[ Search |  Up Level |  Project home |  Index |  Class hierarchy ]

 Data.ox

Overview

The Data class and its derivatives are used to simulate DDP data, compute predictions, read external data sets, and interface with Five-O to evaluate the likelihood or GMM objective.

Data objects store realizations along the path of the program. Unlike other spaces (e.g. the group space \(\Gamma\) and the state space \(\Theta\)), multiple instances of different types of data can be stored simultaneously. That is, there can be more than one data space.

There two parallel kinds of data:

Data							The basis of simulation, prediction, and data.
	Outcome						Data generated from following a realized DP and estimation based on maximum likelihood.
	Prediction						Data based on average or expected results over multiple DP realizations and estimation based on GMM.

However, users of niqlow will typically not interact directly with these base classes. Instead, you will work with panels of data, both external data read in for estimation and data generated from the model through simulation and prediction. If Outcome and Prediction are the lowest level classes, the corresponding highest ones are OutcomeDataSet and PredictionDataSet.

Contents

1. Outcomes
Simulation
Likelihood
2. Prediction
Tracking Objects to Predict
Moments
GMM

Data

1. Outcomes
1. Overview

Outcomes are used for simulating the model and for storing external data interpreted as realizations of the DDP. An outcome is generically denoted \(Y\). \(Y\) contains the realized state vectors, action vectors, auxiliary values, etc. It is described in more detail below.

As discussed elsewhere, the user codes MyModel as a class derived from one of the pre-defined Bellman classes. In contrast, Outcome is designed to be generic so that it can be used for whatever model or data the program uses. A user does not define their own derived class to store outcomes. Only one model can be active in DDP because static data members are used, but multiple data sets can be stored at once. This can be useful when comparing two solution methods applied to one model or two estimation methods applied to two different data sets.

Outcomes are related to points in the state space \(\Theta\), but they are treated distinctly in niqlow. The state space is coded to minimize storage and calculations while solving a model. On the other hand, outcomes are designed to track everything relevant to results of the DDP, and memory would be exhausted if outcomes were defined and stored for each possible state. It would be inefficient to store all potential outcomes when only one can be realized at each point in a single DP process.

A single point in the state space \(\theta\) can have multiple states that could come before it or come after it hypothetically. In contrast, outcomes have a well-defined predecessor and successor. These are tracked as part of the outcome, so that niqlow represents a sequence of outcomes as a doubly-linked list of objects of class Outcome.

An Outcome \(Y\) is a basic unit of data. However, a user will typically not interact with an individual outcome. Data structures are defined to keep track of ordered and unordered groups of outcomes. It is these objects that a user will typically create and use directly. The structure of outcomes is illustrated here:

		Outcome							A single realization of a discrete DP, denoted \(Y\).
			Path						A derived Outcome that points to a sequence of outcomes along a single realized DP path. As an outcome itself, the path is also the first outcome along itself: PATH::Y0 ⇆ Y1 ⇆   …  YT  ⇥ The outcomes on a path are ordered and they "point" to both their predecessor and successor. (In computer speak, a Path is a doubly-linked list.) This is because some processing of paths go forward (such as simulating outcomes) and some backwards (computing likelihood when accounting for unobserved states).
				FPanel					A Path that can point to more paths that all share the same fixed effect value, \(\gamma_f\). As a Path itself, the FPanel is the first path in the panel. FPANEL::PATH0 ↦ PATH1 ↦   …  PATHN  ⇥ The different paths in a FPANEL point to their successor only, because it is typically not necessary to process them in both directions.
					Panel				A heterogenous panel which contains one or more fixed panel. As an object derived from FPanel it also points to more FPanels. The Panel is the first fixed panel in the list. PANEL::FPANEL0 ↦ FPANEL1 ↦   …  FPANELM  ⇥ Different FPANELs in a PANEL point to their successor only, because it is typically not necessary to process them in both directions.
						OutcomeDataSet			A Panel with data-handling tools. Unlike the other derived classes, an OutcomeDataSet does not point to other panels (but the user's code could create more than one data set.) OUTCOMEDATASET::PANEL

2. Examples
To create a data set from external data:

	data = new OutcomeDataSet("DPdata");
	data->ObservedWithLabel(x,d);
	data->IDColumn("id");
	data->Read("data.dta");	

This says that there are two variables from the DP model observed in the data set, say a state variable x and an action variable d. The external data uses the same labels for those variables as the DP model, so the correct columns can be found using them. Each empirical path has to be associated with the value of variable in the data set, so the IDColumn must be specified before reading the data. The data must be sorted within each path starting at \(t=0\). Since no columns in the data were associated with fixed effect variables this must be a homogeneous model. (There can be random effects but niqlow always assumes those are unobserved and will integrate over them.) So the data set will end up having one Panel which contains one FPanel which contains as many paths as Read() finds in the Stata data set.

To create a simulated data set from a solved DP model and save to a Stata data set:

    SimulateOutcomes(400,10,"sim.dta");

This will create a Panel object, which in turn will automatically create a FPanel object for each value of the fixed effects, i.e. each point in the fixed part of \(\Gamma\). It will produce 400 paths, each 10 periods long, for each fixed group. Then it will save the data to the Stata dta file using the labels set for all the DP variables to label the columns. This could then be read back into a OutcomeDataSet during a Monte Carlo exercise. This function is equivalent to these lower-level commands:

    sim = new Panel("DPdata");
    sim -> Simulate(400,10);
	sim -> Print("sim.dta");		
    delete sim;

The reason to create the Panel object instead of using SimulateOutcomes is to keep the simulated data around and to be able to control other aspects of the data produced using methods defined for panels.
3. The Full Outcome

DDP is designed to handle programs for which not everything observed by the agent is observed by the econometrician. In addition, DDP is designed to handle observables that are functions of the underlying outcomes but are not necessarily of any (direct) interest of the agent. To provide these capabilities, first consider the case of the fully observed realization of the DP.

Let the Y^F denote the vector of all actions and all realized states in the model. Generically,

YF ≡ (α ζ ε η θ γ χ YFp YFn )
    α = (a0 … )
    ζ = (z0 … )
    ε = (e0 … )
    η = (h0 … )
    θ = (q0 … )
    γ = (g0 … )
    χ = (x0 … )

For simplicity the length of each of the vectors is not specified and the right-most elements of each of the vectors is not included. Instead, each vector is shown as the leftmost variable with subscript 0 followed by ….

Y^F is referred to as the full outcome because it includes all information that is knowable and relevant about a single period in a realized DDP. Y^F adds a new vector denoted χ with generic elements x. See below. Y^F also includes pointers to the preceding and next outcomes on the path. These pointers can be undefined for various reasons. Obviously the value of Y^F_n is not realized until the end of the current period. And the initial period of decision making obviously has no preceding outcome. However, in the middle of a panel of observation both pointers would be defined.

The full outcome may not be observed in data. The observed outcome, Y, will have the same structure as Y^F but one or more elements of the member vectors will include missing (.NaN) values. Observability is discussed below. Its purpose is to account for gaps between theoretical outcomes and observed results from a DP.

4. Paths

The base Outcome class captures a single period, and it points to the previous outcome and the next outcome along a realized path (if the predecessor and successor exist). A Path is derive from Outcome, so the first outcome on the path is stored directly by the path (since it is an outcome itself). But Path also has a last data member which points to the last outcome on the path. Because both the previous and next outcomes are defined a Path can be processed both forwards and backwards. That is, a path is a doubly-linked list.

