Wilmott Forums - KMV and Merton model

Forum Navigation:

FORUMS > General Forum

< refresh >

Topic Title: KMV and Merton model
Created On Mon Oct 03, 05 03:08 PM

Topic View:

Antonios
Junior Member

Posts: 9
Joined: Oct 2005

Mon Oct 03, 05 03:08 PM

My question is towards the KMV-Merton methodology of calculating the probability of default. What is not clear to me is whether this is a VaR methodology or, in case it is not, how one can convert it into such.

Quote

Top

Bottom

Antonio
Senior Member

Posts: 609
Joined: Jun 2004

Mon Oct 03, 05 03:16 PM

Could you be more precise by "VaR methodology" ? The KMV-Merton model relies on calculating the Distance to Default with assumptions on the dyna;ics of the assets and a certain structure of the debt.

-------------------------
"Beer is proof that God loves us and wants us to be happy."
Benjamin Franklin

Quote

Top

Bottom

Antonios
Junior Member

Posts: 9
Joined: Oct 2005

Mon Oct 03, 05 03:31 PM

To be more precise, can we construct a Loss/Gain distribution via KMV-Merton and then derive a quantile representing the max. loss over a selected time horizon?

Quote

Top

Bottom

Anthis
Senior Member

Posts: 4049
Joined: Oct 2001

Mon Oct 03, 05 05:16 PM

You two are not the same person, right?

Quote

Top

Bottom

Antonios
Junior Member

Posts: 9
Joined: Oct 2005

Mon Oct 03, 05 05:24 PM

No we are not the same person. What about ny question? Do you have any hints?

Posts: 119
Joined: Sep 2003

Wed Oct 05, 05 01:51 PM

Get the Expected Default Frequency (EDF) for each asset, and the default correlations, plot the joint distribution of the expected losses. This would be your loss distribution. Get your percentile off this.

-------------------------
Bash on Regardless

Quote

Top

Bottom

Antonios
Junior Member

Posts: 9
Joined: Oct 2005

Wed Oct 05, 05 02:02 PM

For the case I deal with a single asset, do you think I can produce a distribution of losses/gain through Monte Carlo simulation?

Posts: 26
Joined: Dec 2003

Sat Oct 08, 05 04:07 PM

The KMV-Merton default forecasting model produces a probability of default for each firm in the sample at any given point in time.

Once you have this probability matrix, you can choose a particular point (say, the 95th quantile, for example) and multiply it by you firm?s value ? since at max you can loose the firm. This will be close in function to that of VaR.

Quote

Top

Bottom

erstwhile
Senior Member

Posts: 893
Joined: Mar 2003

Sat Oct 08, 05 10:26 PM

In principle, the advantage of a MC sim like this is that you should be able to simulate the value of all the firms assets and liabilities, and derivatives on all these, self consistently using one simulation!

But sadly this type of structural model is far too simplified and naive.

I think it is best used for educational purposes, or at best for screening large numbers of companies so that unusual debt-equity situation jump out at you. You can then perform fundamental analyses to determine what is going on.

I am by no means the first to make such criticisms - it is pretty widely accepted that a better type of model is needed. Equity vs credit trading highlights such weaknesses.

What exactly are you trying to calculate VaR for? And how would the VaR number be used?

Quote

Top

Bottom

secondMan
Senior Member

Posts: 362
Joined: Aug 2003

Mon Oct 10, 05 08:59 AM

i must damit i do not get the discussion. someone is asking on edf, someone is mixing it with default corr and then erstwhile seems to top it by shifting the level north by a mile.

i am not sure where the initial poster, toni, right?, stands. my understanding is that EDF is already an outcome of a distribution, where the position on the horizontal axis is defined by the default point. IMHo you could see this point as a VAR and the edf is the appropriate likelihood (better to see 1 minus this likelihood). as far as i know in some of their presentations KMV include other such probs in order to illustrate how different positions of the default point result in different ratings. actually i just ask myself what distribution they assume in the first place. since they claim their optionRoots, maybe lognormal?

maybe i am just adding noise here ...

erstwhile
in my opinion the real issue is less the model, but the source of one ingredient: the multiplier of debt in order to derive the default point. it is clearly below 1 (at least it appears to be to me), but the figure is quite debateable IMHO. if KMV derives it out of their database, thus calibrating it to different industries using default data, this might offset the limitations you mention. but i would think that your a mile northwest ...

peace

-------------------------
do it now.

