{smcl}
{* August 2007}{...}
{hline}
help for {hi:mvtobit}{right: Mikkel Barslund (mikkelbarslund@gmail.com)}
{hline}
{title:Multivariate tobit models estimated by maximum simulated likelihood}
{p 4 12 2}{cmd:mvtobit} {it:equation1} {it:equation2} {it: ...} {it:equationM}
[{it:weight}] [{cmd:if} {it:exp}] [{cmd:in} {it:range}] [{cmd:,}
{cmdab:dr:aws(}{it:#}{cmd:)} {cmdab:an} {cmdab:s:eed(}{it:#}{cmd:)}
{cmdab:b:eta0} {cmdab:a:trho0(}{it:matrix_name}{cmd:)}
{cmdab:pre:fix(}{it:string}{cmd:)} {cmdab:bu:rn(}{it:integer}{cmd:)}
{cmdab:ran:dom} {cmdab:hran:dom} {cmdab:sh:uffle} {cmdab:adoon:ly}
{cmdab:prim:es(}{it:matrix_name}{cmd:)} {cmdab:init(}{it:matrix_name}{cmd:)}
{cmdab:r:obust} {cmdab:cl:uster(}{it:varname}{cmd:)}
{cmdab:const:raints(}{it:numlist}{cmd:)}
{cmdab:l:evel(}{it:#}{cmd:)} {it:maximize_options} ]
{p 4 4 2}where each equation is specified as
{p 12 12 2}{cmd:(} [{it:eqname}{cmd::}] {it:depvar} [{cmd:=}] [{it:varlist}]
[{cmd:,} {cmdab:nocon:stant}] {cmd:)}
{p 4 4 2}{cmd:by} {it:...} {cmd::} may be used with {cmd:mvtobit}; see help
{help by}.
{p 4 4 2}{cmd:pweight}s, {cmd:fweight}s, {cmd:aweight}s, and {cmd:iweight}s
are allowed; see help {help weights}.
{p 4 4 2}{cmd:mvtobit} is likely to produce a series of "(not concave)"
statements in the beginning of the estimation process. It is recommended to
specify the {cmd:difficult} option; see help {help maximize}.
{p 4 4 2}{cmd:mvtobit} shares the features of all estimation commands; see help
{help est}.
{p 4 4 2}{cmd:mvtobit} typed without arguments redisplays the last estimates.
The {cmd:level} option may be used.
{p 4 4 2}{cmd:mvtobit} requires {cmd:mdraws} to be installed.
{p 4 4 2}{cmd:Note:} much code in this routine is hacked from or inspired by Cappellari
and Jenkins' {cmd:mvprobit} and {cmd:mdraws} commands (see {help mvprobit} and
{help mdraws} if installed). This in particular applies to the help and syntax handling files.
{cmd:mdraws} must be installed for {cmd:mvtobit} to work. The {cmd:shuffle}
option requires installation of {cmd:_gclsort}. Both are available from SSC.
{p 4 4 2}{it:{cmd: Using Stata version 9 or above? }}{it:Take a look at {stata findit cmp:cmp} and {browse "http://www.cgdev.org/content/publications/detail/1421516/":Roodman (2009)}. }
{title:Description}
{p 4 4 2}{cmd:mvtobit} estimates {it:M}-equation tobit models (including bivariate models),
by the method
of maximum simulated likelihood (MSL). Bivariate tobit models are estimated without
simulation (see also Daniel Lawsons {help bitobit} if installed). A limitation is that
only models left-censored at zero can be estimated, i.e.
{center: y(i) = max[xb(i)+e(i),0]}
{p 4 4 2} where {it:e} is {it:M}-variate normally distributed. Along with
coefficients for each equation {cmd:mvtobit} estimates the cross-equation
error-correlations and the variance of the error terms.
{p 4 4 2}{cmd:mvtobit} uses the Geweke-Hajivassiliou-Keane (GHK) simulator
implemented in {cmd:egen} function {help mvnp} (if installed) and the
related {help mdraws} function to draw random numbers for evaluation of
the multi-dimensional Normal integrals in the likelihood function.