A Path is a derived class from Outcome. The difference is that a path tracks a sequence of outcomes generated by realizations of a single DP problem. The outcomes in a path are a linked list. Since a path is itself an outcome the first outcome is the path itself. Outcome tracks both the next and previous outcome in the realized path. Further, the path stores the last outcome in the path, which makes it possible to work either forwards or backwards in the sequence.

5. Panels

In turn, a fixed panel (FPanel) is a list of Paths that share the same fixed effects. FPanel tracks the following path so a panel can be processed sequentially as a linked list. (Because paths in a panel are not ordered the previous path is not stored.) In turn a Panel is a list of fixed panels.

Finally, OutcomeDataSet is a class derived from Panel that includes methods to read in outcomes from data accounting for partial or complete unobservability of actions and states. This organization can be seen by clicking on hierarchy in the upper right corner of any page of DDP documentation.

6. Simulation

Simulation is a method of each level of Outcome up to and including a Panel. A simulation uses the specified initial conditions, the endogenous conditional choice probabilities, and the transitions of all states to produce simulated realizations of the DP. The simulate panel could then be loaded as data set to conduct a Monte Carlo estimation procedure.

Initial Conditions
Fixed states, γ
Endogenous states, θ
Sampling from the stationary distribution
7. Observed and Partially Observed Outcomes

Recall that Y^F is the full realized outcome of the DP model is a single period. It includes all states, actions and auxiliary variables. In practice the full outcome is not directly observed. The limited information can be variables whose realized values are simply missing, or they only affect an auxiliary variable which combines other elements of the outcome. We can express the observable outcome as

Y = ≡ (x0 … xχ.N  - - … ao0 - … - ao1 - … - zo0 - … - eo0 - … - ho0 - … - qo0 … - go0 - … )

Here - indicates censoring of elements of Y^F, and a_o0 is the first observed action variable. So if o0 = 3 it means a₃ is the first element of α that is in the observed outcome. Between a_o0 and a_o1 are any action variables not observed and indicated by - … -. The same is true of the state vectors.

This is not necessarily a template that all observations must follow. It is possible that some elements of Y^F are observed in some cases but not others. The auxiliary variables provide great flexibility. Suppose the value of an action variable a_d is unobserved, but it is known the realized value is with some sub-range of the potential values. Then x₀ and x₁ could be the min and max values of a_d, respectively. A derived likelihood function would add up the conditional probabilities of actions between those two values.

8. Data Sets

The OutcomeDataSet class is derived from Panel. It adds to the notion of a (heterogeneous) panel methods to read in external data and to account for observability and unobservability of aspects of the DP.

Observability

Getting Data

Data are read in from outside formats such as Stata using Ox's built-in capabilities in Ox's Database class. If the same labels are used for model variables as in the external data set, the methods can seamlessly relate the data to the corresponding element of the model.

A key issue with estimating DDPs is unobservability of actions and states. Outcome handles this two ways. First it treats missing (.NaN) in the data as unobserved. It will then integrate (sum over) values of missing states and actions when computing, say, the likelihood of the data. Second, auxiliary data (sometimes called payoff relevant variables) can be included in the data and then coded so that they indicate which variables and values should be integrated over.

Likelihood

2. Prediction
1. Overview

The Outcome class is designed to work with realized results from a DP model. That is, at one point in time there is a single realization. For this reason, simulation, individual-level data and likelihood functions are related to Outcomes. On the other hand, the Prediction class is designed to handle the distribution of outcomes at a given time. It is designed to handle integrating out uncertainty. For this reason, prediction and generalized method of moments (GMM) are related to Predictions.

The base class Prediction tracks a list of possible state indices and their associated probabilities of occurring. It also tracks choice probabilities. Starting from some initial conditions a PathPrediction is a list of predictions, each providing the distribution across states and choices at that point in time. A PanelPrediction is a set of PathPredictions across fixed effect groups. Data can be read in to match up with predictions using the PredictionDataSet class derived from PanelPrediction.

		Prediction							The expected outcome from a single realization. This expectation will typically integrate over the choice-specific additive shock \(\zeta\) and permanent unobserved heterogeneity \(\gamma_f\). Further, based on information from the user's code the prediction will integrate over missing elements of the current full outcome YF. It would also integrate over the distribution of previous states that could have potentially be the realized previous state.
			PathPrediction						A derived Prediction that points to a sequence of predictions along a single realized DP path. As a prediction, the path itself is the first prediction along the path. Since random effects \(\gamma_r\) are assumed to be unobserved, the path prediction integrates over any random effect groups that are in the model. There is only one prediction for a given fixed effect group \(\gamma_F\), so there is no prediction corresponds to the FPanel class derived from Outcomes. Each PathPrediction is in effect a FixedPathPrediction
				PanelPrediction					A list heterogenous path predictions. It is a PathPrediction which can also point to more PathPredictions. As a PathPrediction itself, the PanelPrediction is the first path in the panel.
					PredictionDataSet				A panel prediction with data-handling tools. Unlike the other derived classes, empirical moments do not point to other panel predictions.

2. Moments
3. GMM

Author:

© 2011-2023 Christopher Ferrall

---

Documentation of Items Defined in Data.ox  ⇧ 

 Global functions

Functions
ComputePredictions		Simple Prediction .
SimulateOutcomes		Simple Panel Simulation.

 DataColumn : Zauxiliary

Public fields
force0	const
incol
ind
label
obj	const
obsv
type	const

 ErgodicOutcomeDataSet : OutcomeDataSet : Panel : FPanel : Path : Outcome : Data : Task : DP

Public methods
ErgodicOutcomeDataSet
SemiClosedForm

Inherited methods from OutcomeDataSet:

EconometricObjective, freqColumn, IDColumn, MatchToColumn, ObservedWithLabel, OutcomeDataSet, Read, Simulate, Summary, tColumn, UnObserved

Inherited methods from Panel:

Collapse, Deep, Flat, LogLikelihood, Panel, Print, SetMethod

Inherited methods from FPanel:

FPanel, GetCur

Inherited methods from Path:

FullLikelihood, Likelihood, PartialObservedLikelihood, Path, PathObjective, ReadObs

Inherited methods from Outcome:

AuxLikelihood, CCLikelihood, FromSim, IIDLikelihood, Outcome

Inherited methods from Data:

SetLog

Inherited methods from Task:

SyncStates, ~Task

Inherited methods from DP:

Actions, AuxiliaryOutcomes, CreateSpaces, DrawFsamp, DrawGroup, DrawOneExogenous, EndogenousStates, ExogenousStates, GetAind, GetPinf, GetPstar, GetTrackTrans, GetUseEps, GroupVariables, Indicators, Initialize, Interactions, KLaggedAction, KLaggedState, MakeGroups, MultiInteractions, onlyDryRun, RecomputeTrans, SemiExogenousStates, SetClock, SetDelta, Settheta, SetUpdateTime, SetVersion, StorePalpha, SubSampleStates, SyncAct, UpdateDistribution, ValuesCounters

Inherited fields from OutcomeDataSet:

dlabels, freqcol, HasFrequencies, ids, label, list, low, LTypes, masked, source

Inherited fields from Panel:

first, flat, Fmtflat, FN, FNT, fparray, LFlat, M, r

Inherited fields from FPanel:

f, FPL, method, N, NT, SD, summand, upddens

Inherited fields from Path:

freq, i, L, LType, rcur, T

Inherited fields from Outcome:

act, Ainds, arows, aux, flnlike, mask, onext, OnlyTransitions, prev, snext, t, viinds, vilikes

Inherited fields from Data:

logf, lognm, Volume

Inherited fields from Task:

caller, d, done, inner, itask, iter, left, MaxTrips, right, state, subspace, trace, trips

Inherited fields from DP:

Blocks, Chi, ClockType, counter, curREdensity, delta, EOoE, EStoS, ETT, gdist, IOE, MyVersion, NKptrans, NKvindex, NxtExog, parents, S, SS, States, SubVectors, tod, userState, XUT

 ExogAuxOut : ExTask : Task : DP

Public fields
auxlike	static
outcm	static

Inherited methods from Task:

list, SyncStates, ~Task

Inherited methods from DP:

Actions, AuxiliaryOutcomes, CreateSpaces, DrawFsamp, DrawGroup, DrawOneExogenous, EndogenousStates, ExogenousStates, GetAind, GetPinf, GetPstar, GetTrackTrans, GetUseEps, GroupVariables, Indicators, Initialize, Interactions, KLaggedAction, KLaggedState, MakeGroups, MultiInteractions, onlyDryRun, RecomputeTrans, SemiExogenousStates, SetClock, SetDelta, Settheta, SetUpdateTime, SetVersion, StorePalpha, SubSampleStates, SyncAct, UpdateDistribution, ValuesCounters

Inherited fields from Task:

caller, d, done, inner, itask, iter, L, left, MaxTrips, right, state, subspace, trace, trips

Inherited fields from DP:

Blocks, Chi, ClockType, counter, curREdensity, delta, EOoE, EStoS, ETT, gdist, IOE, logf, lognm, MyVersion, NKptrans, NKvindex, NxtExog, parents, S, SS, States, SubVectors, tod, tom, userState, Volume, XUT

 FPanel : Path : Outcome : Data : Task : DP

Public fields
f	const	index of Fpanel in a panel.
FPL		fixed panel likelihood vector.
method		method to call for nested solution.
N		Number of paths in the panel.
NT		Total Number of Outcomes in the panel.
SD	const
summand	static
upddens	const
Public methods
Collapse	virtual
Deep	virtual	.
Flat	virtual	Return the fixed panel as a flat matrix.
FPanel		Store a panel of realized paths with common fixed group.
GetCur
LogLikelihood		Compute the vector log-likelihood for paths in the fixed (homogeneous) panel.
Simulate	virtual	Simulate a homogenous panel (fpanel) of paths.

Inherited methods from Path:

FullLikelihood, Likelihood, PartialObservedLikelihood, Path, PathObjective, ReadObs

Inherited methods from Outcome:

AuxLikelihood, CCLikelihood, FromSim, IIDLikelihood, Outcome

Inherited methods from Data:

SetLog

Inherited methods from Task:

list, SyncStates, ~Task

Inherited methods from DP:

Actions, AuxiliaryOutcomes, CreateSpaces, DrawFsamp, DrawGroup, DrawOneExogenous, EndogenousStates, ExogenousStates, GetAind, GetPinf, GetPstar, GetTrackTrans, GetUseEps, GroupVariables, Indicators, Initialize, Interactions, KLaggedAction, KLaggedState, MakeGroups, MultiInteractions, onlyDryRun, RecomputeTrans, SemiExogenousStates, SetClock, SetDelta, Settheta, SetUpdateTime, SetVersion, StorePalpha, SubSampleStates, SyncAct, UpdateDistribution, ValuesCounters

Inherited fields from Path:

flat, freq, i, L, LType, rcur, T

Inherited fields from Outcome:

act, Ainds, arows, aux, flnlike, mask, onext, OnlyTransitions, prev, snext, t, viinds, vilikes

Inherited fields from Data:

logf, lognm, Volume

Inherited fields from Task:

caller, d, done, inner, itask, iter, left, MaxTrips, right, state, subspace, trace, trips

Inherited fields from DP:

Blocks, Chi, ClockType, counter, curREdensity, delta, EOoE, EStoS, ETT, gdist, IOE, MyVersion, NKptrans, NKvindex, NxtExog, parents, S, SS, States, SubVectors, tod, userState, XUT

 Outcome : Data : Task : DP

Public fields
act		\(\alpha\)
Ainds		list of feasible sets consistent w/ data.
arows	static	current likelihood A rows .
aux		auxiliary values.
flnlike	static	current log-likelihood when collapsed
mask	static
onext		pointer to next outcome on the path
OnlyTransitions	static	do not include choice prob for fully observed likelihood, for first stage estimation of transitions.
prev	const	previous outcome.
snext		index of next simulated state
t	const	order on the path .
viinds	static	consistent states now & tom
vilikes	static	contigent likelihood now & tom.
Public methods
AuxLikelihood
CCLikelihood		Compute likelihood of choices and transitions this period assuming full state and action observability.
Deep	virtual	Print the outcome as record.
Flat	virtual	Return the outcome as a (flat) row vector.
FromSim
IIDLikelihood		Compute likelihood of an outcome given observablity of θ and η but integrating over ε $$L(\theta) = $$
Likelihood		Compute likelihood based on the Type.
Outcome		Record everything about a single realization of the DP.
PartialObservedLikelihood		Compute conditional forward likelihood of an outcome.
Simulate	virtual	Simulate the IID stochastic elements of a realization.

Inherited methods from Data:

SetLog

Inherited methods from Task:

list, SyncStates, ~Task

Inherited methods from DP:

Actions, AuxiliaryOutcomes, CreateSpaces, DrawFsamp, DrawGroup, DrawOneExogenous, EndogenousStates, ExogenousStates, GetAind, GetPinf, GetPstar, GetTrackTrans, GetUseEps, GroupVariables, Indicators, Initialize, Interactions, KLaggedAction, KLaggedState, MakeGroups, MultiInteractions, onlyDryRun, RecomputeTrans, SemiExogenousStates, SetClock, SetDelta, Settheta, SetUpdateTime, SetVersion, StorePalpha, SubSampleStates, SyncAct, UpdateDistribution, ValuesCounters

Inherited fields from Data:

logf, lognm, Volume

Inherited fields from Task:

caller, d, done, inner, itask, iter, L, left, MaxTrips, right, state, subspace, trace, trips

Inherited fields from DP:

Blocks, Chi, ClockType, counter, curREdensity, delta, EOoE, EStoS, ETT, gdist, IOE, MyVersion, NKptrans, NKvindex, NxtExog, parents, S, SS, States, SubVectors, tod, userState, XUT

 OutcomeDataSet : Panel : FPanel : Path : Outcome : Data : Task : DP

Example:

d = new OutcomeDataSet("d");

See DataObjective.

Public fields
dlabels		labels
freqcol
HasFrequencies
ids
label	const	Label for the data set.
list
low	const
LTypes
masked
source
Public methods
EconometricObjective	virtual	The default econometric objective: log-likelihood.
freqColumn		set the column label or index of a frequency (weighted data) variable.
IDColumn		set the column label or index of the observation ID.
MatchToColumn		Identify a variable with a data column.
ObservedWithLabel		Mark actions and state variables as observed in data, matched with their internal label.
OutcomeDataSet		Store a Panel as a data set.
Read		Load outcomes into the data set from a (long format) file or an Ox database.
Simulate		Simulate a data .
Summary		Produce a Stata-like summary statistics table.
tColumn		set the column label or index of the time value.
UnObserved		UnMark action and states variables as observed.

