaring a Holt-Winters model for the ipc data.
 sf1 = stan_Hw(ipc,iter = 500,chains = 1)

 # Declaring a Holt-Winters damped trend model for the ipc data.
 sf2 = stan_ssm(ipc,damped = TRUE,iter = 500,chains = 1)

Description

Fitting a Local level state-space model in Stan.

Usage

stan_LocalLevel(
  ts,
  xreg = NULL,
  genT = FALSE,
  chains = 4,
  iter = 2000,
  warmup = floor(iter/2),
  adapt.delta = 0.9,
  tree.depth = 10,
  stepwise = TRUE,
  prior_sigma0 = NULL,
  prior_level = NULL,
  prior_level1 = NULL,
  prior_breg = NULL,
  prior_df = NULL,
  series.name = NULL,
  ...
)

Arguments

Details

The function returns a varstan object with the fitted model.

When genT option is TRUE a t-student innovations ssm model (see Ardia (2010)) is generated see Fonseca, et. al (2019) for more details.

The default priors used in a Local_level( ) model are:

• level ~ normal(0,0.5)

• sigma0 ~ t-student(0,1,7)

• level1 ~ normal(0,1)

• dfv ~ gamma(2,0.1)

• breg ~ t-student(0,2.5,6)

For changing the default prior use the function set_prior().

Value

A varstan object with the fitted Local Level model.

Author(s)

Asael Alonzo Matamoros.

References

Fonseca, T. and Cequeira, V. and Migon, H. and Torres, C. (2019). The effects of degrees of freedom estimation in the Asymmetric GARCH model with Student-t Innovations. arXiv doi: arXiv: 1910.01398.

See Also

Sarima auto.arima set_prior garch

Examples

# Declaring a local level model for the ipc data.
 sf1 = stan_LocalLevel(ipc,iter = 500,chains = 1)

Description

naive is the model constructor for a random walk model applied to y. This is equivalent to an ARIMA(0,1,0) model. naive() is simply a wrapper to maintain forecast package similitude. seasonal returns the model constructor for a seasonal random walk equivalent to an ARIMA(0,0,0)(0,1,0)m model where m is the seasonal period.

Usage

stan_naive(
  ts,
  seasonal = FALSE,
  m = 0,
  chains = 4,
  iter = 2000,
  warmup = floor(iter/2),
  adapt.delta = 0.9,
  tree.depth = 10,
  stepwise = TRUE,
  prior_mu0 = NULL,
  prior_sigma0 = NULL,
  series.name = NULL,
  ...
)

Arguments

Details

The random walk with drift model is

$Y_t = mu_0 + Y_{t-1} + epsilon_t$

where $epsilon_t$ is a normal iid error.

The seasonal naive model is

$Y_t = mu_0 + Y_{t-m} + epsilon_t$

where $epsilon_t$ is a normal iid error.

Value

A varstan object with the fitted naive Random Walk model.

Author(s)

Asael Alonzo Matamoros

References

Hyndman, R. & Khandakar, Y. (2008). Automatic time series forecasting: the forecast package for R. Journal of Statistical Software. 26(3), 1-22.doi: 10.18637/jss.v027.i03.

Box, G. E. P. and Jenkins, G.M. (1978). Time series analysis: Forecasting and control. San Francisco: Holden-Day. Biometrika, 60(2), 297-303. doi:10.1093/biomet/65.2.297.

Kennedy, P. (1992). Forecasting with dynamic regression models: Alan Pankratz, 1991. International Journal of Forecasting. 8(4), 647-648. url: https://EconPapers.repec.org/RePEc:eee:intfor:v:8:y:1992:i:4:p:647-648.

See Also

Sarima

Examples

library(astsa)
 # A seasonal Random-walk model.
 sf1 = stan_naive(birth,seasonal = TRUE,iter = 500,chains = 1)

Description

Fitting a SARIMA model in Stan.

Usage

stan_sarima(
  ts,
  order = c(1, 0, 0),
  seasonal = c(0, 0, 0),
  xreg = NULL,
  period = 0,
  chains = 4,
  iter = 2000,
  warmup = floor(iter/2),
  adapt.delta = 0.9,
  tree.depth = 10,
  stepwise = TRUE,
  prior_mu0 = NULL,
  prior_sigma0 = NULL,
  prior_ar = NULL,
  prior_ma = NULL,
  prior_sar = NULL,
  prior_sma = NULL,
  prior_breg = NULL,
  series.name = NULL,
  ...
)

Arguments

Details

The function returns a varstan object with the fitted model.