Edited: Mon Oct 10, 05 at 11:07 AM by secondMan

Quote

Top

Bottom

erstwhile
Senior Member

Posts: 893
Joined: Mar 2003

Mon Oct 10, 05 11:02 PM

secondman - there are loads of problems with the very simplified KMV type of structural model. one particularly glaring one is the notion that "enterprise value" is well represented by the sum of market capitalisation and debt.

think about Cisco systems in the year 2000 - the market cap was huge - would that make you want to lend them money? were they a better credit then, when their stock price was multiples of what it is now?

then there is the problem of the time distribution of debt - KMV merely adds it all up. some companies have no significant debt due for 20 years, and other have most of their debt due in 3 years. with all else held constant, KMV would not distinguish between these two situations.

never mind all the other issues, like details of when a company can be refinanced, the fact that this may vary between sectors and countries, etc. or there is sometimes the probability of a government bailout, which can be a huge factor, but which would be ignored in a simplified model.

i do think it is good for educational purposes and as a screening tool though.

as far as the original question went, I think that was to do with the KMV model and VaR. i assumed the idea was to convert KMV/Creditgrades into a MC sim for VaR. doable, but not good, IMO.

Quote

Top

Bottom

secondMan
Senior Member

Posts: 362
Joined: Aug 2003

Tue Oct 11, 05 09:52 AM

i thought that KMV reflects the debt structure in their default point calibration. well, not purely idosyncratic since they do that sectorwise i think. IMHO CSCO at 80 [given same absolute level of debt, which will not be the case i guess

] is definitely less likely to default than below 10. but we come from different corners of the world. for me the pure fact that they use market data to determine default debt is already a huge advantage. debt equity modelling is still new to me ...

BTW what data sources on balance sheet data would you recommend, life and historic? compustat? are there others? i am contemplating to revive a multiregression equity market neutral factor model (APT kind of thing) on balance sheet data. (and i am a little annoyed with bloomberg, who corrects data after the fact, without leaving dates of the correction.)

peace

-------------------------
do it now.

Quote

Top

Bottom

erstwhile
Senior Member

Posts: 893
Joined: Mar 2003

Tue Oct 11, 05 10:13 AM

CSCO went from $80 to $15 in a year. i doubt that the company's "true creditworthiness" (vague, but intended to be a common-sense concept) was dramatically different in that period. we can agree to disagree on this point.

personally i haven't looked at historical balance sheet data - i couldn't tell you where to get it from.

i have only used creditgrades types models using Bloomberg balance sheet data - i agree it is pretty chaotic sometimes.

i am giving several talks at conferences on equity-credit trading in the next few months, and my final conclusion will be that fundamentally based analysis on the equity and debt of individual companies needs to be done to form a sound investment strategy. things like KMV/Creditgrades/Merton models, or multifactor regressions should not be the core analysis performed in commiting capital.

Posts: 55
Joined: Aug 2004

Tue Oct 11, 05 04:46 PM

I don't think I agree on the CSCO point. With the stock at $80 the balance sheet is in better shape and they have much better refinancing options - its only with a degree of hindsight that you can dismiss the information content of the ridiculous share price, particularly as to the potential future cash flows this was discounting.

Quote

Top

Bottom

secondMan
Senior Member

Posts: 362
Joined: Aug 2003

Wed Oct 12, 05 08:44 AM

Quote
Originally posted by: erstwhile
CSCO went from $80 to $15 in a year. i doubt that the company's "true creditworthiness" (vague, but intended to be a common-sense concept) was dramatically different in that period. we can agree to disagree on this point.

personally i haven't looked at historical balance sheet data - i couldn't tell you where to get it from.

i have only used creditgrades types models using Bloomberg balance sheet data - i agree it is pretty chaotic sometimes.

i am giving several talks at conferences on equity-credit trading in the next few months, and my final conclusion will be that fundamentally based analysis on the equity and debt of individual companies needs to be done to form a sound investment strategy. things like KMV/Creditgrades/Merton models, or multifactor regressions should not be the core analysis performed in commiting capital.

hmm. i see your point and i sense where you are heading. yes, a bubble is a bubble. and i assume that each and every model will have its flaws under extreme circumstances. it might be completely insane to analyse high growth companies by their current balance sheet, since their value might be always much more driven by forecasts. (not necessarily "fantasy"). and let us not forget: after the fact we all are wise.

i think in principle you will agree that a stock trading at 80 will have different default prob than one at 10. interesting thing BTW, does anyone know studies an on that? (alwys thinking in terms of market cap and not directly price of course.)

can you "publish" material on your talk after you delievered them? would love to here something on that. do you kow a market neutral bond (or cds) fund? pairs trading bonds? that kind of thing ...

[BTW am i the only one who thinks CDS is unethical, given that it is 70% of traded bond volume, yet does not provide a single $ for the companies ...?]

-------------------------
do it now.