Inherited methods from Panel:

Collapse, Deep, Flat, LogLikelihood, Panel, Print, SetMethod

Inherited methods from FPanel:

FPanel, GetCur

Inherited methods from Path:

FullLikelihood, Likelihood, PartialObservedLikelihood, Path, PathObjective, ReadObs

Inherited methods from Outcome:

AuxLikelihood, CCLikelihood, FromSim, IIDLikelihood, Outcome

Inherited methods from Data:

SetLog

Inherited methods from Task:

SyncStates, ~Task

Inherited methods from DP:

Actions, AuxiliaryOutcomes, CreateSpaces, DrawFsamp, DrawGroup, DrawOneExogenous, EndogenousStates, ExogenousStates, GetAind, GetPinf, GetPstar, GetTrackTrans, GetUseEps, GroupVariables, Indicators, Initialize, Interactions, KLaggedAction, KLaggedState, MakeGroups, MultiInteractions, onlyDryRun, RecomputeTrans, SemiExogenousStates, SetClock, SetDelta, Settheta, SetUpdateTime, SetVersion, StorePalpha, SubSampleStates, SyncAct, UpdateDistribution, ValuesCounters

Inherited fields from Panel:

first, flat, Fmtflat, FN, FNT, fparray, LFlat, M, r

Inherited fields from FPanel:

f, FPL, method, N, NT, SD, summand, upddens

Inherited fields from Path:

freq, i, L, LType, rcur, T

Inherited fields from Outcome:

act, Ainds, arows, aux, flnlike, mask, onext, OnlyTransitions, prev, snext, t, viinds, vilikes

Inherited fields from Data:

logf, lognm, Volume

Inherited fields from Task:

caller, d, done, inner, itask, iter, left, MaxTrips, right, state, subspace, trace, trips

Inherited fields from DP:

Blocks, Chi, ClockType, counter, curREdensity, delta, EOoE, EStoS, ETT, gdist, IOE, MyVersion, NKptrans, NKvindex, NxtExog, parents, S, SS, States, SubVectors, tod, userState, XUT

 Panel : FPanel : Path : Outcome : Data : Task : DP

Public fields
first		.
flat		matrix representation of panel.
Fmtflat	static	.
FN		total paths.
FNT		total outcomes in the panel.
fparray
LFlat	static	column labels in flat.
M		panel likelihood vector.
r	const	tag for the panel.
Public methods
Collapse	virtual
Deep		Print the deep view of the panel.
Flat		Store the panel as long flat matrix.
LogLikelihood		Compute the vector of log-likelihoods.
Panel		Store a list of heterogenous fixed panels.
Print		Produce a matrix of the panel.
SetMethod		Set the nested DP solution method to use when evaluating the panel's econometric objective.
Simulate		Simulate a (heterogeneous) panel.

Inherited methods from FPanel:

FPanel, GetCur

Inherited methods from Path:

FullLikelihood, Likelihood, PartialObservedLikelihood, Path, PathObjective, ReadObs

Inherited methods from Outcome:

AuxLikelihood, CCLikelihood, FromSim, IIDLikelihood, Outcome

Inherited methods from Data:

SetLog

Inherited methods from Task:

list, SyncStates, ~Task

Inherited methods from DP:

Actions, AuxiliaryOutcomes, CreateSpaces, DrawFsamp, DrawGroup, DrawOneExogenous, EndogenousStates, ExogenousStates, GetAind, GetPinf, GetPstar, GetTrackTrans, GetUseEps, GroupVariables, Indicators, Initialize, Interactions, KLaggedAction, KLaggedState, MakeGroups, MultiInteractions, onlyDryRun, RecomputeTrans, SemiExogenousStates, SetClock, SetDelta, Settheta, SetUpdateTime, SetVersion, StorePalpha, SubSampleStates, SyncAct, UpdateDistribution, ValuesCounters

Inherited fields from FPanel:

f, FPL, method, N, NT, SD, summand, upddens

Inherited fields from Path:

freq, i, L, LType, rcur, T

Inherited fields from Outcome:

act, Ainds, arows, aux, flnlike, mask, onext, OnlyTransitions, prev, snext, t, viinds, vilikes

Inherited fields from Data:

logf, lognm, Volume

Inherited fields from Task:

caller, d, done, inner, itask, iter, left, MaxTrips, right, state, subspace, trace, trips

Inherited fields from DP:

Blocks, Chi, ClockType, counter, curREdensity, delta, EOoE, EStoS, ETT, gdist, IOE, MyVersion, NKptrans, NKvindex, NxtExog, parents, S, SS, States, SubVectors, tod, userState, XUT

 PanelPrediction : PathPrediction : Prediction : Data : Task : DP

Public fields
aflat		flat matrix version of predictions.
cols		columns in data .
delt		difference between pred.
EverPredicted		Predict() called before.
first	const	either fparray[0] or this.
FN		total number of predictions..
fparray	const	array pointing to (fixed) path predictions.
M		array of GMM vector.
mask		indicator vector for observed moments.
pcount		# of predictions made and saved so far .
PredMomFile	static	file name of the last Panel Prediction Data Set saved.
summand	const	object to integrate over \(\gamma_r\).
tlabels		labels of flat print.
tlist		list of objects to track.
TrackingCalled		Has Tracking() been called.
upddens	const	object to update distribution over r.
Public methods
InitializePath
PanelPrediction		Create a panel of predictions.
Predict		Predict outcomes in the panel.
SetColumns		Set up data columns for tracked variables.
Tracking		Add objects to track mean values over the path.

Inherited methods from PathPrediction:

AppendSimulated, Empirical, GetFlat, PathPrediction, Qcols, SetFlat, SimulateOutcomePaths, tprefix

Inherited methods from Prediction:

GetAcc, Histogram, IncAcc, Prediction, Reset, SetMoms

Inherited methods from Data:

SetLog

Inherited methods from Task:

list, SyncStates, ~Task

Inherited methods from DP:

Actions, AuxiliaryOutcomes, CreateSpaces, DrawFsamp, DrawGroup, DrawOneExogenous, EndogenousStates, ExogenousStates, GetAind, GetPinf, GetPstar, GetTrackTrans, GetUseEps, GroupVariables, Indicators, Interactions, KLaggedAction, KLaggedState, MakeGroups, MultiInteractions, onlyDryRun, RecomputeTrans, SemiExogenousStates, SetClock, SetDelta, Settheta, SetUpdateTime, SetVersion, StorePalpha, SubSampleStates, SyncAct, UpdateDistribution, ValuesCounters

Inherited fields from PathPrediction:

aggexists, cur, dlabels, f, Fcols, firstprediction, flat, fnext, fvals, HasObservations, iDist, inT, L, MakeFlat, method, mother, myshare, pathW, plabels, prtlevel, pstate, rcur, T, vdelt, wght

Inherited fields from Prediction:

accmom, ch, chq, empmom, haz, p, pnext, pq, predmom, q, readmom, sind, t, W

Inherited fields from Data:

logf, lognm, Volume

Inherited fields from Task:

caller, d, done, inner, itask, iter, left, MaxTrips, right, state, subspace, trace, trips

Inherited fields from DP:

Blocks, Chi, ClockType, counter, curREdensity, delta, EOoE, EStoS, ETT, gdist, IOE, MyVersion, NKptrans, NKvindex, NxtExog, parents, S, SS, States, SubVectors, tod, tom, userState, XUT

 Path : Outcome : Data : Task : DP

Public fields
flat		.
freq		frequency of outcomes represented by this.
i		index of path in a panel.
L		likelihood of the path.
LType		type of likelihood calculation.
rcur		current index of random effects.
T		lenth of the path.
Public methods
Collapse	virtual
Deep	virtual	Produce a two dimensional view of the path;
Flat	virtual	Produce a matrix representation of the path.
FullLikelihood
Likelihood		Compute likelihood of a realized path.
PartialObservedLikelihood		Compute conditional forward likelihood of an outcome.
Path		Create a new series of Outcomes along a realized path.
PathObjective
ReadObs
Simulate	virtual	Simulate a list of realized states and actions from an initial state.

