Classes in hpiR

Andy Krause

2020-04-01

Introduction

This vignette explains the various classes of objects and data used in the hpiR package. By executing the code in this vignette you can create examples of all classes of objects used in the hpiR package.

Class Structures

The table below shows the basic information on the various S3 classes used in the package. The naming convention of classes is such that prefixes indicate the domain (or the type) of the information contained in the object. The major domains are:


The hpi collection object is the “core” unit of the hpiR package. The object will contain all the information, temporary objects (data, models, etc.) and analysis about a single house price index.

The serieshpi object is a “super-unit” that contains multiple house price indexes within it. Indexes within a series share the same data, modeling approach and parameters; they differ only in the length of the indexes. Additionally there are two types of analyses that can only be applied to series: forecast accuracy (calcForecastError()) and revision (calcRevision()). Also note that in an serieshpi object the data objects are moved out of the hpi objects and into the serieshpi object to avoid duplication of data storage.

The table below shows the various classes of objects used with hpiR, their type, what they inherit from, what function creates them and a short description. A tilde (~) in front of the inherits value signifies that some of the components of the object inherit from that class, but not all.


Type Class Inherits Created by Description
Super
serieshpi many createSeries() A series of progressively longer indexes
Core
hpi many **Index() wrappers Unified house price index collection object containing data, model and index as well as additional analytics that can be added to it
hpiaccuracy data.frame calcAccuracy() Accuracy of the index
Data
hpidata data.frame dateToPeriod() Transaction data.frame with dates converted to relative periods
rtdata hpidata, data.frame rtCreateTrans() Data.frame of repeat transactions with standardized field names
heddata hpidata, data.frame hedCreatTrans() Data.frame of transactions with some standardized field names for use in hedonic model approach
Model
hpimodel Many (ex. lm, rlm) hpiModel() Flexible list of model results
rtmodel Many (ex. lm, rlm) rtModel() List of repeat transaction model results
hedmodel Many (ex. lm, rlm) hedModel() List of hedonic model results
Index
hpiindex ~ts modelToIndex() List of index information. $index slot is a ts object
indexsmooth ts smoothIndex() Smoothed index
indexvolatility NONE calcVolatility() List of rolling, mean and median index volatility
Series
seriesrevision NONE calcRevision() List of the period, mean and median revision
seriesaccuracy hpiaccuracy, data.frame calcSeriesAccuracy() Accuracy of the series
Visualization
plotindex gg, ggplot plot.hpiindex(), plot.hpi Plotting object for an index
plotseries gg, ggplot plot.serieshpi() Plotting object for an index series
plotrevision gg, ggplot plot.seriesrevision() Plotting object for revision statistics
plotaccuracy gtable, gTree, grob, gDesc plot.hpiaccuracy(), plot.seriesaccuracy Plotting object for accuracy results
plotvolatility gg, ggplot plot.indexvolatility() Plotting object for index volatility

Examples And Explanation

Small examples of any of the classes in the table above are available in the hpiR package as pre-loaded data. Use data(ex_'name of class') to access. Ex: data(ex_hpidata).

Loading

Begin by loading the hpiR package. It is assumed that you’ve followed the instructions in the “introduction” vignette in order to obtain the package.

Next, load the example dataset of Seattle house transactions into your R session. We’ll limit to just the central city (area equal to 13, 14, and 15). This can also be accessed via data(ex_sales).

Data

‘hpidata’ objects

hpidata objects inherit from data.frame, but are very flexible and the collection of fields that may be in these objects is solely dependent on the data being used. The only requirements for creating an hpidata object is that there must be an R formatted date field that can be turned into a period field.

The dateToPeriod() function will create an hpidata object by converting the date into a period. The only difference between the hpidata object and the original data.frame is the addition of trans_date and trans_period columns and the periodicity, min_date, max_date and period_table attributes.

Approach-based data objects

Each of the index creation approaches (currently just rt and hed) will have their own data object classes that include the necessary fields and structures for these approach types. The name format here is “xxdata”", where ‘xx’ is the approach abbreviation.

The rtdata object holds repeat transactions and contains the eight standard fields shown below as well as the period_table attribute.