For each observation, a likelihood contribution is calculated
for each replication, and the simulated likelihood contribution is the
average of the values derived from all the replications. The simulated
likelihood function for the sample as a whole is then maximized using
standard methods ({cmd:ml} in this case). For a brief description of the
GHK smooth recursive simulator, see Greene (2003, 931-933), who also
provides references to the literature. See Cappellari and Jenkins (2006)
for detailed information on implemention of MSL in Stata and the workings
of {help mvnp} and {help mdraws}. Also see Train (2003).
{p 4 4 2}Under standard conditions, the MSL estimator is consistent as the
number of observations and the number of draws tend to infinity and is
asymptotically equivalent to the true maximum likelihood estimator as the
ratio of the square root of the sample size to the number of draws tends to
zero. Thus, other things equal, the more draws, the better. In practice,
however, it has been observed that a relatively small number of draws may work
well for `smooth' likelihoods in the sense that the change in estimates as the
number of draws is increased is negligible. It is the responsibility of the
user to check that this is the case. Simulation variance may be reduced
using antithetic draws in addition to the pseudo-random uniform variates used
in the calculation of the simulated likelihood. The antithetic draws for a
vector of pseudo-random uniform draws, z, are 1-z.
{p 4 4 2}Estimation is numerically intensive and may be very slow if the data
set is large, if the number of draws is large, or (especially) if the
number of equations is large. Users may also need to {cmd:set matsize}
and {cmd:set memory} to values above the default ones. (See help for
{help matsize} and {help memory}.) Use of the {cmd:atrho0} option may
speed up convergence.
{p 4 4 2}Models for which the error variance-covarince matrix is close
to not being positive definite are likely to be difficult to maximize.
(The Cholesky factorization used by MSL requires positive definiteness.)
In these cases, {cmd:ml} may report difficulties calculating numerical
derivatives and a non-concave log likelihood. In difficult maximization
problems, the message "Warning: cannot do Cholesky factorization of rho matrix"
may appear between iterations. It may be safely ignored if the maximization proceeds
to a satisfactory conclusion. Results may differ depending on the sort order
of the data, because the sort order affects which values of the random
variable(s) get allocated to which observation. (Note, {cmd:mvtobit} does
not change the sort order of the data.) This potential
problem is reduced by the larger the number of random draws that is used.
{title:Options}
{p 4 8 2}{cmd:beta0} specifies that the estimates of the marginal tobit
regressions (used to provide starting values) are reported.
{p 4 8 2}{cmd:atrho0(}{it:matrix_name}{cmd:)} allows users to specify
starting values for the standard deviations and correlations that
are different from the default values (zeroes and ones, respectively).
The matrix {it:matrix_name} contains values of the incidental parameters,
/lnsigma{it:i} and /atrho{it:ij}, for the {it:M} equations. Matrix
{it:matrix_name} must have properly named column names. E.g., if
a starting value in /atrho{it:12} is being set, one would first use the
command {cmd:matrix {it:matrix_name} = ({it:value})}, followed by
{cmd:matrix colnames {it:matrix_name} = atrho12:_cons}. Between 1 and
{it:M}({it:M}-1)/2 /atrho{it:ij}, and between 1 and {it:M} /lnsigma{it:i}
starting values may be specified, where {it:i} = 1,...,{it:M-1}, and
{it:j} > {it:i}. One likely source for a non-default starting value
for atrho{it:ji} is the /athrho parameter estimate from a bivariate model
corresponding to equations {it:i} and {it:j} of the full {cmd:mvtobit} model.
{p 4 8 2}{cmd:robust} specifies that the Huber/White/sandwich estimator of
variance is to be used in place of the traditional calculation; see
{hi:[U] 23.11 Obtaining robust variance estimates}. {cmd:robust} combined
with {cmd:cluster()} allows observations that are not independent within
clusters (although they must be independent between clusters). If you
specify {help pweight}s, {cmd:robust} is implied.
{p 4 8 2}{cmd:cluster(}{it:varname}{cmd:)} specifies that the observations are
independent across groups (clusters) but not necessarily within groups.
{it:varname} specifies to which group each observation belongs; e.g.,
{cmd:cluster(personid)} in data with repeated observations on individuals.
See {hi:[U] 23.11 Obtaining robust variance estimates}. {cmd:cluster()} can be
used with {help pweight}s to produce estimates for unstratified
cluster-sampled data. Specifying {cmd:cluster()} implies {cmd:robust}.
