Hello everyone,
I am running a logit model in which I try to forecast insolvency. The coefficients one obtains using standard logit estimation being the effet on the log-odds, I wanted something more interpretable. It seemed to me the best way to go was to run margins, dyex (L.X1) to know, when X1 increases by 1%, what is the resulting increase in the probability to become insolvent (in absolute units, e.g., a .01 increase in probability).
I know margins is a brilliant but sensitive command, so perhaps someone can confirm that this is indeed the tool to use here. Perhaps dydx would be better, given that my covariates are financial ratios (like ROA), I should do dydx and divide by hundred to have percentage point changes (in level)?
My main question, however, concerns the dyex command. I noticed that when I run dydx, I have the expected signs (increase in ROA decreases probability of insolvency) but when I use dyex, the sign changes and is now positive, following the same regression!
Here is a dataex:
----------------------- copy starting from the next line -----------------------
------------------ copy up to and including the previous line ------------------
Thank you in advance for any insight on this.
Respectfully,
John Rivers
I am running a logit model in which I try to forecast insolvency. The coefficients one obtains using standard logit estimation being the effet on the log-odds, I wanted something more interpretable. It seemed to me the best way to go was to run margins, dyex (L.X1) to know, when X1 increases by 1%, what is the resulting increase in the probability to become insolvent (in absolute units, e.g., a .01 increase in probability).
I know margins is a brilliant but sensitive command, so perhaps someone can confirm that this is indeed the tool to use here. Perhaps dydx would be better, given that my covariates are financial ratios (like ROA), I should do dydx and divide by hundred to have percentage point changes (in level)?
My main question, however, concerns the dyex command. I noticed that when I run dydx, I have the expected signs (increase in ROA decreases probability of insolvency) but when I use dyex, the sign changes and is now positive, following the same regression!
Here is a dataex:
----------------------- copy starting from the next line -----------------------
Code:
* Example generated by -dataex-. To install: ssc install dataex clear input int(date comp_num) float(impt ROA1 roa_v log_TA bshare1 opexp_prem_pc part_ceded_pc) long yearcountcode 1996 1 0 . .04198379 . . . . 104 1997 1 0 . .04198379 . . . . 105 1998 1 0 . .04198379 . . . . 106 1999 1 0 . .04198379 . . . . 107 2000 1 0 . .04198379 . . . . 108 2001 1 0 . .04198379 . . . . 109 2002 1 0 . .04198379 . . . . 110 2003 1 0 . .04198379 . . . . 111 2004 1 0 . .04198379 . . . . 112 2005 1 0 . .04198379 . . . . 113 2006 1 0 . .04198379 . . . . 114 2007 1 0 . .04198379 . . . . 115 2008 1 0 . .04198379 . . . . 116 2009 1 0 . .04198379 . . . . 117 2010 1 0 . .04198379 . . . . 118 2011 1 0 . .04198379 . . . . 119 2012 1 0 . .04198379 . . . . 120 2013 1 0 -.05765611 .04198379 7.88339 . 1.3128875 0 121 2014 1 0 .02736935 .04198379 8.080724 . .7207186 0 122 2015 1 0 .011651821 .04198379 8.216535 . .57930404 0 123 2016 1 0 .033698197 .04198379 8.305705 0 .7495359 0 124 1996 2 0 .11384311 .05007874 8.858372 .698357 .12621942 .8667591 104 1997 2 0 .014075192 .05007874 9.156266 .6947108 .09772328 .8628223 105 1998 2 0 -.067483835 .05007874 9.292594 .6842238 .2128755 .4671292 106 1999 2 0 .03868382 .05007874 9.440553 .6038181 .1470378 .5399969 107 2000 2 0 -.015914656 .05007874 9.693195 .6634216 .289766 .03936061 108 2001 2 0 .008663441 .05007874 9.831507 .7028519 .244019 .13513243 109 2002 2 0 -.05088913 .05007874 10.040242 .7829893 .2399435 .12845938 110 2003 2 0 .09041487 .05007874 10.0453 .7907972 .1895433 .4205122 111 2004 2 0 .028980885 .05007874 10.108306 .8368427 .24578042 .345214 112 2005 2 0 .04474738 .05007874 10.23015 .8485531 .26169425 .28048077 113 2006 2 0 .03514605 .05007874 10.314502 .8578923 .3478031 .09495977 114 2007 2 0 .025653735 .05007874 10.34216 .8561094 .47965536 .14605303 115 2008 2 0 .03241699 .05007874 10.304285 .9114216 .3747966 .14372167 116 2009 2 0 .02928237 .05007874 10.324498 .8922768 .4047336 .13893671 117 2010 2 0 .0032238334 .05007874 10.316257 .9090319 .411041 .14052549 118 2011 2 0 -.06658515 .05007874 10.33746 .8957891 .415263 .1407712 119 2012 2 0 -.07770962 .05007874 10.254777 .8948808 .40365475 .1395058 120 2013 2 0 .06433318 .05007874 10.279704 .8699471 .39733365 .13067509 121 2014 2 0 -.0012459718 .05007874 10.3082 .8564203 .4246248 .11834542 122 2015 2 0 .01863146 .05007874 10.261472 .8964072 .45057055 .09769556 123 2016 2 0 -.03054271 .05007874 10.273818 .884986 .4349437 .0811611 124 1996 3 0 . .04233585 . . . . 