If xreg option is used, the model by default will cancel the seasonal differences adjusted (D = 0). If a value d > 0 is used, all the regressor variables in xreg will be difference as well.

The default priors used in Sarima are:

• ar ~ normal(0,0.5)

• ma ~ normal(0,0.5)

• mu0 ~ t-student(0,2.5,6)

• sigma0 ~ t-student(0,1,7)

• sar ~ normal(0,0.5)

• sma ~ normal(0,0.5)

• breg ~ t-student(0,2.5,6)

Value

A varstan object with the fitted SARIMA model.

Author(s)

Asael Alonzo Matamoros

References

Box, G. E. P. and Jenkins, G.M. (1978). Time series analysis: Forecasting and control. San Francisco: Holden-Day. Biometrika, 60(2), 297-303. doi:10.1093/biomet/65.2.297.

Kennedy, P. (1992). Forecasting with dynamic regression models: Alan Pankratz, 1991. International Journal of Forecasting. 8(4), 647-648. url: https://EconPapers.repec.org/RePEc:eee:intfor:v:8:y:1992:i:4:p:647-648.

Hyndman, R. & Khandakar, Y. (2008). Automatic time series forecasting: the forecast package for R. Journal of Statistical Software. 26(3), 1-22.doi: 10.18637/jss.v027.i03

See Also

garch set_prior

Examples

library(astsa)
 # Declare a multiplicative seasonal ARIMA model for the birth data.
 sf1 = stan_sarima(birth,order = c(0,1,2),
                   seasonal = c(1,1,1),iter = 500,chains = 1)


 #Declare an Dynamic Harmonic Regression model for the birth data.
 sf2 = stan_sarima(birth,order = c(1,0,1),
                   xreg = fourier(birth,K = 2),iter = 500,chains = 1)

Description

Fitting an Additive linear State space model in Stan.

Usage

stan_ssm(
  ts,
  trend = FALSE,
  damped = FALSE,
  seasonal = FALSE,
  xreg = NULL,
  period = 0,
  genT = FALSE,
  chains = 4,
  iter = 2000,
  warmup = floor(iter/2),
  adapt.delta = 0.9,
  tree.depth = 10,
  stepwise = TRUE,
  prior_sigma0 = NULL,
  prior_level = NULL,
  prior_level1 = NULL,
  prior_trend = NULL,
  prior_trend1 = NULL,
  prior_damped = NULL,
  prior_seasonal = NULL,
  prior_seasonal1 = NULL,
  prior_breg = NULL,
  prior_df = NULL,
  series.name = NULL,
  ...
)

Arguments

Details

The function returns a varstan object with the fitted model.

By default the ssm() function generates a local level model (or a ets("A","N","N") or exponential smoothing model from the forecast package). If trend is set TRUE, then a local trend ssm model is defined (a equivalent ets("A","A","N") or Holt model from the forecast package). For damped trend models set damped to TRUE. If seasonal is set to TRUE a seasonal local level model is defined (a equivalent ets("A","N","A") model from the forecast package). For a Holt-Winters method (ets("A","A","A")) set Trend and seasonal to TRUE.

When genT option is TRUE a t-student innovations ssm model (see Ardia (2010)) is generated see Fonseca, et. al (2019) for more details.

The default priors used in a ssm( ) model are:

• level ~ normal(0,0.5)

• Trend ~ normal(0,0.5)

• damped~ normal(0,0.5)

• Seasonal ~ normal(0,0.5)

• sigma0 ~ t-student(0,1,7)

• level1 ~ normal(0,1)

• trend1 ~ normal(0,1)

• seasonal1 ~ normal(0,1)

• dfv ~ gamma(2,0.1)

• breg ~ t-student(0,2.5,6)

For changing the default prior use the function set_prior().

Value

A varstan object with the fitted SSM model.

Author(s)

Asael Alonzo Matamoros.