{p 4 8 2}{cmd:noconstant} suppresses the constant term (intercept) in the
relevant regression.
{p 4 8 2}{cmd:constraints(}{it:numlist}{cmd:)} specifies the linear constraints
to be applied during estimation. Constraints are defined using the
{cmd:constraint} command and are numbered; see help {help constraint}. The
default is to perform unconstrained estimation.
{p 4 8 2}{cmd:level(}{it:#}{cmd:)} specifies the confidence level, in percent,
for the confidence intervals of the coefficients; see help {help level}.
{p 4 8 2}{cmd:init(}{it:matrix_name}{cmd:)} specifies a matrix of starting values.
Options from {cmd:ml init} can be specified inside the parenthesis.
{p 4 8 2}{it:maximize_options} control the maximization process; see help
{help maximize}. Use of them is likely to be rare.
{title:Options related to random number generation}
See also {help mdraws}. The explanations below are taken from the {cmd:mdraws} helpfile.
{p 4 8 2}{cmd:draws(}{it:#}{cmd:)} specifies the number of pseudo-random
standard uniform variates drawn when calculating the simulated likelihood.
The default is 5. (See the discussion above concerning the choice of the
number of draws.) If the {cmd:an} option is specified, the total number of
draws used in the calculations is twice the number specified in
{cmd:draws(}{it:#}{cmd:)}.
{p 4 8 2}{cmd:prefix(}{it:string}{cmd:)} specifies the prefix common
to the names of each of the created variables containing the random
numbers used by the {cmd:egen} function {help mvnp:mvnp()}. The default
prefix is X_MVT.
{p 4 8 2}{cmd:an} specifies that antithetic draws is to be used by {cmd:mdraws}.
The antithetic draw for a vector of uniform draws, z, is 1-z.
{p 4 8 2}{cmd:random} specifies that pseudorandom number sequences
are created rather than Halton sequences (the default).
{p 4 8 2}{cmd:seed(}{it:#}{cmd:)} specifies the initial value of the
(pseudo-)random-number seed used by the {cmd:mdraws} function
in the simulation process. The value should be an integer (the
default value is 123456789). Warning: if the number of draws is 'small',
changes in the seed value may lead to surprisingly large changes in
estimates. {cmd:seed(}{it:#}{cmd:)} only has effect when {cmd:random},
{cmd:hrandom} or {cmd:shuffle} are specified.
{p 4 8 2}{cmd:primes(}{it:matrix_name}{cmd:)} specifies the name of an
existing 1 x {it:M} or {it:M} x 1 matrix containing
{it:M} different prime numbers. If the option is not specified and
as long as {it:M} <= 20, the program uses the first {it:M} prime numbers
in ascending order to generate the Halton sequences.
{p 4 8 2}{cmd:burn(}{it:#}{cmd:)} specifies the number of initial
sequence elements to drop for each equation when creating Halton
sequences. The default is zero, and the option is ignored if
{cmd:random} is specified. Specification of this option
reduces the correlation between the sequences in each dimension.
Train (2003, 230) recommends that {it:#} should be at least as
large as the prime number used to generate the sequences.
{p 4 8 2}{cmd:hrandom} specifies that each Halton sequence should be
transformed by a random perturbation. For each dimension, a draw,
{it:u}, is taken from the standard uniform distribution.
Each sequence element has {it:u} added to it. If the sum is greater
than 1, the element is transformed to the sum minus 1; otherwise, the
element is transformed to the sum. See Train (2003, 234).
{p 4 8 2}{cmd:shuffle} specifies that "shuffled" Halton draws should be
created, as proposed by Hess and Polak (2003). Each Halton sequence in
each dimension is randomly shuffled before sequence elements are
allocated to observations. Philippe Van Kerm's program {cmd: _gclsort},
available via SSC, must be installed for this option to work.
{p 4 8 2}{cmd:adoonly} prevents using the Stata plugin to perform the
intensive numerical calculations. Specifying this option results in
slower-running code but may be necessary if the plugin is not available
for your platform. This option is also useful if you like to do speed
comparisons!