104 1997 3 0 . .04233585 . . . . 105 1998 3 0 . .04233585 . . . . 106 1999 3 0 . .04233585 . . . . 107 2000 3 0 .01655973 .04233585 7.254329 1 .07175426 .7799376 108 2003 3 0 -.07925805 .04233585 7.725895 1 .1011939 .25212994 111 2004 3 0 .003809625 .04233585 8.12962 1 .22251 .195481 112 2005 3 0 .05451892 .04233585 8.352901 1 .212595 .09391081 113 2006 3 0 .06793278 .04233585 8.559239 1 .20486717 .09431782 114 2007 3 0 .05297162 .04233585 9.100768 .9280499 .2315477 .0950568 115 2008 3 0 .06102193 .04233585 9.187486 .9913161 .2998646 .09803285 116 2009 3 0 .05971158 .04233585 9.291459 .8570565 .2482366 .10014564 117 2010 3 0 .032083604 .04233585 9.362007 .7935006 .3057735 .11595336 118 2011 3 0 -.0018972246 .04233585 9.421637 .8444695 .2989494 .11024439 119 2012 3 0 -.0017617185 .04233585 9.511732 .8016667 .3201108 .10802703 120 2013 3 0 -.0023159585 .04233585 9.655608 .7689856 .2910347 .09641418 121 2014 3 0 -.005131893 .04233585 9.771913 .7489496 .317059 .09962157 122 2015 3 0 -.00922158 .04233585 9.841464 .7384962 .3164138 .09748398 123 2016 3 0 -.04862303 .04233585 9.918364 .7321743 .29802006 .10116556 124 2001 4 0 . .052996 . . . . 109 2010 4 0 . .052996 . . . . 118 2011 4 0 . .052996 . . . . 119 2012 4 0 . .052996 . . . . 120 2013 4 0 . .052996 . . . . 121 2014 4 0 . .052996 . . . . 122 2015 4 0 . .052996 . . . . 123 2016 4 0 . .052996 . . . . 124 1996 5 0 . .03864212 . . . . 104 1997 5 0 . .03864212 . . . . 105 1998 5 0 . .03864212 . . . . 106 1999 5 0 . .03864212 . . . . 107 2000 5 0 . .03864212 . . . . 108 2004 5 0 .07557066 .03864212 7.082619 1 .563084 0 112 2006 5 0 . .03864212 . . . . 114 2007 5 0 . .03864212 . . . . 115 2008 5 0 . .03864212 . . . . 116 2009 5 0 . .03864212 . . . . 117 2010 5 0 . .03864212 . . . . 118 2011 5 0 . .03864212 . . . . 119 2012 5 0 . .03864212 . . . . 120 2013 5 0 . .03864212 . . . . 121 2014 5 0 . .03864212 . . . . 122 2015 5 0 . .03864212 . . . . 123 2016 5 0 . .03864212 . . . . 124 1996 6 0 . .04523665 . . . . 104 1997 6 0 . .04523665 . . . . 105 1998 6 0 . .04523665 . . . . 106 1999 6 0 . .04523665 . . . . 107 2000 6 0 . .04523665 . . . . 108 2001 6 0 . .04523665 . . . . 109 2002 6 0 .0572899 .04523665 8.969716 .7638934 .4895625 0 110 2003 6 0 .1005542 .04523665 9.080692 .9828112 .3894725 0 111 2004 6 0 .08958496 .04523665 9.236841 .9994179 .3882671 0 112 2005 6 0 -.0002067057 .04523665 9.323193 1 .4095679 0 113 2006 6 0 . .04523665 . . . . 114 2007 6 0 . .04523665 . . . . 115 2008 6 0 . .04523665 . . . . 116 2009 6 0 . .04523665 . . . . 117 end format %ty date label values yearcountcode fixedston label def fixedston 104 "USA1996Y", modify label def fixedston 105 "USA1997Y", modify label def fixedston 106 "USA1998Y", modify label def fixedston 107 "USA1999Y", modify label def fixedston 108 "USA2000Y", modify label def fixedston 109 "USA2001Y", modify label def fixedston 110 "USA2002Y", modify label def fixedston 111 "USA2003Y", modify label def fixedston 112 "USA2004Y", modify label def fixedston 113 "USA2005Y", modify label def fixedston 114 "USA2006Y", modify label def fixedston 115 "USA2007Y", modify label def fixedston 116 "USA2008Y", modify label def fixedston 117 "USA2009Y", modify label def fixedston 118 "USA2010Y", modify label def fixedston 119 "USA2011Y", modify label def fixedston 120 "USA2012Y", modify label def fixedston 121 "USA2013Y", modify label def fixedston 122 "USA2014Y", modify label def fixedston 123 "USA2015Y", modify label def fixedston 124 "USA2016Y", modify
Thank you in advance for any insight on this.
Respectfully,
John Rivers
Comment