CASTrader Blog

Book Review: The Intelligent Universe - Part II

In Part I, I discussed The Intelligent Universe, by James Gardner.  Specific takeaways for CASTrader, mostly in my words, not Gardners:

• Evolution really is the ultimate computer algorithm.  It has the best hope of advancing the software at a rapid enough pace to take advantage of the hardware, as well as discovery.
• Evolution is pervasive and much greater than Darwin imagined: Gardner thinks the universe is an evolutionary organism making babies with physical laws as DNA.  The human mind is an evolutionary playground of ideas, including mathematics - we are enticed by something to explore it.  Perhaps evolution is the ultimate fractal property of the universe, self similar across dimensional and even conceptual scales.  By extension, financial markets would be expected to evolve.

Book Review: The Intelligent Universe - Part I

At the risk of straying too far off-topic, I'll discuss The Intelligent Universe, by James Gardner, a book I just finished.  This book has little to do with the markets, but certainly relates peripherally to CASTrader in terms of machine intelligence.  It features an introduction by Ray Kurzweil, inventor, futurist author, supplement popper, and hedge fund starter with "80-100%" returns.  Gardner's book certainly sent my mind spinning off in all sorts of directions, imagining all sorts of possibilities about the meaning of the universe and the possibilities thereof.  It is certainly mind-expanding for those who don't think about this sort of stuff often.  In this Part I, I'll review the book.  In Part II, I'll summarize the takeaways for CASTrader.

The book can be divided roughly into two parts, the first part being an exploration of the work and thoughts of others in fields ranging from machine intelligence to religion to extraterrestrial intelligence to cosmology to string theory and quantum mechanics.  It's a good overview, and this is the part that sent my mind spinning. Gardner makes a brave attempt to weave all these together into an all-encompassing theory of the universe in the second part of the book.  The low point of the book was suddenly realizing that I was being given a "sales pitch" on Gardner's theory called Biocosm, whereby "the universe is coming to life," which seems to be a popular meme.  He definately reminds you often of his earlier book on the subject.  The lowest of the low was his very brief characterization of Stephen Hawking's competitive theory as a "desperate attempt to come up with a plausible explanation" of why our universe is "improbably biofriendly."  Do I sense some animosity there?  One can't help but wonder, however, if Gardner's theory is true that the universe and life are so deeply linked, why E.T.s from all over the universe haven't contacted us yet (the Fermi Paradox).   Gardner's treatment of the Fermi Paradox is quite inadequate in my view compared to the treatment on Wikipedia. 

I got the sense that there is an even bigger paradox out there.  Gardner seems to imply via his extensive quoting of Kurzweil amongs others, and his own theory that the ultimate destiny of the universe is for life/intelligence to remake and repurpose it, perhaps into a giant computing machine for the creation of new and better universes in the most fantastic application of Darwin yet.   Kurzweil, meanwhile, would have us believe we'll produce massively intelligent machines within the next 30 years or so.  These machines will in turn produce more intelligent machines it an incomprehensible pace (albeit with an upper limit) and presumably figure out how to harness quantum weirdness and remake the universe to Gardner's vision. If so, given that our solar system is billions of years younger than others, forget why E.T. hasn't phoned home and ask why E.T. hasn't reinvented the future of the universe yet.  Perhaps he has and like some Matrix movie nightmare, we haven't figured it out yet.  Something doesn't quite add up.  The only other takeaway I got from all this is that the SETI project is woefully inadequate and too myopic to detect signs of E.T.

