There is an interesting article on TechCrunch about the author’s advice to a Facebook employee on how to manage his newly liquid wealth. He believes software will beat the active investor 99% of the time for the following reasons.
Will the active investment manager go the way of the MTV VJ?
I don’t see how they can fully replace humans. What if another flash crash occurs and you have everyone running a rules based trading system? How does a system handle movements not attributable to market mechanics? I’m very skeptical of this.
I find it interesting that there is no discussion of what type of software to use. Are they really all the same? It doesn’t matter what the software is, as long as “it’s software”?? You might as well say, “it doesn’t matter who your portfolio manager is, as long as he’s white.”
While the points about how software doesn’t get scared and tricked into the behavioral biases make a lot of sense, I still think:
“To err is human; to bring the S&P down by 40% over the course of 15 minutes requires a computer”
Ultimately, someone has to be there at the on-off switch, and that guy has to be able to throw the switch at the right time.
who makes the computer isn’t it human? The point here is that the computer can have a flaw that continues on making the error the design had at inception.
another is with the computer flaw, the computer can’t make you your money back with a loss.
Computers may fail to spot opportunity.
If everyone uses a computer to auto-trade then they all gonna do the same thing!
I think computer can on some days beat the human in trading but it’ll be rare because humans understand the other humans’ greed 
I think in some way it is possible. It just takes a super computer that today is the size of maybe 3 acres. The computer must be able to formulate its own logarithms that allow it to evaluate the probabilities of future events given today’s circumstances. Some computers have been predicting major catastrophes lately - they are called ‘crawlers’ or ‘spiders’ or something like that. Given a powerful enough computer, I think it’s possible to predict likely events to a high degree of certainty.
Also the computer has to be capable of self-evaluation and self-teaching. Although that is possible today with AI, it has its limits.
And as the second poster said, if everyone uses computers than there would be no arbitrage. In essence we would create the first perfectly efficient market, and therefore no one - not active traders or computer would be able to profit.
I’d trust the computer compared to the run of the mill CFA. Fear and Greed. I focus on firm asset allocation. A computer will hold one to said allocations as opposed to a PM who ‘knows’ gold is about to spike.
A computer will also force once to allocate even when the market is crashing, which essentially means you’ll be buying as the prices go low. Furthermore, the computer will sell when the prices go high to maintain said allocations meaning you’ll sell high. Isn’t that what a good investor does? Transaction costs and tax implication aside, I feel a solid program maintaing fixed percentages in asset classes would do very well over the long haul.
I actually agree that software systems have a lot of advantages when it comes to investing.
I just think that the article is a shoddily presented argument in favor of it.
I think the author is using software to refer to programs such as Wealthfront which are basically asset allocation models that claim to be along the efficient frontier using ETFs. You basically tell the system how much risk you are willing to accept and the system selects the portfolio at that point along the hypothetical efficient frontier.
This seems very reasonable but how many of us actually believe in the EMH?
It seems if computers are trading, one with enough knowledge of how they invest could exploit that by taking advantage of things it’ll miss; however, it may take forever for the market to accurately price it due to the software always overlooking it.
its possible…so long as the computer sticks to simple rules (low p/e stocks, etc) and not rely on some new discovery…
The efficient frontier is different from efficient markets. You don’t have to believe in efficient markets to believe that the efficient frontier is where you might want set an allocation.
Efficient Markets = information is already reflected in current prices, so no need to do intense analysis on expected returns.
Efficient Frontier = the portfolio is allocated so that you can’t get get any more return for the same amount of risk, and can’t take any less risk for the same amount of return. (risk defined as portfolio volatility)
If you think that markets aren’t efficient, you may still have some way to generate expected returns (it’s called securities analysis). Once you have that, you can still decide to choose a portfolio allocation that is on the efficient frontier, given what your analysis of inefficient markets tells you about expected risk and return and correlations.
You are confusing the efficient frontier with the idea that if markets are efficient and the other assumptions of CAPM hold, then the market portfolio is on the efficent frontier.
I think this article misunderstands how traders make money. Most traders do not make money by making genius trade choices. They make money by knowing when and how to screw their customers. Can computers buy pizza for their clients and bribe them to forget that their commission is higher than a competitors’? Can a computer call a client and keep rambling about local sports teams until the client finally agrees to place an order?
In a zero sum trading game, computers might beat humans on average (although, 99% is a pretty high estimate). However, other traders are not the only source of profit.
If you have FB type money and you are relying on standard mean variance optimization like Wealthfront then you are doing something wrong.
If I had FB type money I’d invest in T-bonds, large cap blue chip stocks and real estate.
So you are talking about seperating security selection from asset allocation, whereas the invalidity of EMH will allow one to benefit from security selection they would still benefit from an AA that was mathematically derived. Is that right?
Alot of people would argue that the assumptions underpinning CAPM and Modern Portfolio Theory are also incorrect. Now I kind of like MPT because it means I can use a mathematically driven optimiztion to determine the appropriate asset allocation. This is good for me beacuse it professionalises the investment process, removes some tough decisions and provides me an analytical framework upon which I can justify my decisions. The only problem is that coming from a fairly dogmatic value oriented background I have some fairly big reservations as to wether MPT is actually true or not.
I may be totally out to lunch on this, but my biggest reservation about MPT and mathematically driven optimization like black litterman and monte carlo is that it turns everyone into closet indexers. If people are paying for active management or expecting alpha, how can investment managers achieve either if they all dogmatically follow modern portfolio theory?
Regarding security selection and asset allocation, yes, that’s a way to look at it.
I’m totally on board with the assumptions underpinning CAPM being out of sync with reality (although how out of sync versus alternative sets of assumptions can be debated). I’m not sure what assumptions of MPT are incorrect. The way most MPT theories are applied tend to assume that returns are normally distributed, which is indeed untrue, but I don’t see that that is an essential assumption of MPT. You don’t have to measure risk as covariance in MPT, it’s just that the first presentation of it by Markowitz did, and many people just run with that version.
You don’t have to be a closet indexer. I actually think you could take a value approach like FrankArabia’s for security selection and then put it into an MPT optimized portfolio. I personally do something a little different because I find that correlations are often unstable and so I often use volatility weighting over true MPT optimization with covariance, but it still respects MPT’s claim that portfolio risks do not add up in the same way that portfolio returns do, and therefore there is an optimized weighting that maximizes return for unit of risk.
It’s also true that the MPT portfolio optimization framework does not insist on covariance as the measure of risk. In theory, you can have different measures of risk and optimize based on those metrics instead. Downside deviation, VaR, or any number of risk measures. The only real requirement is that there has to be some algorithm for calculating the total (estimated) risk of a portfolio consisting of two or more assets. Covariance is the simplest and most common measure, but its not the only measure permitted by MPT.
MPT does have problems to the extent that there are multiple sources of risk that cannot be boiled down into a single measure. You can try to fix this by fiddling with assumptions about the utility function, but ultimately that ends up being an exercise in choosing a utility function that gives you the answers you like.
Nice!
What software do you use for your portfolio optimization?
Also I need to work this out in my head, if you pick say 15 stocks on a value basis and throw them into an optimized portfolio what is the end result since the optimization isn’t driving the security selection. Would it simply adjust the weights based on whatever I choose as my risk metric (which in my case would be downside deviation)?
I’d like to see what the output/results would be.