References

Fonseca, T. and Cequeira, V. and Migon, H. and Torres, C. (2019). The effects of degrees of freedom estimation in the Asymmetric GARCH model with Student-t Innovations. arXiv doi: arXiv: 1910.01398.

See Also

Sarima auto.arima set_prior garch

Examples

# Declaring a local level model for the ipc data.
 sf1 = stan_ssm(ipc,iter = 500,chains = 1)

 # Declaring a Holt model for the ipc data.
 sf2 = stan_ssm(ipc,trend = TRUE,damped = TRUE,iter = 500,chains = 1)

Description

Fitting a Stochastic Volatility model (SVM) in Stan.

Usage

stan_SVM(
  ts,
  arma = c(0, 0),
  xreg = NULL,
  chains = 4,
  iter = 2000,
  warmup = floor(iter/2),
  adapt.delta = 0.9,
  tree.depth = 10,
  stepwise = TRUE,
  prior_mu0 = NULL,
  prior_sigma0 = NULL,
  prior_ar = NULL,
  prior_ma = NULL,
  prior_alpha = NULL,
  prior_beta = NULL,
  prior_breg = NULL,
  series.name = NULL,
  ...
)

Arguments

Details

The function returns a varstan object with the fitted model.

Value

A varstan object with the fitted SVM model.

Author(s)

Asael Alonzo Matamoros

References

Sangjoon,K. and Shephard, N. and Chib.S (1998). Stochastic Volatility: Likelihood Inference and Comparison with ARCH Models. Review of Economic Studies. 65(1), 361-93. url: https://www.jstor.org/stable/2566931.

Tsay, R (2010). Analysis of Financial Time Series. Wiley-Interscience. 978-0470414354, second edition.

Shumway, R.H. and Stoffer, D.S. (2010).Time Series Analysis and Its Applications: With R Examples. Springer Texts in Statistics. isbn: 9781441978646. First edition.

See Also

garch set_prior

Examples

# Declares a SVM model for the IPC data
 sf1 = stan_SVM(ipc,arma = c(1,1),iter = 500,chains = 1)

Description

student(mu,sd)

Usage

student(mu = 0, sd = 1, df = 5)

Arguments

Details

Define a t student prior distribution using the hyper parameters mu, sigma and df as degree freedom, by default a standard t-student(0,1,5) distribution with 5 degree freedom is return.

Value

a numerical vector interpreted as a prior in Stan

Description

Summaries of parameter estimates and MCMC convergence diagnostics (Monte Carlo error, effective sample size, Rhat).

Usage

## S3 method for class 'varstan'
summary(object, robust = FALSE, prob = 0.9, ...)

Arguments

Value

A data.frame with the posterior mean, standard error, credible intervals, effective sample size (ess),and Rhat for all the model parameters in a varstan model, if robust is TRUE then the posterior mean and standard error, are replaced by the posterior mean and MAD.

Author(s)

Asael Alonzo Matamoros.

Description

Constructor of the Stochastic Volatility model (SVM) for Bayesian estimation in Stan.

Usage

SVM(ts,arma = c(0,0),xreg = NULL,series.name = NULL)

Arguments

Details

The function returns a list with the data for running stan() function of rstan package.

Value

The function returns a list with the data for running stan() function of rstan package.

Author(s)

Asael Alonzo Matamoros

References

Sangjoon,K. and Shephard, N. and Chib.S (1998). Stochastic Volatility: Likelihood Inference and Comparison with ARCH Models. Review of Economic Studies. 65(1), 361-93. url: https://www.jstor.org/stable/2566931.

Tsay, R (2010). Analysis of Financial Time Series. Wiley-Interscience. 978-0470414354, second edition.

Shumway, R.H. and Stoffer, D.S. (2010).Time Series Analysis and Its Applications: With R Examples. Springer Texts in Statistics. isbn: 9781441978646. First edition.

See Also

garch set_prior

Examples

# Declares a SVM model for the IPC data

model = SVM(ipc,arma = c(1,1))
model

Description

uniform(shape1,shape2)

Usage

uniform(min = 0, max = 1)

Arguments

Details

Define a beta prior distribution using the hyper parameters min and max, by default a uniform(0,1) distribution is return.

Value

a numerical vector interpreted as a prior in Stan

Description

Constructor of the varstan object for Bayesian estimation in Stan.

