For many probability of defaults models in credit risk it is needed to use data observed from a "full" business cycles. Usually a business cycle is defined as a recurring (not necessarily periodic) economic cycle of recession and expansion.

My question now is pretty general but maybe there is some knowledge of it. Are there any available methods to compensate if not a full business cycle is observed? And how can one tell if a full business cycle is observed? Thanks for any help!


General questions beget general answers but hopefully mine offers some additional insight. If you do not have full business cycles, how about different phases in the business cycle (generally one could say there are 4)? Recovery, Expansion, Slowdown, Recession.

If you are missing recession data you can generally fill those in using comparable industry/sector data. There has to be SOMETHING that exists in SOME shape or form for the Great Recession of 08, Dot Com bubble of 01-02... and you can always look at smaller downturns that affect sector specific spreads. The judgement call here would be how much do you tweak that.

If you are missing "the good times" that is not so important. One could always be conservative and retain upside (especially in the case of default assumption).

Again, really not sure what you are looking at, so specific cases might require specific techniques.

  • 1
    $\begingroup$ Yes this is also the way I'm considering this problem. In fact I'm using a regime switching model on GDP to find movement of business cycles. Then I fit a Transfer Function model (or distributed lag) to backward forecast the observed default frequencies. I have obtained some good results on S&P odf rates. Still thanks for your input! $\endgroup$
    – user18514
    Mar 3 '16 at 7:35
  • $\begingroup$ Thanks for sharing the follow up, glad I helped somewhat. $\endgroup$
    – Larasing
    Mar 3 '16 at 17:29

I am not clear with what you mean. You usually have data on business cycle up until two quarters before the actual date. Check the NBER: https://research.stlouisfed.org/fred2/series/USRECM

If you want to know the current business cycle you can just extrapolate it from the last data point, or you can compute a probability transition matrix.

  • $\begingroup$ Ideally, credit risk models are based on data covering at least a full business cycle. If too little data is available, one must use alternative methods to compensate for the missing part of the cycle. But how could one compensate for this missing part? $\endgroup$
    – user18514
    Dec 2 '15 at 15:34
  • $\begingroup$ Can you be more specific on which data you need? Without knowing the model requirements is hard to help. $\endgroup$
    – phdstudent
    Dec 2 '15 at 15:40
  • $\begingroup$ I guess it would be macro economic data like GDP, which is an indicator for business cycles. For example it is usually assumed that having atleast 8 years of data will give a good representation of a business cycle. But how could one prove that atleast one business cycle is observed? $\endgroup$
    – user18514
    Dec 2 '15 at 15:48
  • $\begingroup$ I'm not sure where you are getting the 8 year figure. Measuring business cycles is subjective, and no two are alike. The best common measure as mentioned above is NBER, while some others may disagree for good reasons. For example, if I recall correctly, the NBER lists both the 1980 and 1981 as two different cycles, while others would say it is one cycle. So for the industry standard NBER measure, using 8 years of data from each of those would introduce some rather unfortunate overlapping data. $\endgroup$
    – horseless
    Mar 1 '16 at 20:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy