R TimeSeries - Revision history

Rb: /* Additional Resources */

2017-08-28T13:01:09Z

‎Additional Resources

Rb: /* Additional Resources */

2017-08-28T12:57:18Z

‎Additional Resources

Rb: /* ARMA models */

2017-01-26T20:16:40Z

‎ARMA models

Rb: /* Online Demo */

2016-02-17T23:00:49Z

‎Online Demo

Rb at 23:00, 17 February 2016

2016-02-17T23:00:32Z

Rb: /* Importing Data */

2016-02-17T22:19:57Z

‎Importing Data

Rb: /* Importing Data */

2016-02-17T22:16:45Z

‎Importing Data

Rb: /* Forecasting ARMA Models */

2016-02-17T22:06:58Z

‎Forecasting ARMA Models

Rb: /* ARMA models */

2015-07-17T13:58:51Z

‎ARMA models

Rb: /* Selecting subsets of data */

2015-07-15T13:32:42Z

‎Selecting subsets of data

← Older revision		Revision as of 13:01, 28 August 2017
Line 141:		Line 141:
	* To access some very useful data-series in a very convenient way we will also use the [https://www.quandl.com/help/r QUANDL package].		* To access some very useful data-series in a very convenient way we will also use the [https://www.quandl.com/help/r QUANDL package].
	* A good intro into a date-time format which you need if you have intra-day data can be found [http://neondataskills.org/R/time-series-convert-date-time-class-POSIX/ here]		* A good intro into a date-time format which you need if you have intra-day data can be found [http://neondataskills.org/R/time-series-convert-date-time-class-POSIX/ here]
		+	* [https://faculty.washington.edu/ezivot/econ424/Working%20with%20Time%20Series%20Data%20in%20R.pdf Here] is another tutorial on how to deal with time-series data .

← Older revision		Revision as of 12:57, 28 August 2017
Line 140:		Line 140:
	* We are using the package "forecast" authored by [http://robjhyndman.com Rob Hyndman] who has also written an online textbook on the topic of forecasting [https://www.otexts.org/fpp]		* We are using the package "forecast" authored by [http://robjhyndman.com Rob Hyndman] who has also written an online textbook on the topic of forecasting [https://www.otexts.org/fpp]
	* To access some very useful data-series in a very convenient way we will also use the [https://www.quandl.com/help/r QUANDL package].		* To access some very useful data-series in a very convenient way we will also use the [https://www.quandl.com/help/r QUANDL package].
		+	* A good intro into a date-time format which you need if you have intra-day data can be found [http://neondataskills.org/R/time-series-convert-date-time-class-POSIX/ here]

← Older revision		Revision as of 20:16, 26 January 2017
Line 84:		Line 84:

	delivers exactly the same results. If you don't believe it, try it!		delivers exactly the same results. If you don't believe it, try it!
		+
		+	Before we discuss the output of such an ARIMA estimation there is one further little detail that is worth mentioning. Whenever we are using autoregressive lags, e.g. <math>p</math> lags of <math>y_t</math>, we are in some sense loosing <math>p</math> observations. The values of <math>y_1</math> to <math>y_p</math> only appear on the right hand side but not on the left hand side as dependent variable as we do not have all conditioning information. If we were to estimate the model only for the <math>y_{p+1}</math> onwards, we are essentially treating <math>y_1</math> to <math>y_p</math> as conditioning information only and we are not trying to fit the model to these values. This means that we are not using all information efficiently. It is not straightforward to explain how to use these data well (read James Hamilton's textbook on Time-Series Analysis if you want to know), but we are lucky, the clever people who wrote the forecast package, did the right thing. So when you use the <source enclose=none>arima</source> function as described above all is good.