Inherited methods from Outcome:

AuxLikelihood, CCLikelihood, FromSim, IIDLikelihood, Outcome

Inherited methods from Data:

SetLog

Inherited methods from Task:

list, SyncStates, ~Task

Inherited methods from DP:

Actions, AuxiliaryOutcomes, CreateSpaces, DrawFsamp, DrawGroup, DrawOneExogenous, EndogenousStates, ExogenousStates, GetAind, GetPinf, GetPstar, GetTrackTrans, GetUseEps, GroupVariables, Indicators, Initialize, Interactions, KLaggedAction, KLaggedState, MakeGroups, MultiInteractions, onlyDryRun, RecomputeTrans, SemiExogenousStates, SetClock, SetDelta, Settheta, SetUpdateTime, SetVersion, StorePalpha, SubSampleStates, SyncAct, UpdateDistribution, ValuesCounters

Inherited fields from Outcome:

act, Ainds, arows, aux, flnlike, mask, onext, OnlyTransitions, prev, snext, t, viinds, vilikes

Inherited fields from Data:

logf, lognm, Volume

Inherited fields from Task:

caller, d, done, inner, itask, iter, left, MaxTrips, right, state, subspace, trace, trips

Inherited fields from DP:

Blocks, Chi, ClockType, counter, curREdensity, delta, EOoE, EStoS, ETT, gdist, IOE, MyVersion, NKptrans, NKvindex, NxtExog, parents, S, SS, States, SubVectors, tod, userState, XUT

 PathPrediction : Prediction : Data : Task : DP

Public fields
aggexists	static
cur		the current prediction
dlabels		labels for vdelt.
f	const	fixed index.
Fcols	const
firstprediction		first prediction.
flat		flat prediction matrix.
fnext		the next PathPrediction
fvals	const	for tracking.
HasObservations		Empirical moments read in.
iDist	const	initial distribution.
inT		Path length sent it.
L		Distance between predictions and emp.mom.
MakeFlat		Create flat prediction always .
method		method to call for nested solution.
mother	const	panel I belong to.
myshare	const	My share of population.
pathW		Weighting matrix for GMM for full path.
plabels		labels for simulated path.
prtlevel		.
pstate	const	pstate .
rcur		current index of random effects.
T		length of the path.
vdelt		wide delta vector.
wght	const	Weight Moments for GMM if empirical moments include.
Public methods
AppendSimulated
Empirical		Add empirical values to a path of predicted values.
GetFlat		Get selected elements of the flat path prediction after it is made.
PathPrediction		Create a path of predictions - should only be called by PanelPrediction.
Predict		Create or update a path of predicted distributions.
Qcols
SetFlat
SimulateOutcomePaths		Compute and store path weighting matrix for fixed effect group.
tprefix	static

Inherited methods from Prediction:

GetAcc, Histogram, IncAcc, Prediction, Reset, SetMoms

Inherited methods from Data:

SetLog

Inherited methods from Task:

list, SyncStates, ~Task

Inherited methods from DP:

Actions, AuxiliaryOutcomes, CreateSpaces, DrawFsamp, DrawGroup, DrawOneExogenous, EndogenousStates, ExogenousStates, GetAind, GetPinf, GetPstar, GetTrackTrans, GetUseEps, GroupVariables, Indicators, Interactions, KLaggedAction, KLaggedState, MakeGroups, MultiInteractions, onlyDryRun, RecomputeTrans, SemiExogenousStates, SetClock, SetDelta, Settheta, SetUpdateTime, SetVersion, StorePalpha, SubSampleStates, SyncAct, UpdateDistribution, ValuesCounters

Inherited fields from Prediction:

accmom, ch, chq, empmom, haz, p, pnext, pq, predmom, q, readmom, sind, t, W

Inherited fields from Data:

logf, lognm, Volume

Inherited fields from Task:

caller, d, done, inner, itask, iter, left, MaxTrips, right, state, subspace, trace, trips

Inherited fields from DP:

Blocks, Chi, ClockType, counter, curREdensity, delta, EOoE, EStoS, ETT, gdist, IOE, MyVersion, NKptrans, NKvindex, NxtExog, parents, S, SS, States, SubVectors, tod, tom, userState, XUT

 Prediction : Data : Task : DP

Public fields
accmom		accumulated predicted moments across r
ch		Expanded ch.
chq		current ch.
empmom		used (masked) empiricalmoments.
haz		mass hazard at t.
p
pnext		next prediction on the path
pq		current p.
predmom		(unmasked) predicted moment vector
q		index into sind.
readmom		(unmasked) empirical moments.
sind		state index
t	const	rank in a pathprediction.
W		masked weight to put on distance.
Public methods
GetAcc
Histogram		Compute the histogram of tracked object at the prediction.
IncAcc
Predict		Compute the predicted distribution of actions and states.
Prediction		Create a new prediction.
Reset
SetMoms		Initialize and if necessary set moms vectors.

Inherited methods from Data:

SetLog

Inherited methods from Task:

list, SyncStates, ~Task

Inherited methods from DP:

Actions, AuxiliaryOutcomes, CreateSpaces, DrawFsamp, DrawGroup, DrawOneExogenous, EndogenousStates, ExogenousStates, GetAind, GetPinf, GetPstar, GetTrackTrans, GetUseEps, GroupVariables, Indicators, Initialize, Interactions, KLaggedAction, KLaggedState, MakeGroups, MultiInteractions, onlyDryRun, RecomputeTrans, SemiExogenousStates, SetClock, SetDelta, Settheta, SetUpdateTime, SetVersion, StorePalpha, SubSampleStates, SyncAct, UpdateDistribution, ValuesCounters

Inherited fields from Data:

logf, lognm, Volume

Inherited fields from Task:

caller, d, done, inner, itask, iter, L, left, MaxTrips, right, state, subspace, trace, trips

Inherited fields from DP:

Blocks, Chi, ClockType, counter, curREdensity, delta, EOoE, EStoS, ETT, gdist, IOE, MyVersion, NKptrans, NKvindex, NxtExog, parents, S, SS, States, SubVectors, tod, tom, userState, XUT

 PredictionDataSet : PanelPrediction : PathPrediction : Prediction : Data : Task : DP

Public fields
flist
FMethod
Nplace		observations column (index or label).
Tplace		time column (index or label).
UorCorL		matrix of indices or array of labels or UseLabel
Public methods
EconometricObjective	virtual	The default econometric objective for a panel prediction: the overall GMM objective.
Observations		Define columns where observations and time values appear.
Observed
PredictionDataSet		Create a panel prediction that is matched with external data.
Read		Read in external moments of tracked objects.
SimulateMomentVariances		Simulate sample of outcomes compute path Variance matrix and save inverse.
TrackingMatchToColumn		Track an object that is matched to column in the data.
TrackingWithLabel		Track one or more objects that are matched to columns using the object's label.