Gardner's conclusions are mindblowing, and I won't ruin the ending, since the book seems to have been designed for the "big ending."  The most mind-blowing thing for me, however, was the odd sense this book gave me that something very big is going to happen this century (be it good or bad), not just because of Kurweil's influence, but also because of the work of others that Gardner cites as well.  Gardner even dredges up that Isaac Newton, arguably the father of physics, apparently concluded Armageddon is due in the year 2060.  If Gardner intended to give his readers this feeling, it's subliminal, but he adds in plenty of scientifically inspired doomsday scenarios to stir the pot.  Maybe he didn't want to get too metaphysical and say it outright.  Sometimes he brushes away these doomsday scenarios with beliefs that we will "live on" while also going to great lengths to point out the insignificance of mankind in the overall plan (we are but a minor subroutine in the cosmic computer).  In the afterword, he ends with an an argument that like a butterfly is enticed to a flower, our minds are being enticed by something to discover the laws that govern the universe - enticed to our doom, or something better?  You can sense Gardner saying "Dammit! the universe is not about us, and neither is this theory!" while at the same time saying the universe is all about intelligent life and struggling to explain why we Earthlings appear to be the only intelligent life.

Fortune's Formula by William Poundstone - Part III, Useful tidbits applicable to CASTrader

In Part I, we reviewed Poundstone's book, Fortune's Formula, and in Part II attempted to answer whether Fortune's Formula (aka the Kelly Criterion) really is the best way to manage money.  In this final part, I'll summarize what parts of Poundstone's book are applicable to the development of CASTrader. 

To Kelly or not to Kelly.  The CASTrader platform, being composed of many autonomous complex adaptive agents is the perfect environment to apply the Kelly Criterion.  I personally have a utility function, but the thousands (millions?) or more trading agents that will eventually "live" in CASTrader won't have a utility function unless I give them one.  In other words, they won't care about drawdowns or get psyched out by a losing streak unless I program them to - and right now I see no reason to.  Their sole purpose in life is to maximize compound growth, which is exactly what the Kelly Criterion achieves.  Therefore, the Kelly Criterion will be a key component of the CAS Trader platform for now.  Ideally, many of the traders will trade frequently, allowing the Kelly Criterion to work it's magic. My assumption is that the diverse set of agents will be largely uncorrelated and thus mute the Kelly volatility and drawdowns so that the overall community portfolio returns do not violate my personal utility function.  This will hopefully achieve the Holy Grail - Kelly optimal growth without the ulcers.  I'm betting that there will be no need to limit individual traders to a fractional Kelly system.

Constant-proportion portfolio. Claude Shannon once advocated a constant-proportion rebalanced portfolio, which is a special case of the Kelly Criterion.  Such a portfolio would keep say 50% in cash and 50% in a particular stock rather than putting everything in the stock, and you rebalance frequently as the stock changes in price.  Shannon dismissed the idea due to commissions, but recently Mark Rubinstein and Eugene Fama, unaware of Shannon's unpublished work, have apparently published research that the optimal portfolio is always a constant-proportion portfolio.  Unaware of either Shannon's and Fama's work, I independently investigated the constant-proportion rebalanced portfolio several years ago and it is close to how I manage my personal money now.  In some future post, I'll detail my findings and experiences.  I figure I get 1-4% additional return over a straight buy-and-hold strategy using this approach.  On certain stocks, the returns from this approach can be 30% even if the stock returns 10% or less.  The approach works by profiting from volatility - a kind of volatility arbitrage.  Needless to say, this was an interesting part of the book for me.  Variations on this approach will be incorporated into CASTrader.

Fortune's Formula by William Poundstone - Part II, Is the Kelly approach really the best way to manage money?

In Part I, I reviewed William Poundstone's book, Fortune's Formula, which discusses the Kelly Criterion.  The great part of the book for me is the detailing of fight between the Efficient Market Hypothesis (and Nobel Prize winning) Robert Merton/Paul Samuelson crowd who call the Kelly Criterion a "fallacy" and the Thorp/Shannon Pro-Kelly camp who have astonishing investment returns that appear to simultaneously validate the Kelly approach and disprove the Efficient Market Hypothesis.

Most money managers today are using the Markowitz Criterion, as Pounstone calls it.  Michael Mauboussin of Legg Mason writes a great expanded summary of the Kelly fight, and also explains why "modern" money managers don't typically utilize the Kelly Criterion.  The short answer: Kelly is only good when applied over and over for a long period of time and you reinvest your proceeds to achieve compound returns, and Kelly losses and volatility (during unlucky streaks) are hard to deal with psychologically.  Neither of these features is suited to most money managers who have a short-term focus because they risk losing assets otherwise, therefore most money managers accept the mediocre returns (over the long run) of the mean/variance (Markowitz) approach.  Read Mauboussin after you read Poundstone's explanation.  It is interesting that Poundstone and others speculate that Warren Buffett and Charlie Munger are closer adherents of the Kelly approach than the Markowitz approach.

