• C++ Programming for Financial Engineering
    Highly recommended by thousands of MFE students. Covers essential C++ topics with applications to financial engineering. Learn more Join!
    Python for Finance with Intro to Data Science
    Gain practical understanding of Python to read, understand, and write professional Python code for your first day on the job. Learn more Join!
    An Intuition-Based Options Primer for FE
    Ideal for entry level positions interviews and graduate studies, specializing in options trading arbitrage and options valuation models. Learn more Join!

Review of Jim Liew's Stat Arb Class

I’m currently a student at Baruch MFE and had the pleasure of taking Dr. Liew’s 5-course statistical arbitrage lecture series in Baruch’s inaugural Time Series Analysis and Algorithmic Trading course. The purpose of this post is to give readers a student’s perspective of what I believe to be one of the best quant courses on Wall Street.

Dr. Jim Liew offers this intensive statistical arbitrage course to students at Baruch MFE, Columbia MFE, and JHU MSE. He also has made this course available to industry practitioners for free, given that they complete a statistical arbitrage project for him after the course is finished. Dr. Liew received his PhD in Finance from Columbia Business School and has been an industry practitioner of systematic trading strategies ranging from macro to ultra high frequency.

Dr. Liew began the lecture series in earnest, giving background into statistical arbitrage, including its history, current state, and a top 10 list of the delusions of statistical arbitrage novices. In retrospect, I should have plastered this above my computer monitor because I personally experienced nearly every one of them without remembering Dr. Liew’s prophetic advice. Right off the bat, Dr. Liew let us know that his course would change our entire career paths, pending that we worked harder for it than we’ve ever worked in our lives. Students were to research and backtest strategies, incorporating transaction costs and volume constraints. Dr. Liew provided some initial direction, and a pinch of “secret sauce”, but ultimately the strategy and its backtesting framework were up to our own discretion. PCA, machine learning, neural nets, technical analysis, fundamental analysis, s-scores, and pairs trading were all suggested as fair game.

The first project we were given was creating a momentum strategy, trading on daily bars in 40 different markets. Dr. Liew wasted no time getting us up the learning curve, assigning the first of three projects during the first lecture. All of the teams would be ranked by highest sharpe ratio and grades would be diviid out accordingly. We were told that we could work with each partner on only one of the projects, which forced us to work with people who we wouldn’t normally have chosen. We were told we had two weeks to work on it, picked our own teams of two, and were off to the races.

The environment of the class was already highly competitive, which anecdotally led to terse phone conservations beginning in “what’s your sharpe ratio!?!?” breaking out among colleagues. I loved this aspect of the class, but there were some who had complaints about this intensity. Personally, I chose to have no partner for this first project in order to develop my own Matlab backtesting framework. I put in 120 hours of work during these two weeks and ultimately got the top sharpe ratio in the class, a measly 1.86. I believe Dr. Liew intended for the first project to provide us with a backtesting framework for the rest of the course, as well as a lesson in overfitting and optimization. Now I know what having 100+ parameters, no out-of-sample period, and limitless capacity can do to inflate a lackluster strategy.

The second project was using daily bars from SPDRs to create pairs under a mean reversion framework. For this lecture, Dr. Liew provided some “secret sauce”: the Avellaneda and Lee strategy from their 2008 paper. At first the model seemed easy enough to code up, but my team soon realized that keeping track of a universe of 100,000+ pairs was no trivial task. Also, as this time we were to run the strategy out of sample, the parameters needed to be endogenized. This made simulations significantly more computationally intensive. The project was due in the middle of Spring break and I had already booked train tickets home so I battled severe eye strain as I tried to finish the project on my 12 hour train ride the day before it was due. Ultimately, my team failed to endogenize enough parameters and we ended with a 0.50 sharpe ratio. Knowing that the third project would make this one look like cake, I knew I had to step it up.

For the final project we had three weeks to build and backtest a strategy trading anything we wanted on minute bars. There was an imposed volume constraint of 5%, a $1 ticket t-cost, and a $0.003/share t-cost. The final class was to be a presentation in front of a mock investment committee of funders of statistical arbitrage strategies, as well as executives at leading desks. Dr. Liew kept reiterating that this was our “shot” at breaking into statistical arbitrage. Any one of these practitioners could easily bring us on board to their respective desk. My team (David Rappaport, Michael Lwin, Yike Lu, and myself) was united in one goal: cranking the project and obtaining our dream jobs.

Before my team had even truly begun the project, Yike set up a Q/KDB database hosted on an Amazon EC2 cloud. My team spent a couple of days brainstorming strategies to get a sense of direction for the project. Using my improved Matlab skills, I developed a vectorized backtesting framework in which I fast-prototyped technical analysis strategies and extensions during the full course of the project. Ultimately, we all contributed ideas to improve the single-stock counter-trend strategy which I had discovered during this process. David developed a limit order framework which met the rigorous constraints Dr. Liew had assigned and Michael researched our market impact model.

We worked on the project every day for three weeks straight, without fail, for an average of 13 hours per day. After pushing us harder than we had ever worked in our lives, Dr. Liew’s promises held true. We received the highest scores from the mock investment panel, with the accompanying prize of getting to pitch to one of the top high frequency desks in the world. Our presentation can be viewed in the attached document. My team has since been contacted for interviews at other leading desks. Dr. Liew’s class inspired many people to seek positions on the buy-side and now, thanks to his class, we have the avenues opened to do so. I highly recommend this course to anyone looking to break into statistical arbitrage. Prepare to be pushed to your limit.
 

