All the programs can be freely downloaded for the purpose of research and academic communication.
It cannot be resold or used for any commercial purpose.
Comments and suggestions are sincerely appreciated.
Microeconometrics Tools (Classical Inference, Simulation)
Ordinary Least Squares (OLS) and More
The OLS and More package includes baseline OLS regression as well as White robust standard error, Cochrane-Orcutt AR(p) disturbances, and TSLS with instruments. A graphical user interface is added for beginners in econometrics. It is intuitive and virtually requires no knowledge in Matlab. Users only need to prepare EXCEL data and regressors names consistent with the EXCEL heading. Variable names of international languages are supported, so you can name your regressors in your favored language. Variable names with lags are supported, so you can name your variable like GDP(-1), stock(-2) etc.
Baseline Logit and Probit Model
The baseline Logit and Probit model package includes a graphical user interface to estimate binary or ordered discrete dependent variable model.
The latent variable and threshold framework is used, and log likelihood is maximized by the observable variates.
It can estimate
1) Binary Probit Model
2) Binary Logit Model
3) Ordered Probit Model
4) Ordered Logit Model
Unordered Logit Model with RUM
Unordered Logit model package contains two Matlab functions. It can estimate: 1) multinomial Logit model (i.i.d. extreme value disturbances), 2) nested multinomial Logit model (gereralized extreme value disturbances). The two functions use the random utility framework and maximize likelihood of realized choice.
The Tobit package includes a graphical user interface to facilitate estimation.
It can estimate:
1) Baseline Tobit model with a known lower and / or upper bound due to censoring,
2) Tobit with unknown lower bound, using maximum likelihood,
3) Tobit with unknown lower bound, using Heckman two steps.
Hidden Markov Chain (HMM) package includes several functions to decode and estimate the HMM model. The model is driven by Markovian switching regimes which are unobservable. Conditional on a latent state, an emission signal is observed. The signal variable can be normal, Poisson, multinomial or user specified p.d.f. If the model parameters are known, the program can decode the model using Baum-Welch algorithm and find the most likely path using the Viterbi algorithm. If the model parameters are unknown, the program can estimate the model with E-M algorithm and Bayesian Gibbs sampler.
Illustration of Expectation—Maximization (EM) Algorithm
Expectation—Maximization package contains 3 simple applications of the EM algorithm.
It can estimate:
1) Classic regression model with missing dependent variable data,
2) Tobit model with lower and upper censoring,
3) Regression model with Student-t error.
Aggregated Covariate Data Model (Qian, 2010)
The package contains several MATLAB functions pertaining to aggregated covariate data (ACD) regression
discussed in the paper Linear Regression Using Both Temporally Aggregated and Temporally Disaggregated Data: Revisited (Qian, 2010).
In the ACD model, one key covariate in the regression is aggregated by household, or by group, by region, by time, and so on.
There is also another equation to impute the missing disaggregated covariate.
AGGREGATE_ML1.m implements the analytic maximum likelihood to the ACD model with instruments.
AGGREGATE_ML2.m implements the analytic maximum likelihood to the ACD model without instruments.
AGGREGATE_ML1.m implements the numerical maximum likelihood to the ACD model with many instruments.
AGGREGATE_MSE_TSLS.m implements the minimum MSE TSLS (Hsiao, 1979) to the ACD model.
AGGREGATE_TSLS.m implements the TSLS (Dagenais, 1973) to the ACD model.
AGGREGATE_GIBBS1 is the Bayesian estimation of the ACD model via Gibbs sampler.
SIM1.m and SIM2.m are simulation studies of those estimators as in Qian (2010).
Bootstrap Bias Correction with MIV (Qian,2011)
The package contains several MATLAB functions pertaining to bootstrap bias correction in the presence of monotone instrumental variables (MIV).
The MIV models feature a supremum operator in the lower bound and an infimum operator in the upper bound.
However, by Jensen’s inequality, the analogue estimate of the lower bound is biased upwards and upper bound biased downwards,
resulting in estimates that are narrower than the true bounds.
Qian (2011) proposes a multi-level simultaneous bootstrap procedure to correct the bias.
BIAS_CORRECT_MAX.m implements the multi-level simultaneous bootstrap procedure to correct the finite sample bias of max(.).
BIAS_CORRECT_MIN.m implements the multi-level simultaneous bootstrap procedure to correct the finite sample bias of min(.).
B2_shape.m illustrates the shape of the first and second level of the bias function induced by max(x1,x2),
where x1 and x2 are two independent normal variates, as discussed in Qian (2011).
Bayesian Inference with MIV (Qian,2011)
The package contains several MATLAB functions pertaining to Bayesian Inference with MIV. The algorithm is based on "Bayesian Inference with Monotone Instrumental Variables" by Hang Qian (2011). It also contains an application on the effects of taking extra classes on high school students' test scores.
These programs were originally written by Dr. Hang Qian of Iowa State University.
Since these codes are not available on the author's website, they were reproduced here.