There appears to be no dispute that over the long haul, the Kelly approach is mathematically proven to outperform all other strategies in terms of total return.  Samuelson's critique of the Kelly approach boils down to the fact that Samuelson believes the Kelly approach is not consistent with the utility function of most investors.  Many investor's utility functions will not tolerate the inevitable drawdowns and volatility of the Kelly approach.  Utility functions, however are somewhat subjective.  The pro-Kelly camp believes that maximizing the compound annual growth rate of assets is the most rational approach (utility function) in the long run.  Others disagree. Another question is how long is the long run?  Morten Christensen gives excellent details in a long but mostly readable and up-to-date paper titled On the History of the Growth Optimal Portfolio.  This is an excellent read for those desiring a lot more technical details than Poundstone provides.  Recent theoretical studies show that the Kelly Approach can require an investing lifetime or more before it can be said to beat other "safer" approaches with a 95% confidence level. 

The answer of whether the Kelly approach is better, in my opinion, is that it depends a lot on your situation.  As Poundstone points out, it is probably more suited to traders than individual investors. In Part III, I'll discuss some other tidbits from the book and whether the Kelly criterion is appropriate for CASTrader.

Fortune's Formula by William Poundstone - Part I, Review

The magic formula in William Poundstone's book, Fortune's Formula is the Kelly Criterion, named after John Kelly who discovered it while working at Bell Labs.  Want to maximize your capital growth in any repeated game of chance where you have an edge you can calculate?  The surprisingly simple Kelly Formula tells you how much of your bankroll to risk on each bet, and in doing so, saves you from bankruptcy in the process if applied correctly.  Poundstone basically summarizes the Kelly Formula on one page, and many gamblers and even some investors are already familiar with the Kelly Criterion - so why the 386 page book?  The answer is two-fold:

First, Poundstone details the lives of John Kelly and his colleagues at Bell Labs: Ed Thorp and Claude Shannon.  The story of Ed Thorp applying the Kelly Criterion (plus card counting) to the blackjack tables in early Las Vegas as well as the world of hedge funds (and cleaning house in both cases) is particularly fascinating, as is the the life of Claude Shannon, the father of information theory which is the very foundation of computers and the internet.  Just as fascinating is the link between these three and various individuals/entities such as the mob, Micheal Milken, Ivan Boesky, Fischer Black, Robert Merton, Myron Scholes, Paul Samuelson, LTCM, Rudy Guiliani and even Warren Buffett.

Second, Poundstone details the outright hostility towards the Kelly Formula from the financial scholars, including Nobel prize winners Merton and Paul Samuelson.  Paul Samuelson goes so far as to call the Kelly Criterion a "fallacy," although he technically does not disagree with the mathematical proof of the formula, rather he disagrees with it's application to the real world of finance, pointing out that while the Kelly Criterion will not make you go broke, it can result in serious drawdowns.  Indeed, search around the internet, and you will find many gamblers and horse track bettors who seem to utilize the Kelly Formula but will tell you the same thing.  Many of these people advocate using "1/2 Kelly" which trades less capital growth in return for much reduced drawdown during the inevitable unlucky streak.  The explanation of the catfight between the Pro-Kelly Thorp/Shannon camp and the Anti-Kelly Merton/Samuelson camps is worth the price of the book alone.  It's fascinating how Kelly Criterion fight seems to be a proxy for the Efficient Market Hypothesis (EMH) battle.  Poundstone details how the amazing investing returns of Thorp and Shannon make a mockery of the Efficient Market Hypothesis.  In fact, Poundstone details how the Kelly Criterion (which uses geometric mean) is superior to the darling of the EMH crowd: Harry Markowitz's Modern Portfolio Theory (which uses an arithmetic mean).  More on this in Part II.