Online course of the day, investing department

By Felix Salmon
November 21, 2012

Would you like to take a free online university course which teaches you the basics of quantitative analysis and also helps you manage your money so that you get high returns with low risk? Of course you would. Let me introduce you to Computational Investing, Part I, taught by Tucker Balch, Ph.D., on the Coursera website.

Under “Recommended Background” we’re told that “the primary prerequisite is an excitement about the stock market”. And there are two recommendations under “Suggested Readings”, including All About Hedge Funds : The Easy Way to Get Started, by Robert Jaeger. (Apparently it “explains how any investor can take advantage of the high-potential returns of hedge funds while incorporating safeguards to limit their volatility and risk”.)

This is a genuine university course: it’s the same one that Balch teaches at Georgia Tech. And so you’d expect a few disclaimers, at least, along the lines of “this is an introductory course, it’ll help you understand a few concepts, and maybe be the first step on the road to becoming a quantitative analyst yourself one day, but please, kids, don’t try this at home”.

You might expect such a thing, but you’d be disappointed. Instead, you get the exact opposite. Check out Week 4 (you might have to register; it’s easy and free) and then “Lecture Video 1.2: Response to Questions from Students”. According to Balch, the “number one most popular question” he gets asked is “Do I use these techniques to manage my own funds?”. He responds as forthrightly as he can:

The answer is yes.

Balch continues:

I have a number of different investments that I use these approaches for. With regard to my company, Lucena Research, we manage a few small funds as a way to test our techniques and validate them. One of them in particular I’ll show you in just a moment.

It’s far from clear how a student who has merely taken an online course might ever hope to replicate the returns that Balch manages to generate at Lucena (“Hedge Fund Technology for the Strategic Investor”). But in any case Balch does share with us a Lucena portfolio which “was developed specifically to be low risk”. It looks like this:

I look at this and I immediately get suspicious: there’s something quite Madoff-like about the way in which Balch’s returns go steadily up and to the right regardless of what the stock market is doing. Here’s how Balch explains what’s going on in there:

This approach was developed specifically to be low risk. It includes a basket of less than 20 equities that are traded about every 2 weeks. It’s 2X leveraged, meaning that half of the money is borrowed investment.

So this approach is a 2X levered fund with less than 20 stocks? Sounds very risky to me. But Balch shows us the numbers to prove that it isn’t:

The first thing to note here is that although Balch told us he was going to show us one of the “small funds” that he uses “to test our techniques and validate them”, this does not look like a real-money fund. There’s no indication, for starters, of what the borrowing costs are: if the fund is indeed 2X leveraged, how much does it cost to borrow $10 million on an ongoing basis?

Maybe those numbers are somehow incorporated into the returns — but then there’s the very odd section on “Transaction Costs”. The commissions bit makes sense: if you trade 10 times a week on average for 20 months, then that’s about 860 trades in all, and the commissions add up to about $20 per trade.

But then there’s the “slippage”, which doesn’t make sense. Commissions are real costs: they’re the amount of money you have to pay your broker to execute your trades. Slippage, on the other hand, is not a real cost, but rather a theoretical cost: it’s the difference between the official market price of a security, and the price you actually end up paying. It’s a way of taking a theoretical portfolio, which always trades at the market price, and adjusting the returns to make them more realistic. If you have a real portfolio, as Balch suggests that he does, then there’s no “slippage”: the slippage is built in to your actual returns.

So it seems that Balch, after promising to show us the returns that one of his “small funds” has generated, ends up doing no such thing. (And also, I don’t think that a $20 million fund would count as “small” for a college professor who tells us that most of his money is in his TIAA-CREF retirement account.) Still, he says:

This is a conservative approach which nets about 15%-20% per year. You can absolutely follow more risky approaches that’ll provide higher returns. This is the kind of approach I follow.

In other words, if you take what Balch is saying at face value, he’s managed to come up with a conservative investment strategy, which is levered 2-to-1, which generates returns of more than 15% per year, which he follows himself. And he encourages his students to try to do the exact same thing.

