**Additions and improvements**- When you estimate a Logit or Probit model (via module LIMDEP), the Hausman test
is now automatically conducted, based on the comparison of the NLLS and ML parameter
estimates, provided that the difference of the variance matrices involved is non-singular.
Reference:

Hausman, J. A. (1978): "Specification Testing in Econometrics",*Econometrica*46,1251-1271. - It appears that the instructions regarding the allowable formats of CSV files in EasyReg module DATACSV were not clear enough. If Windows uses a dot (.) as decimal delimiter, the data entries are expected by EasyReg to be separated by commas (,), and if Windows uses a comma (,) as decimal delimiter, the data entries are expected by EasyReg to be separated by semi-colons (;). However, some versions of Excel use semi-colons as data entry separators even if Windows uses a dot as decimal delimiter. If the CSV file format does not correspond to the number setting of Windows, you will now get an error message window with a diagnosis of the problem, and the option to overrule the EasyReg default CSV formats.
- The module CSVSUBSAMPLE, which allows you take a subsample of a CSV file, will now ask whether the CSV file involved uses a comma or a semi-colon as data entry separator, in order to compute the number of records containing missing values correctly if the CSV file format does not correspond to the number setting of Windows.
- If a large data file in former EasyReg default format is imported, it may appear that EasyReg module DATAOLD freezes up (which is not the case, though). Now the reading and writing activities of EasyReg are traced in the menu bar of the window involved, and the mouse pointer will turn into a hour glass while EasyReg is digesting the file.
- All the maximum likelihood modules (ML, LIMDEP, TOBIT, HECKMAN, NEXTGARCH) now display the Akaike, Hannan-Quinn and Schwarz information criteria.
- The information criteria in the case of ARCH or GARCH models [module NEXTGARCH] were not comparable with those for models without ARCH or GARCH. They are now. See the updated lecture notes on information criteria and model selection.
- The integrated conditional moment test was calculated under the homoskedasticity assumption. It now assumes heteroskedasticity, which is the rule rather than the exception. See the EasyReg help file involved.
- The Hausman test for Probit and Logit models has been improved. Now the maximum likelihoods parameters are used for the calculation of the NLSS variance matrix, to guarantee (or at least to make it more likely) that its difference with the ML variance matrix is positive definite.
- The minimum number
*p*of lagged differences in the Augmented Dickey-Fuller unit root test was one and is now zero. Also, the order*q*of the AR model for the differenced time series used for simulation of the small sample distributions of various unit root tests can now be chosen equal to zero.

- When you estimate a Logit or Probit model (via module LIMDEP), the Hausman test
is now automatically conducted, based on the comparison of the NLLS and ML parameter
estimates, provided that the difference of the variance matrices involved is non-singular.
Reference:
**Bug repairs**- The nonlinear least squares module (NLLS) gave occasionally an error message (error 6: overflow) due to an obsolete integer-valued counter.
- A non-essential bug in the matrix calculation module (MATRIX) has been repaired.
- Two bugs in module ARIMAMODSEL, which determines the orders
*p*and*q*of anARMA( via the Akaike, Hannan-Quinn and Schwarz information criteria, have been repaired. The optimal model according to the Hannan-Quinn and Schwarz information criteria was always the same as the optimal model according to the Akaike criterion, although these criteria were correctly computed for each model. Moreover, when you repeated the model selection a second time you got an error message. The latter bug has been repaired by jumping back to the EasyReg main module when done.*p*,*q*) - Also in the ARIMA estimation and forecasting module you will now jump back to the EasyReg main module when done.
- There was another bug in module ARIMAMODSEL, which determines the orders
*p*and*q*of anARMA( via the Akaike, Hannan-Quinn and Schwarz information criteria. If no intercept, time trend or seasonal dummy variables are selected for the deterministic part of the model, the initial model has no parameters. This caused an error, "Run-time error 9: Subscript of range", in the simplex iteration procedure. The bug has been repaired by skipping the simplex procedure if there are no parameters.*p*,*q*) - If after restarting the simplex iteration in module ARIMA you try to uncheck "Auto restart", EasyReg stalled. It now finishes the simplex iteration and pauses.