Inherited methods from PanelPrediction:

InitializePath, PanelPrediction, Predict, SetColumns, Tracking

Inherited methods from PathPrediction:

AppendSimulated, Empirical, GetFlat, PathPrediction, Qcols, SetFlat, SimulateOutcomePaths, tprefix

Inherited methods from Prediction:

GetAcc, Histogram, IncAcc, Prediction, Reset, SetMoms

Inherited methods from Data:

SetLog

Inherited methods from Task:

list, SyncStates, ~Task

Inherited methods from DP:

Actions, AuxiliaryOutcomes, CreateSpaces, DrawFsamp, DrawGroup, DrawOneExogenous, EndogenousStates, ExogenousStates, GetAind, GetPinf, GetPstar, GetTrackTrans, GetUseEps, GroupVariables, Indicators, Interactions, KLaggedAction, KLaggedState, MakeGroups, MultiInteractions, onlyDryRun, RecomputeTrans, SemiExogenousStates, SetClock, SetDelta, Settheta, SetUpdateTime, SetVersion, StorePalpha, SubSampleStates, SyncAct, UpdateDistribution, ValuesCounters

Inherited fields from PanelPrediction:

aflat, cols, delt, EverPredicted, first, FN, fparray, M, mask, pcount, PredMomFile, summand, tlabels, tlist, TrackingCalled, upddens

Inherited fields from PathPrediction:

aggexists, cur, dlabels, f, Fcols, firstprediction, flat, fnext, fvals, HasObservations, iDist, inT, L, MakeFlat, method, mother, myshare, pathW, plabels, prtlevel, pstate, rcur, T, vdelt, wght

Inherited fields from Prediction:

accmom, ch, chq, empmom, haz, p, pnext, pq, predmom, q, readmom, sind, t, W

Inherited fields from Data:

logf, lognm, Volume

Inherited fields from Task:

caller, d, done, inner, itask, iter, left, MaxTrips, right, state, subspace, trace, trips

Inherited fields from DP:

Blocks, Chi, ClockType, counter, curREdensity, delta, EOoE, EStoS, ETT, gdist, IOE, MyVersion, NKptrans, NKvindex, NxtExog, parents, S, SS, States, SubVectors, tod, tom, userState, XUT

 TrackObj

Public methods
print		Print mean and histogram of tracked object.

 Global

 ComputePredictions

Simple Prediction .

Parameters:

T	integer, length of panel UseDefault [default], length of lifecycle or 10 probabilities Two print predictions
prtlevel	Two [default] print predictions One print state and choice probabilities Zero do not print, instead save to prediction moment file This creates a PanelPrediction object, creates the prediction tracking all varaibles and prints out. Object is then deleted

Returns:

flat prediction

 SimulateOutcomes

Simple Panel Simulation.

Parameters:

Nsim	integer, number of paths to simulate per fixed group [default] UseDefault, whic is 1
T	integer, length of panel [default], length of lifecycle or 10
outopt	integer [default] print to screen or string name of file to save to
ErgOrStateMat	0 [default]: find lowest reachable indexed state to start from 1: draw from stationary distribution (must be ergodic) matrix of initial states to draw from (each column is a different starting value)
DropTerminal	TRUE: eliminate termainl states from the data set FALSE: [default] include terminal states. This routine simplifies simulating a solved DP model. Simply call it instead of creating an Panel object.

 DataColumn

 force0

 incol

 ind

 label

 obj

 obsv

 type

 ErgodicOutcomeDataSet

 ErgodicOutcomeDataSet

 SemiClosedForm

 ExogAuxOut

 auxlike

 outcm

 FPanel

 Collapse

 Deep

.

 f

index of Fpanel in a panel.

 Flat

Return the fixed panel as a flat matrix. index of panel

Returns:

long matrix of panels

 FPanel

Store a panel of realized paths with common fixed group.

Parameters:

f	integer tag for the panel (such as replication index) [default=0]
method	Method to call to solve 0 [default] do nothing, something else handles solution

 FPL

fixed panel likelihood vector.

 GetCur

 LogLikelihood

Compute the vector log-likelihood for paths in the fixed (homogeneous) panel. The vector of path log-likelihoods is stored in FPL.

If the method is a class

Solve() is called first.

If done equals IterationFailed the likelihood is not computed. FPL is set to a vector of .NaN.

 method

method to call for nested solution.

 N

Number of paths in the panel.

 NT

Total Number of Outcomes in the panel.

 SD

 Simulate

Simulate a homogenous panel (fpanel) of paths.

Parameters:

Nsim	> 0, number of paths to simulate
Tmax	maximum path length 0 no maximum length.
ErgOrStateMat	0 [default]: find lowest reachable indexed state to start from 1: draw from stationary distribution (must be ergodic) matrix of initial states to draw from (each column is a different starting value)
DropTerminal	TRUE: eliminate termainl states from the data set FALSE: [default] include terminal states.
pathpred	0 [default] or PathPrediction object to filter simulated values

Comments:

γ region of state is masked out.

 summand

 upddens

 Outcome

 act

\(\alpha\)

 Ainds

list of feasible sets consistent w/ data.

 arows

current likelihood A rows .

 aux

auxiliary values.

 AuxLikelihood

 CCLikelihood

Compute likelihood of choices and transitions this period assuming full state and action observability.

 Deep

Print the outcome as record.

 Flat

Return the outcome as a (flat) row vector.

Used to print or save a series or panel as a matrix.

Columns:

t ~ State_Ind ~ Type ~ Aind ~ ε ~ eta; ~ θ ~ γ ~ α

 flnlike

current log-likelihood when collapsed

 FromSim

 IIDLikelihood

Compute likelihood of an outcome given observablity of θ and η but integrating over ε

$$L(\theta) = $$

 Likelihood

Compute likelihood based on the Type.

 mask

 onext

pointer to next outcome on the path

 OnlyTransitions

do not include choice prob for fully observed likelihood, for first stage estimation of transitions.

 Outcome

Record everything about a single realization of the DP. This is not usually called by the user code. Outcomes are created along paths.

Parameters:

prior

Outcome object, the previous realization along the path integer, this is first realization on the path. state uninitialized. vector, initial value of state

 PartialObservedLikelihood

Compute conditional forward likelihood of an outcome.

$$L(\theta) = \sum_{\eta} \sum_{\alpha} P*() P()$$

 prev

previous outcome.

 Simulate

Simulate the IID stochastic elements of a realization. This is usually called along a path not by the user's code

θ and γ vectors already set. ε and η elements are simulated from their transitions. Then Simulate called to simulate &apha;, and Υ.

Returns:

TRUE if path is ended, FALSE otherwise

See also:

DrawOneExogenous, Simulate

 snext

index of next simulated state

 t

order on the path .

 viinds

consistent states now & tom

 vilikes

contigent likelihood now & tom.

 OutcomeDataSet

 dlabels

labels

 EconometricObjective

The default econometric objective: log-likelihood.

Parameters:

subp	DoAll (default), solve all subproblems and return likelihood vector Non-negative integer, solve only subproblem, return contribution to overall L

Returns:

M, lnL = (lnL₁ lnL₂ …)

See also:

LogLikelihood

 freqcol

 freqColumn

set the column label or index of a frequency (weighted data) variable.

Parameters:

lORind

string, column label integer≥0 column index; Likelihoods are multiplied by this value. If no column is specified the default is 1.

 HasFrequencies

 IDColumn

set the column label or index of the observation ID.

Parameters:

lORind

string, column label integer≥0 column index;

 ids

 label

Label for the data set.

 list

 low

 LTypes

 masked

 MatchToColumn

Identify a variable with a data column.

Parameters:

aORs	Either an ActionVariable, element of \(\alpha\), or a StateVariable, element of one of the state vectors, or a AuxiliaryValue, element of \(\chi\) OR
LorC	UseLabel, variable's label to denote column of data with observations integer ≥ 0, column of data matrix that contains observations string, label of column with observations.