Usage

varstan(model,chains=4,iter=2000,warmup=floor(iter/2),
               adapt.delta = 0.90,tree.depth =10,...)

Arguments

Details

The function estimates one of the defined models in Stan using the stan() function for sampling.

This is the principal package's function and the link with Stan, this function fits the posterior distribution of every parameter for a defined model using a HMC method.

Every estimated model become a varstan object, with different methods for summary, diagnostic, forecast and plotting.

Defining priors

Default priors are chosen to be non or very weakly informative so that their influence on the results will. However, after getting more familiar with Bayesian statistics, I recommend you to start thinking about reasonable informative priors for your model parameters.

Those can be changed using the function set_prior() before estimating the model with the varstan() function. For checking the defined priors use get_prior() and report() functions.

Adjusting the sampling behavior of Stan

In addition to choosing the number of iterations, warmup samples, and chains, users can control the behavior of the NUTS sampler, by using the control argument. The most important reason to use control is to decrease (or eliminate at best) the number of divergent transitions that cause a bias in the obtained posterior samples. Whenever you see the warning "There were x divergent transitions after warmup." you should really think about increasing adapt_delta. Increasing adapt_delta will slow down the sampler but will decrease the number of divergent transitions threatening the validity of your posterior samples.

Another problem arises when the depth of the tree being evaluated in each iteration is exceeded. This is less common than having divergent transitions, but may also bias the posterior samples. When it happens, Stan will throw out a warning suggesting to increase max_treedepth. For more details on the control argument see stan.

Value

a varstan object with the estimated time series model.

A varstan object is a list that contains the following values:

• Stanfit a Stanfit object returned by rstan pacakge.

• stan.parmaters The parameters used in Stan for the sample.

• model The defined model for the time series.

• series.name The time series' name.

• ts The provided time series data.

Author(s)

Asael Alonzo Matamoros

References

Carpenter, B. and Gelman, A. and Hoffman, D. and Lee, D. and Goodrich, B. and Betancourt, M. and Brubaker, and Guo, L. and Riddell. 2017. Stan: A probabilistic programming language. Journal of Statistical Software 76(1). doi: 10.18637/jss.v076.i01.

Stan Development Team. (2018). Stan Modeling Language Users Guide and Reference Manual, Version 2.18.0. url: http://mc-stan.org.

Paul-Christian Buerkner (2017). brms: An R Package for Bayesian Multilevel Models Using Stan. Journal of Statistical Software, 80(1), 1-28. doi:10.18637/jss.v080.i01

Hyndman, R. & Khandakar, Y. (2008). Automatic time series forecasting: the forecast package for R. Journal of Statistical Software. 26(3), 1-22.doi: 10.18637/jss.v027.i03.

See Also

rstan:stan.

Examples

library(astsa)
 # Fitting a seasonal ARIMA model
 mod1 = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
 fit1 = varstan(mod1,iter = 500,chains = 1)
 fit1

 # Fitting a GARCH(1,1) model
 dat = garch(ipc,order = c(1,1,0))
 fit2 = varstan(dat,iter = 500,chains = 1)
 fit2

Description

Compute the widely applicable information criterion (WAIC) based on the posterior likelihood using the loo package. For more details see waic.

Usage

## S3 method for class 'varstan'
waic(x, ...)

Arguments

Details

See the loo_compare function of the loo package for more details on model comparisons.

Value

An object of class loo. With the estimates of the Watanabe-Akaike Information criteria.

References

Vehtari, A., Gelman, A., & Gabry J. (2016). Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. In Statistics and Computing, doi:10.1007/s11222-016-9696-4.

Gelman, A., Hwang, J., & Vehtari, A. (2014). Understanding predictive information criteria for Bayesian models. Statistics and Computing. 24, 997-1016.

Watanabe, S. (2010). Asymptotic equivalence of Bayes cross validation and widely applicable information criterion in singular learning theory. The Journal of Machine Learning Research. 11, 3571-3594.

Examples

library(astsa)
 model = Sarima(birth,order = c(0,1,2),seasonal = c(1,1,1))
 fit1 = varstan(model,iter = 500,chains = 1)

 waic1 = waic(fit1)
 waic1