cusum, structure change, idiosyncratic risk
February 5, 2010
Northfield used an interesting example (river nile v.s. aswan dam) to explain how they identify structural change through performance time series of fund manager ( a lot of assumptions here, but on balance I mean the idea capture the essense). Using 1-factor risk model and assuem excessive risk prepresent everything else (using a statistical multifactor model will be more accurate but…) I copied the idea here and there is some interesting finding:
Google’s cumulative execess return over QQQQ has been flat since 2006, one year after its IPO, according to wiki the buying mainly came from individual investor:
while Apple’s cusum took off some time in 2004:
Interstingly, ipod sales chart indicated that its sale start to took off ( 4G and first mini(my first as well) hit the mass market) in 2004 as well.
ipod sales (250 m cumulative uptodate) has been tailing off since 2008, but this has been substituted by sales growth in iphone:
a particular inflection point in Q$ 08 in cusum data coincide with strong Q4 08 iphone sales.
the bias in idiosyncratic risk should be slow moving and more closely reflecting the underlying factors of a particular company. should this be a better bet?
the caveat of using cusum is the effect of leverage, the cusum will be positively biased in bull market and viceverse, this effect is especially pronounced in small cap. In this case, the statistically significant ‘structural’ change is more difficult to spot, taking BITSTREAM for example:
At least based on the public information reflected in price, it’s safe to say no structurally change has happend to this company yet:
in case of IBM, it’s a story of reinvent from hardware manufacturer to technology consulting. SP 500 is not a perfect proxy for systematic risk, but a little reassuring is the fact that residue show little correlation with market.