 ObservedWithLabel

Mark actions and state variables as observed in data, matched with their internal label.

Parameters:

aORs	Either an ActionVariable, element of \(\alpha\), or a StateVariable, element of one of the state vectors, or a AuxiliaryValue, element of \(\chi\) OR array of the form {v1,v2,…}. In this case all other arguments are ignored.
...	continues with object2, LoC2, object3, LorC3, etc.

 OutcomeDataSet

Store a Panel as a data set.

Parameters:

label	string, tag for the data set UseDefault [default] user classname
method,	solution method to be used as data set is processed. 0 [default], no solution

 Read

Load outcomes into the data set from a (long format) file or an Ox database.

Parameters:

FNorDB	string, file name with extension that can be read by OX::Database::Load Database object
SearchLabels	TRUE: search data set labels and use any matches as observed.

Example:

  d = new OutcomeDataSet();
  d -> Read("data.dta");

 Simulate

Simulate a data . Each value of fixed \(\gamma\) is simulated N times, drawing the random effects in \(\gamma_r\) from their density.

Parameters:

N	integer number of paths to simulate in each FPanel.
T	Integer, max length of each path vector, max length for each FPanel.
ErgOrStateMat	0: find lowest reachable indexed state to start from 1: draw from stationary distribution (must be ergodic) matrix: initial states to draw from (each column is a different starting value)
DropTerminal	TRUE: eliminate termainl states from the data set
pathpred	Integer [default] or PathPrediction object that is simulating (used when computing efficient GMM weights)

 source

 Summary

Produce a Stata-like summary statistics table.

Parameters:

data	matrix, data to summarize integer, summarize flat
rlabels	[default=0], array of labels

 tColumn

set the column label or index of the time value.

Parameters:

lORind

string, column label integer≥0 column index;

 UnObserved

UnMark action and states variables as observed.

Parameters:

as1	Discrete object, either an ActionVariable, element of \(\alpha\), or a StateVariable, element of one of the state vectors StateBlock: each variable in the block will be marked unobserved.
...	as2, etc.

Comments:

Does nothing unless variable was already sent to ObservedWithLabel();

 Panel

 Collapse

 Deep

Print the deep view of the panel.

 first

.

 Flat

Store the panel as long flat matrix.

 flat

matrix representation of panel.

See also:

Flat

 Fmtflat

.

 FN

total paths.

 FNT

total outcomes in the panel.

 fparray

 LFlat

column labels in flat.

 LogLikelihood

Compute the vector of log-likelihoods. The vector of path log-likelihoods is stored in M, it is constructed by appending each FPL. If method is an object, then method->Solve() is called.

See also:

EconometricObjective

 M

panel likelihood vector.

 Panel

Store a list of heterogenous fixed panels.

Parameters:

r	integer tag for the panel (such as replication index)
method	Method object, the DP solution to call to solve FPanel problem. (default) 0 do nothing, something else handles solution

 Print

Produce a matrix of the panel. If flatis an uninitialized then Flat() is called first. Flat version of the data set is stored in flat.

Parameters:

fn	0: do not print or save, just return 1 print to data log file 2 print to screen string: save to a file
Orientation	LONG or WIDE

Returns:

long matrix of panels

 r

tag for the panel.

 SetMethod

Set the nested DP solution method to use when evaluating the panel's econometric objective.

Parameters:

method

Method

 Simulate

Simulate a (heterogeneous) panel. Each value of fixed γ is simulated N times, drawing the random effects in γ from their density.

Parameters:

N	integer number of paths to simulate in each FPanel.
T	Integer, max length of each path vector, max length for each FPanel.
ErgOrStateMat	0: find lowest reachable indexed state to start from 1: draw from stationary distribution (must be ergodic) matrix of initial states to draw from (each column is a different starting value)
DropTerminal	TRUE: eliminate termainl states from the data set
pathpred	Integer [default] or PathPrediction object that is simulating

 PanelPrediction

 aflat

flat matrix version of predictions.

 cols

columns in data .

 delt

difference between pred. & data.

 EverPredicted

Predict() called before.

 first

either fparray[0] or this.

 FN

total number of predictions..

 fparray

array pointing to (fixed) path predictions.

 InitializePath

 M

array of GMM vector.

 mask

indicator vector for observed moments.

 PanelPrediction

Create a panel of predictions.

Parameters:

label	name for the data UseDefault [default] use classname of model class.
method	Method to be called before predictions.
iDist	initial conditions for PathPredictions ErgodicDist : use computed stationary distribution in ergodic dist. 0 [default]: start the prediction at the lowest-indexed reachable state in Θ. non-negative integer: start at iDist and increment until a reachable state index is found. matrix: a list of states to start the prediction from, uniform distribution array: a list of state INDICES and a list of probabilities, can be the output of a IID Transit() object of Prediction class: use sind as the initial state for this prediction.
wght	[default=UNCORRELATED]
aggshares	0 [default] equal shares of averaged moments over fixed groups -1 [UnInitialized] do not compute averaged moments Fx1 vector, share of population

 pcount

# of predictions made and saved so far .

 Predict

Predict outcomes in the panel.

Parameters:

T	: positive integer or matrix of lengths of paths to predict (same length as number of paths in then panel)
prtlevel	: Zero [default] do not print One print state and choice probabilities Two print predictions
outmat	matrix, predictions already made, just process contributions

Returns:

succ TRUE no problems
FALSE prediction or solution failed.

 PredMomFile

file name of the last Panel Prediction Data Set saved.

 SetColumns

Set up data columns for tracked variables.

Parameters:

dlabels	array of column labels in the data.
Nplace	number of observations (row weight) column UnInitialized no row weights
tplace	model t column UnInitialized

 summand

object to integrate over \(\gamma_r\).

 tlabels

labels of flat print.

 tlist

list of objects to track.

 Tracking

Add objects to track mean values over the path.

Parameters:

LorC	UseLabel: use object label to match to column. NotInData unmatched to data. integer: column in data set string: column label TrackAll : track all actions, endogenous states and auxiliaries
...	Discrete objects and/or arrays or objects to track

Returns:

a row vector of the positions of the added objects that can be sent to GetFlat Thus if different sets of outcomes should be extracted together as vectors call Tracking() for each set and store the return value.

Comments:

This routine can be called more than once, but once Predict() has been called no more objects can be added to the list.

 TrackingCalled

Has Tracking() been called.

 upddens

object to update distribution over r.

 Path

 Collapse

 Deep

Produce a two dimensional view of the path;

 Flat

Produce a matrix representation of the path. Path id (i) is appended as the first column. Each row is an Outcome.

Returns:

TxM matrix

 flat

.

 freq

frequency of outcomes represented by this.

 FullLikelihood

 i

index of path in a panel.

 L

likelihood of the path.

 Likelihood

Compute likelihood of a realized path.

 LType

type of likelihood calculation.

 PartialObservedLikelihood

Compute conditional forward likelihood of an outcome.

$$L(\theta) = \sum_{\eta} \sum_{\alpha} P*() P()$$

 Path

Create a new series of Outcomes along a realized path.

Parameters:

id	integer, id or tag for the path.
state0	UnInitialized (-1), set state to uninitalized non-negative fixed effect index to use, draw and random effect from current distribution vector, initial state vector

 PathObjective

 rcur

current index of random effects.

 ReadObs

 Simulate

Simulate a list of realized states and actions from an initial state.