Attachments

  • Krause_Lu_Lwin_Rappaport_Stat_Arb_Project.PDF
    975.2 KB · Views: 598
As Alex mentioned in his post, I was also in Jim Liew's class as well as a member of the winning team. I strongly agree with everything that was said about the caliber of the teaching and the intensity of the assignments. The projects are geared towards self-starters and the amount of time needed to complete them should not be underestimated.

A little more about the final project, mock investment committee, and the prize...
The final presentation was 20 minutes long with an additional 10 minutes for questions. A total of four teams presented with either four or five people per team. The committee had members from: Goldman Sachs, SAC, Tradeworx, Athena, Credit Suisse, GETCO, Millennium, etc. Each provided invaluable feedback that will no doubt help us as we pitch other strategies in the future. The presentations concluded with the chance to socialize with the committee at a local pub.

The prize for first was the opportunity to present to Peter Muller's legendary Process Driven Trading (PDT) group at Morgan Stanley. The entire experience was both surreal and inspiring. We were graciously invited to present to four of PDT's senior members (including their chief scientist). Every aspect of PDT was impressive and well-exceeded our highest expectations. After hearing many stories of the “siloed” environments at other high frequency firms, it was refreshing to learn of such transparency and cohesiveness at PDT.

Jim Liew rewards hard work with opportunities you won't find anywhere else on Wall Street. If interested, more information about his innovative course can be found here: http://www.alphaquantclub.com/statarb.php
 
How is the material taught by Jim Liew as part of this Baruch's algo trading class different from the independent workshop that he runs?
http://www.quantnet.com/forum/threads/do-you-want-to-learn-stat-arb-quant-trading.5019/
How do you have time to study for other courses?

I took professor's Liew independent workshop, and I also sat at the final presentation that Alex and David are talking about (great job! btw). I also had access to both lecture notes, home works and final presentations, and I must say that they are pretty much the same. Of course, for obvious reasons, the quality of the presentations/techniques/strategies that I saw at Baruch is definitely higher than the one that I saw when I took the independent workshop but in terms of syllabi they look the same to me.
 

dstefan

Baruch MFE Director
I have a feeling the feedback from PDT is confidential.

From personal conversations, the feedback from the people on the panel at the final presentations was uniformly positive, and very supportive. In particular, the winning team (of which Alex was part of) made a very strong positive and professional impression.

Well done!
 
Correction to previous post...

"The committee had members from: Goldman Sachs, SAC, Tradeworx, Athena, Credit Suisse, GETCO, Millennium, etc."

Is now...

The committee had members from: Goldman Sachs, SAC, Tradeworx, Athena, Credit Suisse, ex-GETCO, Millennium, etc.
 
I can see that from the earlier post but I'd like to get a sense of how applicable what the students learn in this class is to the "real world".

The model which we created is far too simplistic to generate alpha in the "real world". It's probably 1-2 years of work from full implementation. However, the process in developing this model brings students up the learning curve very quickly. My team has internalized a lot of the lessons we learned while working on this project.

In the real world, no one is going to give you their "secret sauce" and the class emulates this very well. Jim gave us some guidance at the start but finding the alpha was up to us. It feels as much like performing novel (although it's not) scientific research as it does quantitative finance. A lot of what you learn will have to be discovered by yourself. The approach we took to creating the strategy is probably along the lines of what we would do on our second iteration of the project, except now we are a little wiser and more efficient.
 
First of all, congratulation guys on winning the competition! Great work.

Regarding real world application, the guys a blunder choosing in-sample data (2004), do you recall what changed in 2005 (do you need a hint)? Better approach would be to go for 2007/2008 the make the market environment more up to date. Furthermore, (realistically) they should have gone for a strategy with much lower Sharpe ratio and smoother pnl curve over the years. When you find that, you just leverage like crazy. Finally, my opinion is the strategy would work much better in FX world (non-commodities like rates).

Again, congrats!
 
First of all, congratulation guys on winning the competition! Great work.

Regarding real world application, the guys a blunder choosing in-sample data (2004), do you recall what changed in 2005 (do you need a hint)? Better approach would be to go for 2007/2008 the make the market environment more up to date. Furthermore, (realistically) they should have gone for a strategy with much lower Sharpe ratio and smoother pnl curve over the years. When you find that, you just leverage like crazy. Finally, my opinion is the strategy would work much better in FX world (non-commodities like rates).

Again, congrats!

Thanks for your input. We had a similar thought about it performing better in FX. Do you believe this solely because of increased capacity or are there additional reasons?
 
Yes, liquidity would be case, but since it's intended to be HF somewhere along 10k-100k units shouldn't be issue. Most likely you would pick only one pool of liquidity, or trade against a dealer (then there is no market impact but no limit orders as well), so you wouldn't have to navigate 11 exchanges and 50+ dark pools; that would take some burden of your order being flashed (aka front-run) or sub-pennied by ultra low-latency guys. I do not thing that FX is less of a "wild west" when it comes to trading, but at least I know whats up against me. Finally, you have massive swings on a daily basis in FX, and I feel hedging would be easier to set up.
Last thing, when I saw the strategy it came up to me: "this would work better in FX", I don't know how much intuition counts these days.
 
Top