{title:Saved results}
{p 4 4 2} In addition to the usual results saved after {cmd:ml},
{cmd:mvtobit} also saves the following:
{p 4 8 2}{cmd:e(draws)} is the number of pseudo-random draws used when simulating
probabilities. If the {cmd:an} option is specified, {cmd:e(draws)} is twice the
number specified in {cmd:draws(}{it:#}{cmd:)}, rather than equal to the number.
{p 4 8 2}{cmd:e(an)} is a local macro containing "yes" if the {cmd:an} option is
specified, and containing "no" otherwise.
{p 4 8 2}{cmd:e(seed)} is the initial seed value used by the random-number
generator.
{p 4 8 2}{cmd:e(neqs)} is the number of equations in the {it:M}-equation model.
{p 4 8 2}{cmd:e(ll0)} is the log likehood for the comparison model (the sum of
the log likelihoods from the marginal univariate tobit models corresponding
to each equation).
{p 4 8 2}{cmd:e(chi2_c)} is chi-square test statistic for the likelihood ratio
test of the multivariate tobit model against the comparison model.
{p 4 8 2}{cmd:e(nrho)} is the number of estimated rhos (the degrees of freedom
for the likelihood ratio test against the comparison model).
{p 4 8 2}{cmd:e(sigma{it:i})} is the estimate of the standard deviation of the {it:i}'th
error term.
{p 4 8 2}{cmd:e(sesigma{it:i})} is the estimated standard error of sigma{it:i}.
{p 4 8 2}{cmd:e(rho{it:ji})} is the estimate of correlation {it:ji} in the
variance-covariance matrix of cross-equation error terms.
{p 4 8 2}{cmd:e(serho{it:ji})} is the estimated standard error of
correlation {it:ji}.
{p 4 8 2}{cmd:e(rhs{it:i})} is the list of explanatory variables used in
equation {it:i}. This list does not include the constant term, regardless
of whether there is one is implied by equation {it:i}.
{p 4 8 2}{cmd:e(nrhs{it:i})} is the number of explanatory variables in equation
{it:i}. This number includes the constant term if there is one implied by
equation {it:i}.
{title:Examples}
{p 8 12 2}{cmd:. mvtobit (y1 = x11 x12) (y2 = x21 x22) }
{p 8 12 2}{cmd:. mvtobit (y1 = x11 x12) (y2 = x21 x22) (y3 = x31 x32), dr(20) an }
{p 8 12 2}{cmd:. constraint define 1 [y1]x11 = [y2]x22 }
{p 8 12 2}{cmd:. mvtobit (y1 = x11 x12) (y2 = x21 x22) (y3 = x31 x32), dr(20) an constraints(1) }
{title:Authors}
{p 4 4 2}{browse "http://ideas.repec.org/f/pba424.html":Mikkel Barslund}, Danish Economic Councils, Denmark{break}
{title:Acknowledgments}
{p 4 4 2}I have hacked a large amount of code from Cappellari and Jenkins {cmd:mvprobit}
(Cappellari and Jenkins, 2003). In addition most of the heavy work in this routine is
performed by their {cmd:mdraws} command. All errors are, of course, mine.
{title:Version}
{p 4 4 2}Version 1.0, August, 2007.
{title:References}
{p 4 8 2} Cappellari, L. and S.P. Jenkins. 2003. Multivariate probit regression
using simulated maximum likelihood. {it:The Stata Journal} 3(3): 278{c -}294.
{p 4 8 2} Cappellari, L. and S.P. Jenkins. 2006. Calculation of multivariate
normal probabilities by simulation, with applications to maximum simulated
likelihood estimation. {it:The Stata Journal} 6(2): 156{c -}189.
{p 4 8 2} Greene, W.H. 2003. {it:Econometric Analysis}, 5th ed.
Upper Saddle River, NJ: Prentice-Hall.
{p 4 8 2} Roodman, D. 2009. {browse "http://www.cgdev.org/content/publications/detail/1421516/":Estimating Fully Observed Recursive Mixed-Process Models with cmp.}
Working Paper 168. Center for Global Development.
{title:Also see}
{p 4 19 2}Manual: {hi:[R] intreg},{p_end}
{p 13 13 2}{hi:[R] tobit}
{p 4 19 2}Online: help for {help constraint}, {help est}, {help ereturn}, {help postest},
{help ml}, {help tobit}, and (if installed)
{help bitobit}, {help mdraws}, {help mvnp}.{p_end}