Checks to see if transition is Ρ is tracking. If not, process span the state space with EndogTrans.

Parameters:

newstate	UnInitialized (default) state already set state to add to group state
T	integer, max. length of the panel 0, no maximum lenth; simulation goes on until a Terminawl State is reached.
DropTerminal	drop states that are terminal

Example:

 T

lenth of the path.

 PathPrediction

 aggexists

 AppendSimulated

 cur

the current prediction

 dlabels

labels for vdelt.

 Empirical

Add empirical values to a path of predicted values.

Parameters:

inmom	Txm matrix of values.
hasN	FALSE: no row observation column TRUE: second-to-last column that contains observation count used for weighting of distances.
hasT	FALSE: no model t column TRUE: last column contains observation count
MaxT	path length will equal max{MaxT,maxc{inNandMom[][C]}}

Comments:

If T is greater than the current length of the path additional predictions are concatenated to the path

 f

fixed index.

 Fcols

 firstprediction

first prediction.

 flat

flat prediction matrix.

 fnext

the next PathPrediction

 fvals

for tracking.

 GetFlat

Get selected elements of the flat path prediction after it is made.

Parameters:

tvals	DoAll (default) return all time periods(rows) integer or vector of t indices to report
mvals	DoAll (default) return all moments (columns) integer or vector of moments to report

Returns:

flat[tvals][mvals]

 HasObservations

Empirical moments read in.

 iDist

initial distribution.

 inT

Path length sent it.

 L

Distance between predictions and emp.mom.

 MakeFlat

Create flat prediction always . Usually flat prediction matrix only created for data output. The flat prediction is used by classes such as Equilibrium. Then in this case set this to TRUE.

 method

method to call for nested solution.

 mother

panel I belong to.

 myshare

My share of population.

 PathPrediction

Create a path of predictions - should only be called by PanelPrediction.

Parameters:

mother	object of PanelPrediction that is the mother to this path.
f	integer: fixed group index [default=0] AlLFixed (-1): this aggregates (averages) predictions over NotInData (-2): no prediction stored here.
method	0: do not call a nested solution [default] a solution Method object to be called before making predictions
iDist	initial distribution. ErgodicDist : use computed stationary distribution in ergodic dist. non-negative integer: start at iDist and increment until a reachable state index is found. So PathPrediction(0) [default] will start the prediction at the lowest-indexed reachable state in Θ. matrix: a list of states to start the prediction from object of Prediction class: use sind as the initial state for this prediction.
wght	Code for weighting scheme for distance between empirical and predicted paths.
myshare	either 0 or a fraction of the population if an aggregate (overall) prediction is being stored. The prediction is not made until Predict() is called.

 pathW

Weighting matrix for GMM for full path.

 plabels

labels for simulated path.

 Predict

Create or update a path of predicted distributions.

Parameters:

inT	integer length of the path 0 (default) : predict only for existing predictions on the Path. If existing path is longer than inT (and inT > 0) then extra predictions are deleted.
prtlevel	Zero [default] do not print One print state and choice probabilities Two print predictions

Returns:

TRUE if everything succeeds
FALSE if either the solution method or the prediction fails

Predictions are averaged over random effect groups.

Example:

  p = new PanelPrediction();
  p->Predict(10);

 prtlevel

.

 pstate

pstate .

 Qcols

 rcur

current index of random effects.

 SetFlat

 SimulateOutcomePaths

Compute and store path weighting matrix for fixed effect group.

Parameters:

curfpanel	FPanel object to simulate
N	sample size
ErogOrStateMat	simulation initial conditions. Let SM be the matrix of simulated outcomes (concatenation over time) pathW = generalized inverse of Var(SM). Save simulated data to logs/flat_??.dta Save weight matrix in pathW_??.mat

 T

length of the path.

 tprefix

 vdelt

wide delta vector.

 wght

Weight Moments for GMM if empirical moments include.

See also:

GMMWeightOptions

 Prediction

 accmom

accumulated predicted moments across r

 ch

Expanded ch. prob.

 chq

current ch. prob.

 empmom

used (masked) empiricalmoments.

 GetAcc

 haz

mass hazard at t.

 Histogram

Compute the histogram of tracked object at the prediction.

Parameters:

printit

TRUE=output; FALSE=quiet

Comments:

output will also be produced for any objects in tlist with Volume > SILENT

 IncAcc

 p

 pnext

next prediction on the path

 pq

current p.

 Predict

Compute the predicted distribution of actions and states. Average values of tracked objects are stored in predmom Transitions to unreachable states is tracked and logged in the Data logfile.

Returns:

TRUE if all current states are termimal or last states.

See also:

Distribution

 Prediction

Create a new prediction.

Typically a user would create a PathPrediction or further derived classes which in turn creates predictions.

Parameters:

t	integer, position in the path.

 predmom

(unmasked) predicted moment vector

 q

index into sind.

 readmom

(unmasked) empirical moments.

 Reset

 SetMoms

Initialize and if necessary set moms vectors.

Parameters:

sz	length of current ctlist.
firsttype	first or only pass integrating over gamma_r This was added to reduce vector creation/destruction

 sind

state index

 t

rank in a pathprediction.

 W

masked weight to put on distance.

 PredictionDataSet

 EconometricObjective

The default econometric objective for a panel prediction: the overall GMM objective.

Parameters:

subp	DoAll (default), solve all subproblems and return likelihood vector Non-negative integer, solve only subproblem, return contribution to overall GMM

Returns:

M

 flist

 FMethod

 Nplace

observations column (index or label).

 Observations

Define columns where observations and time values appear.

Parameters:

NLabelOrColumn	integer or string for column of data containing observation counts. This is used to adjust weight of rows in moments.
TLabelOrColumn	integer or string for column of data containing model t value. Data must still be sorted in time! However, this allows missing time periods to be skipped. Further, it allows for a special row containing moment influence/importance adjustments.

 Observed

 PredictionDataSet

Create a panel prediction that is matched with external data.

Parameters:

UorCorL	where to get fixed-effect values matrix of indices or array of labels UseLabel [default] NotInData only allowed if F=1, then no column contains fixed variables
label	name for the data UseDefault [default] use classname of model class.
method	solution method to call before predict UnInitialized [default] no method (warning if there is heterogeneity)
iDist	initial conditions set to PathPredictions
wght	see GMMWeightOptions
aggshares	FALSE [default] or a Fx1 vector of population shares for groups to be used when creating an aggregate prediction.

 Read

Read in external moments of tracked objects.

Parameters:

FNorDB	string, name of file that contains the data. A Ox database object.
MaxT	TRUE: set length of prediction equal to the max{T} for all Paths, otherwise use individual lengths

 SimulateMomentVariances

Simulate sample of outcomes compute path Variance matrix and save inverse.

Parameters:

N	size of simulated sample for each PathPrediction.
ErogOrStateMat	initial condition from simulated outcome paths
@fvals	either DoALL or a a vector of fixed effect indices to compute.

 Tplace

time column (index or label).

 TrackingMatchToColumn

Track an object that is matched to column in the data.

Parameters:

LorC	label or column index in the data to associate with this moment.
mom	Discrete object to track

 TrackingWithLabel

Track one or more objects that are matched to columns using the object's label.

Parameters:

InDataOrNot	TRUE: the UseLabel tag will be passed to PathPrediction::Tracking() FALSE: the NotInData tag will be sent.
...	objects or array of objects to track

 UorCorL

matrix of indices or array of labels or UseLabel

 TrackObj

 print

Print mean and histogram of tracked object.