The heddata object requires no changes from the original hpidata object and looks the same, with the exception of field standardization (pinx is now prop_id, sale_id is now trans_id, sale_price is now price, sale_date is now date. )

Modeling

Results from modeling the data prior to creating indexes is stored in hpimodel objects. These objects contain eight standard slots, shown below. The slots are standardized across approaches, with the exception of the model_obj slot which will contain an object that is specific to the approach being used to estimate the index.

Below we see an example of a model_obj from a repeat transactions model. This is very similar to what you’d get from a call to a linear model (lm(...))

We also show an example of a model_obj from a hedonic price method, again similar to an lm(...) output.

Indexes and Smoothing

The modelToIndex() function will take an hpimodel object and turn it into an hpiindex object. hpiindex objects contain five standard slots.

If we want to add a smoothed index to this object, we use the smoothIndex() function. In its default state, this function just return a simple ts (time-series) object of the smoothed index.

To add it directly to the hpiindex object, we add the “in_place = TRUE” argument which will add it to the $smooth slot in the hpiindex object. Note that the class of temp_index stays the same (hpiindex), only the $smooth slot is added with an object of class smoothindex (inheriting from ts).

If you want to store plots of hpiindex objects, they are returned as plotindex objects, inheriting from gg and ggplot (from the ggplot2 package)

Wrapper

The index level wrappers (rtIndex() and hedIndex() – more to come) combine the above steps – data prep, modeling and index creation – into a single function, the output of which is an hpi collection, or simply, an hpi object. Directly after creation with a wrapper this object is a list of:

  1. A data object (hpidata)
  2. A model object (hpimodel)
  3. An index object (hpiindex)

Analyzing Indexes

Volatility

Index volatility can be calculated by providing an hpiindex object to the calcVolatility() function. The result is an object of class indexvolatility, containing three vector objects as well as two attributes indicating the original index and the window used.

The indexvolatility object can be nested within an hpiindex object by setting in_place = TRUE.

Volatility plots can be saved in objects of class plotvolatility, inheriting from gg and ggplot

Accuracy (Fitting)

Calculating an index’s (fitting) accuracy through either insample or kfold (out of sample) methods via calcAccuracy() returns an object of class hpiaccuracy, inheriting from data.frame. These object contain four fields, shown below. In hpiaccuracy objects, each observation is that is being tested for accuracy is only found once. This differs from the seriesaccuracy objects which are discussed later in which an observation may be tested against an index for each index in the series and therefore may appear multiple times in the seriesaccuracy object.

hpiaccuracy objects can be added directly to an hpi object with the in_place = TRUE argument.2

An hpiaccuracy object can be plotted and saved to a plotaccuracy object, but need the return_plot=TRUE argument to return to an object.

Series

A series of hpi objects are created with createSeries(). This results in a serieshpi object, with two initial slots: $data (hpidata object) and $hpis a list of hpi objects (minus the data component which has been removed to avoid duplication.)

Series can be plotted and saved into a plotseries object.

Revision

The revision of a series is calculated with calcRevision() and returns an object of class seriesrevision. This object contains a data.frame of revision by period and median and mean summaries of overall revision levels.

To add the revision to the serieshpi object, use in_place = TRUE.

Plotting revisions can be saved to a plotrevision object.

Series Accuracy

Finally, accuracy can be calculated at the series level by providing a serieshpi object to calcSeriesAccuracy(), resulting in a seriesaccuracy object. seriesaccuracy objects look nearly identical to hpiaccuracy objects, however they differ in two respects: 1) they may include estimate of accuracy for the same observation multiple times 3; 2) they will have a series field, indicating from which index in the series the accuracy estimate is derived.

Using test_method = ‘forecast’ will create a true, prediction-type test of accuracy (out of sample and into the future), where kfold is out of sample (but with knowledge of the future) and insample is just a simple fitting measure.

This object can be added directly to the serieshpi object by using the in_place = TRUE argument.

Finally, seriesaccuracy objects can be saved to a plot of class plotaccuracy

For more information on using the above functions please see the “Introduction” Vignette also provided with this package.


  1. Additional approaches will be added here in the future.

  2. Accuracy results are added to hpi objects and not hpindex object because they require data that is outside of the hpiindex object and therefore relate to the hpi collection more than just the index.

  3. This will not happen only when using a test_method of type “forecast” and a “forecast_length” of 1.