There are lots of courses on Coursera, and most of them aren’t as sketchy as this. But I do think that what we’re seeing here is the beginning of a serious problem with online universities like Coursera: you can never be sure about their quality control. And in general, if you’re taking a college course where the professor encourages you to lever up a small number of stock-market investments in the hope of getting low-volatility 20% returns, I’d advise thinking twice about that professor, and that course. Because it just doesn’t pass the smell test.

More From Felix Salmon
Post Felix
The Piketty pessimist
The most expensive lottery ticket in the world
The problems of HFT, Joe Stiglitz edition
Private equity math, Nuveen edition
Five explanations for Greece’s bond yield
Comments
13 comments so far

The conclusion here is off: Balch is primarily representing Georgia Tech’s brand; it’s Georgia Tech’s brand that lends his teaching legitimacy and allows him to access the Coursera platform. If his class is dodgy then Coursera should be policing its higher ed partners, whose job it is to vet faculty in the first place.

Posted by dlinde | Report as abusive

Coursera would probably welcome Professor Salmon to teach a course ..

Posted by Woltmann | Report as abusive

As anyone who’s attended a non-elite university can attest, there’s very little in the way of quality control at brick-and-mortar schools, as well.

Posted by coreymull | Report as abusive

Did you dig into the strategies he’s actually proposing? I’m guessing it’s just some weird model trained on historical data and then simulated against that same data.

Posted by absinthe | Report as abusive

Agreed. I’d subscribed to the course and dropped after the second week. The guy just looks like an artist.

Posted by Alex314159 | Report as abusive

There were, and maybe still are, college adjuncts teaching dodgy real estate extension seminars to adults promising great returns without disclosing that the teacher gets a finder’s fee if a student buys a property from one of the companies touted during the class.

Posted by logicus | Report as abusive

If he could net 15-20% per year, what the heck is he doing still working as a college professor?

He should either be rich enough to have retired, or have started his own hedge fund.

Posted by mfw13 | Report as abusive

Not that I disagree, but some possible counterarguments:

*He could be getting leverage via a leveraged ETF/Mutual Fund without margin costs. It could also be leveraged via the use of options.

*Slippage could refer to the underlying model’s price and the actual price received when the trade was executed.

*$20 million may be small once you take into consideration operation costs. Based on my limited observations, it seems the cusp of solid returns lie somewhere north of $20 to $30 million AUM (and again, this is just from my limited observations).

This all being said, some things I would also ask Professor Balch:

*While the returns may be low risk now, are they low risk always (i.e. does the underlying model adapt to different market regimes) This is akin to cherry-picking a time period to run a model off of.

*Beta/Sharpe/Correlation are all useful measures for normally distributed returns, but unfortunately the model’s returns are likely not normally distributed. If you want to see how this works, take historical market returns and square them; compare the correlation between the returns and the squared returns. It’ll be close to zero, but it is hard to say there isn’t market risk in the model.

*Given the above, how well does a non-linear model, such as local regression (LOESS), fit the data (and thus indicate market risk)?

Sorry for the lengthy post; just thought it may be useful to contribute some more points on both sides.

Posted by finn0123 | Report as abusive

I am in that course. I took it as a skeptic, and Taleb reader. I think Balch has touched on what you have to believe, to believe in the methods. (Lecture Video 2.2 : Efficient Markets Hypothesis) He has not told everyone to forget it and buy index funds, of course. It will be interesting to see where it leads

The class is not as finished and polished as An Introduction to Interactive Programming in Python, also at Coursera, but it’s the first run of Balch’s class.

As an aside, Balch says he is a former fighter pilot, which makes me think from the get-go that he has a different risk profile than me.