	The output of this function is a list (just as the output of the <source enclose=none>LM()</source> function. To figure out what information you can extract from this estimated model type		The output of this function is a list (just as the output of the <source enclose=none>LM()</source> function. To figure out what information you can extract from this estimated model type

@@ Line 131: / Line 131: @@
 == Online Demo ==
-Here is an online clip which goes through this material [https://youtu.be/FYwTl9s4lVc].
+Here is an online clip which goes through this material [https://youtu.be/FYwTl9s4lVc?hd=1].
 == Additional Resources ==

← Older revision		Revision as of 23:00, 17 February 2016
Line 128:		Line 128:

	then we will only see the last 40 observations of the series used to estimate the model (this plot is not shown here).		then we will only see the last 40 observations of the series used to estimate the model (this plot is not shown here).
		+
		+	== Online Demo ==
		+
		+	Here is an online clip which goes through this material [https://youtu.be/FYwTl9s4lVc].

	== Additional Resources ==		== Additional Resources ==

@@ Line 18: / Line 18: @@
      data <- read.csv("UKCPI.csv")  # Opens UKCPI.csv from working directory
-which will produce a dataframe (<source enclose=none>data</source>) with two variables, one giving dates (<source enclose=none>DATE</source>) and the other containing the actual CPI data for 1988Q1 to 2013Q4 (<source enclose=none>UKCPI</source>).
+which will produce a dataframe (<source enclose=none>data</source>) with two variables, one giving dates (<source enclose=none>DATE</source>) and the other containing the actual CPI data for 1988Q1 to 2013Q4 (<source enclose=none>UKCPI</source>). You could find more recent data e.g. from the excellent FRED database [https://research.stlouisfed.org/fred2/].
 == Basic Time-Series Data Transformations ==

@@ Line 13: / Line 13: @@
 == Importing Data ==
-Here we will initially use a dataset on UK CPI [[media:UKCPI.xls|UKCPI.xls]]. Doenload this and save it as a csv file as this will facilitate the upload to R.
+Here we will initially use a dataset on UK CPI [[media:UKCPI.xls|UKCPI.xls]]. Download this and save it as a csv file as this will facilitate the upload to R.
      setwd("YOUR DIRECTORY")              # This sets the working directory

@@ Line 127: / Line 127: @@
      plot(forecast(fit_diff_400,h=10), include = 40)
-then we will only see the last 40 observations of the series used to estimate the model.
+then we will only see the last 40 observations of the series used to estimate the model (this plot is not shown here).
 == Additional Resources ==

← Older revision		Revision as of 13:58, 17 July 2015
Line 66:		Line 66:

	If the series that is being modelled with this ARMA(p,q) process is a differenced series, e.g. <source enclose=none>dUKCPI</source>, then we would actually call this a ARIMA(p,1,q) model. If however we model the UK CPI directly then we would also call it ARIMA(p,0,q).		If the series that is being modelled with this ARMA(p,q) process is a differenced series, e.g. <source enclose=none>dUKCPI</source>, then we would actually call this a ARIMA(p,1,q) model. If however we model the UK CPI directly then we would also call it ARIMA(p,0,q).
		+
		+	It should be noted that the forecast package which we use actually estimates the following model:
		+
		+	<math>(y_t-m) = \alpha_1 (y_{t-1}-m) + ... + \alpha_p (y_{t-p}-m) + \epsilon_t + \delta_1 \epsilon_{t-1} + ... + \delta_q \epsilon_{t-q} </math>
		+
		+	in other words, we drop the constant but instead estimate a mean parameter m. This has the advantage that the estimated value for m (it will be reported as the constant) can actually be interpreted as a mean value for the modelled series.

	=== Estimating ARMA Models ===		=== Estimating ARMA Models ===

@@ Line 57: / Line 57: @@
      dUKCPI_subset <- window(dUKCPI, start=c(2001, 1), end=c(2012, 4))
-It should be obvious what this command does. A note of caution is due, however, the window command only works for series that are not part of a dataframe. As you can see above we defined <code>dUKCPI</code> outside the dataframe. If you wanted to apply the <code>window</code> function to <code>UKCPI</code> series you would first have to define it outside the dataframe as follows: <code></code>
+It should be obvious what this command does. A note of caution is due, however, the window command only works for series that are not part of a dataframe. As you can see above we defined <code>dUKCPI</code> outside the dataframe. If you wanted to apply the <code>window</code> function to <code>UKCPI</code> series you would first have to define it outside the dataframe as follows: <code> UKCPI <- data$UKCPI</code>.
 == ARMA models ==