It is not really possible to know exactly what caused this, but you can try various combinations of the TECH= and RIDGING= options. You could also try using the FIRTH option which maximizes a penalized likelihood. Since you have a fair number of predictors, you might be flirting with separation issues, so the the FIRTH option might help. Also you could consider using a model selection method (but don't use BACKWARD). The more modern Lasso method is available in PROC HPGENSELECT. ... View more

To build on what @lvm said, a parameter is a population value. A STATISTIC (which lvm calls a parameter estimate) is a random variable that depends on a random sample that is drawn from a population. The statistic has a distribution, which reflects that fact that if you choose a different random sample, you will obtain a different estimate. The collection of all estimates, taken over all random samples of a particular size, can be described by the probability distribution of the statistic, which is called the sampling distribution.    The standard error of a statistic is the standard deviation of the samping distribution. For simple statistics, the sampling distribution is known asymptotically (that is, for very large samples). For other statistics, you need to assume that the population is normally distributed if you want an approximate formula for the standard error. ... View more

Check out a similar thread here that discusses LGD:   https://communities.sas.com/t5/SAS-Risk-Management/Simulating-LGD-Dataset/m-p/271640#U271640   EAD is often simply just the loan exposure - both drawn and undrawn added. There is really no need to model this. By analysing in-default loans it is usually quite easy to confirm if EAD can be simply the total exposure or not. Sometimes this figure needs to be adjusted for various products. For example credit cards are often allowed to go significantly overlimit so in this case EAD might be set to 105 or 110 percent of exposure. Also with transaction accounts with shadow limits then EAD may need to be adjusted.  ... View more

In Linear regression, SAS use Least Square Method to estimate the coefficient, so you can use VIF to check collinearity. But in Logistic regression, SAS use Maximize Likelihood Method to estimate the coefficient. SAS will automatically check the  linear correlation between two variables,once SAS found collinearity ,SAS will set the coefficient of one of them to be zero .  ... View more

Simpsons Paradox   http://stats.stackexchange.com/questions/1580/regression-coefficients-that-flip-sign-after-including-other-predictors ... View more

You can take eigenvalue as the variance of covariance matrix . 1.0 means this eigenvalue is big enough to use its eigen vector to map something. or you can say this eigenvalue can explain/represent a big part of variance of matrix . ... View more

I don't think the KS test is an option in PROC LOGISTIC (see http://support.sas.com/kb/23/231.html).   However, it can be used in NPAIR1WAY. See this for more explanation: http://www.sas.com/content/dam/SAS/en_ca/User%20Group%20Presentations/Toronto-Data-Mining-Forum/MarkAn-CreditRiskQuality-Spring2010.pdf ... View more

I'm not sure why you're adding <div> tags around your titles. What are you trying to do? In general I recommend not inserting raw HTML into your output if it can be avoided. If you need custom styling in your titles, use the style inline function. ... View more

The chi-squared statistic in the output is just the Wald statistic, (estimate/SE)^2. The square root of the chi-squared statistic for a single parameter estimate is just the t statistic (with infinite denominator df).  ... View more

These two sections will give some hints: SAS(R) 9.4 OLAP Server: User's Guide, Second Edition SAS(R) 9.4 OLAP Server: User's Guide, Second Edition ... View more

Hi, Subquery is not required here , also it slows down the performance. Proc Sql; select distinct House_ID from have where House_ID <> '02' and Apple=1 ; quit; Just confirm the syntax for not equal to expression : <> Vishal ... View more

I'm a Psychologist, not a statistician, thus I'll leave it to the statisticians to either verify or critique the statements.  However, that said, the following statement from the article I linked said: "The alpha value gives us the probability of a type I error. Type I errors occur when we reject a null hypothesis that is actually true. Thus, in the long run, for a test with level of significance of 0.05 = 1/20, a true null hypothesis will be rejected one out of every 20 times." Why 0.05?  The article goes on to state: "Although in theory and practice many numbers can be used for alpha, the most commonly used is 0.05. The reason for this both because consensus shows that this level is appropriate, and historically it has been accepted as the standard." ... View more