Posted by jpersonna | Report as abusive

oh thank you so much for this!! i signed up for this course a few weeks ago and immediately thought these very same things.

this is not a finance course.

this is a computer programming course with a minor focus on financial applications.

there are no disclaimers about the risks involved in implementing any of the strategies or ideas he talks about. the concepts related to finance and the financial markets are merely glossed over as unimportant details. in one lecture he even goes so far as to say that markets are efficient because of high frequency traders.

the course really feels like it’s being taught by someone who had a lot of success with machine learning algorithms and thinks he’s absolutely conquered the world of finance, and he’s here to teach you how to get rich quick with these techniques!

as someone who has studied finance in college and grad school and has been involved with the industry for several years, the way these topics are being approached in this course is an absolute travesty. i was worried a few weeks ago about the nonchalant style in which the materials were being offered. i’m glad i wasn’t the only one who thought it was problematic.

Posted by angels13 | Report as abusive

It looks not entirely unlike the “Computational Finance” course I took at LBS in the 90s with Apostolos Refenes and Neil Burgess. That was a good course, but they were pretty clear on the limitations.

Posted by dsquared | Report as abusive

I don’t get your complaint, or what metric you are using to determine whether Balch is teaching a good/shady course or not. If it is an educational metric, and learning concepts, it seems like Balch is just doing that by using a compelling example. If it is to duplicate a portfolio, perhaps not by using the exact variables, but there are many educators who use examples that illustrates points better than an exact example. But he does seem to give you the feel for variables that matter, isn’t that what the course is for?

From reading *your* article Blach course seems like it an exciting course that use examples that draws you in.

Posted by MijaMoja | Report as abusive

Hi Felix, Your post raises some provocative questions. I’m glad to have an opportunity to respond.

You focus on a lecture in which I am responding to student questions 3 weeks into the course. Here is some context:

Engagement is one of the key challenges in teaching a MOOC. It’s much tougher than in person teaching. In order to build that engagement I invited the students to post questions in the course forum and to vote for the questions they were most interested in. I promised to answer the 10 questions with the most votes.

The question with the most votes by far was “Do you manage your own money using computational investment techniques?”

This is not a topic I planned to address in the syllabus. However, the question is fair enough, and I felt it deserved an answer. You raised some questions about the details of the strategy I described, and I’ll address those further down. But the point here is that this was a response to questions from the students.

With regard to goals for this course: The course is not intended to provide comprehensive coverage of quantitative techniques. It’s intended to offer an introduction to the most important topics (CAPM, EMH, risk/reward, survivor bias) and to provide some hands-on experience with historical data. The goal is to spark interest with the hope that some students will carry that forward to deeper study. I think that is pretty clear from the course description materials. I do not recommend or suggest that anybody rush out and start managing a hedge fund on the basis of this course.

Also, the course is not meant to be a replacement for the course I teach in person at Georgia Tech. The content represents only about 1/3 of the course I teach at GT. We do not provide course credit for completing this course.

You criticized the recommended reading “All about Hedge Funds” by Jaeger. Remember that one goal is to make the subject accessible, and Jaeger’s book provides a readable introduction to many of the details of the industry. You didn’t mention my other recommendation, “Active Portfolio Management” by Grinold and Kahn. This is a substantial tome viewed by many as a standard reference for portfolio management. I think it would have been fair to mention both.

You go on to comment on the presentation of a strategy I trade. And you make some good points.

Let me first be more specific about what is depicted. The chart and analysis are a back test of a strategy simulated since January 2011. The back test simulates a $20M initial investment at 2X leverage. The strategy has been traded live with a more modest sum over the last 4 months. Return over that period is 2.7% (without leverage). We plan to lever up soon.

With regard to slippage: You are correct that in practice this “cost” is built into the results. The slippage value reported in the chart is an estimate provided by the simulation.

Best regards,

Tucker Balch

Posted by TuckerBalch | Report as abusive
Post Your Comment

We welcome comments that advance the story through relevant opinion, anecdotes, links and data. If you see a comment that you believe is irrelevant or inappropriate, you can flag it to our editors by using the report abuse links. Views expressed in the comments do not represent those of Reuters. For more information on our comment policy, see http://blogs.reuters.com/fulldisclosure/2010/09/27/toward-a-more-thoughtful-conversation-on-stories/