Brief To become a good machine learning practitioner, simply finishing school work is not enough. I recommend you find a topic you truly love and try to adopt theoretical framework to that topic.
Take myself for an example, I love looking at stock charts. When I was a child, my dad would discuss Security Analysis with me instead of Bible. I grew up with the passion of loving stock market and it is part of my identity. Later I started going to school and I discovered that I am able to relate almost all mathematical theorem with stock market. I test them out with real money. Of course I made mistakes and lost money before. I record the action and decision making process. I come back to computer and simulate the event again and again to myself until I figure out something right. Then I use real money again. This cycle repeated since 2010 and I pocketed a unique pool of experience no one has done before in capital market. Nobody pays me anything and I don't get high grades from this. I do this because I truly like it. I don't need other incentives.
First, I started by looking at correlation. This links to running regression and I have conducted experiment to learn that myself when I was a freshman. What is the linear relationship between market return and a stock return? How about fundamental values?
Next, I started to size up. A data composed of stock-to-market comparison seemed too easy, so I started looking at cross-sectional finance data. The common data set that is used for PhD students in finance is CRSP stock universe. This is where I started to work with large-scale matrices that are 2GB in size and when I learned to work with different forms of data frames. It started to turn into tedious data clean-up work. However, the results are interesting.
- I wrote Cross-section Study on Stock Returns; and
- Another similar project that examines stock market through market value balance sheet, click here if interested.
After looking at tons of stock returns data set, I got bored at conventional way of analyzing stock market. I invented this term "greed" so that I can turn this into a supervised model for myself to study. This is not publishable idea, but quite interesting to me.
- Paper is here.
In 2016 I took a little digression to study macroeconomics. I did a project on contact rate in game theory. The paper was too hard, but the coding part is easy. I got stuck on dynamic model in macro material and it took me 3 months to figure this out. It leads to nowhere, but I am grateful to the professor who instructed me and him spending time with me on this project.
- Trade Dynamics with Endogenous Contact Rate is here.
After trading floor, I took a little bit of time and summarizing my experiences on the trading floor and they become the following series of papers:
- Buy Signal from Limit Theorem
- Time Series Analysis on Stock Returns
- Martingale to Optimal Trading
- Anomaly Correction by Trading Frequency
- Absolute Alpha with Moving Averages: A Consistent Trading Strategy
- Absolute Alpha with Limited Leverage
- Absolute Alpha by Beta Manipulation and this when I started to build up intra-discplinary experience across the fields of (1) trading, (2) probability theory, (3) accounting, and (4) game theory. Most of what I am right now depend on these past couple of papers.
Interesting phenomenon (and you probably saw it too) was that I did not have any machine learning technique discussed yet. However, the few papers above gave me a unique view so that I can create any stock data however I want. This is when machine learning comes in, which leads to this paper.
- Robust Portfolio by Influence Measure with presentation
- A small project that comes along this paper is finished as an Applied Data Science project
I am equipped with two unique skills: (1) I can use probability theory to manipulate and invent any stock data helpful for me; and (2) I can invent a machine learning algorithm that solves that problem. All of these ideas are updated frequently here.
The above stories are mine to told and I hope you enjoy reading them. There is still much to learn for me, but I want to let you know that you can get it started just like I did without going to prestigious universities. As long as you have the will, you can keep on learning. That, my friend, is the beautiful thing about learning.
Now I am a PhD student going to Columbia University and I continue to write my stories and legacies. I encourage you to write yours, and perhaps one day you can look back and enjoy the legacies you left as well.