Four Whole Years…Really?
Yesterday, in Fat-Tails, Normal Distributions, Random Walks, and all that Jazz I wrote that one of the underlying assumptions that form the backbone of “modern” finance (namely, that market risk can be accurately modeled) was fatally flawed.
I then stumbled on a post from Don Fishback’s Market Update in which Don discusses how, in the good times, banks used Value at Risk (VaR) metrics to justify taking on ever increasing leverage. There are many flavors of VaR of varying degrees of sophistication, but all inherently make a rather large (and fatal) assumption – that market risk can be accurately modeled.
Now I know all of this has been discussed before in the financial community, but bear with me…I’m getting to the punch line.
Fishback referenced a Bloomberg article from earlier this year on this topic that says:
“…Lehman uses four years of historical data to calculate VaR, with a higher weighting given to more recent time periods, while Morgan Stanley provides VaR calculations using both four years and one year of market data.”
(cue cymbal crash)
Really? Four years? You mean four whole years?
I don’t know whether to laugh or throw up on my keyboard. I wouldn’t risk my lunch money, much less the fate of two of America’s most venerable institutions on any analysis based on four years of market data.
Even a starry-eyed MBA grad should understand that the tepid volatility of those previous four years (or any four years for that matter) could not possibly hope to model the financial markets under stress.
Wall Street’s dilemma extends far beyond debunked theories - it is applying debunked theories poorly.
I often hear, “How can you possibly hope to beat the market, to compete with Wall Street? They have the brightest minds, bottomless wallets, the inside edge!”
Stories like this are why.
Happy Trading,
ms
To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.
Filed under: Random Stuff | 9 Comments
Hey Michael, Don here. Just wanted to throw in one more point to make you or anyone else who reads this puke. And that is how the ratings agencies picked and chose volatility assumptions to grade credit derivatives (both mortgage derivatives and others types). Moody’s was so conflicted that when they found a typo in their computer’s rating program that would result in previously rated AAA paper getting downgraded, the staff at Moody’s looked at assuming a lower volatility in order to keep the rating at the safest level.
In other words, the entire credit system upon which mortgages, credit cards, auto loans, etc., was based on volatility assumptions made by these morons. I was able to find what Lehman and Morgan Stanley used. And I was able to find that Moody’s was thinking about making a volatility modification. What I want to know is, what assumptions was Moody’s using in the first place? What was S&P using? What were MER, AIG, ABK, FNM, FRE, BSC and all those other supposedly smart people using?
Some HF’s and institutions are using less than four years of data. It is really not that simple – the books that HF’s and institutions trade are massive and a lot of the time contain instruments with much less market data than four years (think corporate bonds/credit default swaps etc). Some systems that calculate VAR will simply flatline volatility in case of missing market data, so Relative Value strategies will show severely skewed risk profiles. Some prime brokers calculate margin using a fixed correlation matrix and fixed factors, and for HF’s this is preferred method to VAR as it is more stable. VAR can be a very volatile measure in particular if you use exponentially weighted time series to generate volatility and correlations, and this is not really a useful measure as your VAR will spike in volatile times and then mean revert. It would be very difficult to manage your capital requirements using a VAR measure that volatile. A longer term time series such as four years and often with weekly returns rather than daily gives a more smooth time series and thus VAR. This is a more useful measure to use as risk forecast and most market participants know that VAR is only useful in normal market conditions. Many PB’s using the fixed volatility measures are now simply doubling their margins requirements to protect themselves. This is causing HF’s to be further capital restrained in particular with redemptions requests coming from irrational investors.
Having said that, this obviously does NOT justify the reckless practices by the investments banks. There has also been a HUGE conflict of interest between investment banks who created all the CDO/CLO’s and the rating agencies. The fact that rating agencies got paid by the investment banks is just insane. But in the good times no one seems to question this, it is only now that people are pointing fingers and rightly so. There are also some good developments in the CDS area, and hopefully before not too long we will see CME or another large player create a solid exchange that will make CDS trading and price discovery much more transaparent than it is today.
Don/Michael, it is really pretty sickening.
A normal recession/business cycle notwithstanding, these criminals have very easily put America at risk.
Was there an image originally associated with this post, highlighting the volatility trough in recent years? I found it really powerful – why was it taken down?
RE to aaronb: sorry for the head fake. I liked the image too, but after I thought about it a bit, I decided it was intellectually dishonest to include the graph. These guys were speculating in a whole host of assets and to just show equities didn’t tell the whole story. Now, I could have shown many graphs with many different asset classes – that would have proven the same point because volatility in almost every asset class over that five years was historically low. But since I didn’t want to go that route (it would mess up the flow of the post), I thought it was better just to remove it all together. Hope that helps.
michael
But 4 years of intraday data is something like 1.73748 gazillion data points so it has to be valid right?
RE to justdoug: I hope that’s sarcasm I smell =)
Re to Jens: I appreciate the insight offered in your comments. There is one piece that seems out of place however.
Your comment (paraphrased),
‘It is really not that simple – the books …..contain instruments with much less market data than four years. … Some systems that calculate VAR….in case(s) of missing data…..will show severely skewed risk profiles’
seems to suggest that using an alternate less volatile calculation or a fixed matrix for assumptions, is reasonable and prudent when insufficient data exists than would allow for sufficiently accurate assumptions, or when a widely used system is ineffectual with the amount of data available.
Isn’t that one of the fallacies that has exacerbated our current problem?
Defaulting to a fixed matrix because either the preferred system doesn’t work well with the available data, or the amount data is insufficient to model with the desired accuracy, is just as wrong headed as lowering volatility assumptions to maintain a derivative’s credit rating.
I would argue for a position holding that: the portion of a book that is comprised of instruments with much less market data can only be modeled with much less accuracy, and so should only be allowed much less leverage. Otherwise, unquantified risks compound other unquantified risks, which…. leads to our current predicament.
Do you disagree, or was I misinterpreting your comment?