What's the best way to learn portfolio optimization? by Laurent Bernut
Answer by Laurent Bernut:
In theory, theory and practice are the same. In practice, they are not. Most market participants optimise the wrong thing for the wrong reasons.
Modern Portfolio Theory was revolutionary
Half a century Modern Portfolio Theory was revolutionary in its days, that was before they sent someone to the moon, and Nikita Khrushchev became the Soviet Supreme, and Fidel was kicking Batista’s ass.
It made a couple of assumptions that were good back then but that have plagued generations of traders. It equates volatility with risk. Volatility is not risk and risk is not necessarily volatile.
LTCM had low volatility and so did all the Volatility fund (short Gamma funds) back in 2008. They had exceptionally low volatility until the day before the blew up… Similarly, Paul Tudor Jones, Ed Seykota and a lot of the major CTAs have high volatility but no-one will argue with their returns…
Sharpe ratio, Jensen alpha, Treynor and all that jazz are the right mathematical answers to the wrong question. I wrote a post about that. Conclusion: i actively de-friend the savant clowns who ask for my Sharpe.
Second assumption is linearity of risk. This one is not apparent. Mean variance is mathematically elegant, but it assumes strategies work throughout the cycle. Mean variance is observed through the period and when bad, go back to the drawing board. Old school academic laziness
Example: small caps growth do extremely well at the beginning and at the end of a bull phase. Rest of the time, they oscillate between volatile and outright dangerous. Mean variance optimisation would triumphantly conclude small cap growth is a bad strategy. Good luck explaining that to the cohort of managers baby sitting billions on small caps
What to optimise then
Trading edge = Win% *Avg Win% -Loss % * Avg Loss%
Optimisation and back-testing serve three purposes:
- Kill your baby: the earlier you disprove your idea, the less time you waste on it. So, electrocute, drown and punch your baby until it either dies or survives
- Identify when a strategy works and when it stops: no strategy works all the time. Laziness is to optimise so as to make it fit through the cycle.
- Identify design flows: dropping all the indicators, factors into the masala smoothie blender and press play hoping something will stick is unforgivable academic laziness. Optimisation will give a correct mathematical answer to a myopic wrong question.
Torpedoing decades of layered platitudes probably requires a little more substance.
- When: There are times when buying penny stocks or IPOS are spectacular strategies. Other times, it is suicidal. Difficulty is to know when. Optimisation will have you size small or discard the strategy altogether. Buying penny stocks is like wearing bikinis: it is a bad idea to insist on wearing them through the winter
- Design flows: many optimisations go equal weight on all factors and equal weight position sizing. When it does not spit out the right answer, add another factor. Sounds familiar ? That is one bad “bad idea”. Design flows are about subtracting factors, simplification, not addition of complex fragility.
How to optimise then ?
- Segment time periods: run through very long period, then segment by market regime. Run optmisied series through various regimes. That will tell you when regime has changed. The biggest mistake people commit is to re-optimise when strategy stops working. If you have identified when it does not work, then it is easier to accept
- Universe: run large, segment and then large universes again. Few strategies work across large universes. Conversely, humans are notoriously bad at defining the scope of the strategy. If your stuff is mean for small caps, run it for large caps as well. It will give invaluable perspective
- IMPORTANT POINT HERE: Long Short strategy conception starts from the short side and then considers long side. My personal favourite mistake of all times is the demise of quants in the summer of 2007 (this is a chapter in the upcoming book). They went Long quality and short bad quality, logical right ? They assumed shorts were the inverse of Longs. Everyone buys Quality of course, but they failed to notice that no Long holder was selling bad quality anymore. Yet, in order to stay market neutral, they had to short even larger quantities of stocks that no Long Holder owned anymore. Shorts became illiquid and hard to borrow. Risk management as in days to liquidate became an issue, so they had to cover which triggered chain reaction short squeezes. They lost -4-5% in seemingly calm markets, when they were making +0.5%. Investors started to redeem. Those guys were leveraged up to the hilt. VAR went up so PBs rquired more collateral which brought down leverage. Redemption + lower leverage made returns look even more underwhelming. Game over. (Oh and by the way, they had low volatility and excellent Sharpe until they really did not)
Personally, i optimise for different purposes than most people. I optimise for buying power. Performance is not a function of stock picking , it is a function of position szing (i am picking a fight here). Problem becomes a buying power or concentration issue. This type of optimisation looks at exits, sizes of exits, position sizing. Answer is not easy as it searches for regime disruptions. That is too tough of a problem to visualise. I need the machine to show transitions.
Sorry if i appear pugnacious today (might also be those stiff cocktails they serve in NYC). MPT did a great job. It showed the way, thank You very much, but now it is now time to innovate and move on. The problem is that it requires a different way of thinking, involving both hemispheres of the brain.