Self-start projects for ML quant aspirant

[SRQuant]

I'm hoping to transition from CS to a quant position in finance. My PhD currently is heavy with Machine Learning and I have about 5 years experience as a software engineer previously. I'm mainly looking at quant jobs that will let me use my ML skills in algorithmic trading or risk assessment ( not sure about other ML quants) , whether in investment banks or in Hedge funds/Prop-Trading firms.

I was hoping to get some hands on experience, (for getting a good hang of things and for strengthening my CV) and am wondering if anyone here can suggest good projects that I can do in my spare time. I believe this will give me a good talking point during interviews and help me be more confident.

Being from a CS background I'm not very familiar with trading strategies but would like to start out implementing something basic as a good start. What's the best way to go about this? I'm aware of websites like Quandl.com from where I can get data to play with.

Should I start out implementing basic buy/sell strategies based on historic time-series data? I've come across simple algorithms that use basic features like changes in stock prices and commodities over small intervals and use these to predict future buy sell decisions. Is this a good enough project to implement? I want to start out with something simple and maybe move on to something complex as I proceed.

Thanks for your help.

 

[yetanotherquant]

I'm hoping to transition from CS to a quant position in finance.

For what? Do you want much money?
Well, quant are still relatively well-paid but the learning curve is steep and it is not so easy to find a job.

I was hoping to get some hands on experience, (for getting a good hang of things and for strengthening my CV) and am wondering if anyone here can suggest good projects that I can do in my spare time.

Yes, I can. Here are 6 years charts of Lufthansa.
Can you automatically detect the distinct peaks and dips automatically (and fit trendlines)?

I'm aware of websites like Quandl.com from where I can get data to play with.

You can also get data from yahoo.finance, in R there is quantmod package that does it automatically (in particular, the chart above was generated with it).
You likely also heard about quantopian (if not, google!)

Should I start out implementing basic buy/sell strategies based on historic time-series data? I've come across simple algorithms that use basic features like changes in stock prices and commodities over small intervals and use these to predict future buy sell decisions. Is this a good enough project to implement? I want to start out with something simple and maybe move on to something complex as I proceed.

In either case it is a big project. Start with a task I have formulated above. If you cope with it, we find the next.

 

[Zaubertrank]

What kind of ML are you doing in your PhD?

 

[SRQuant]

For what? Do you want much money?
Well, quant are still relatively well-paid but the learning curve is steep and it is not so easy to find a job.


Yes, I can. Here are 6 years charts of Lufthansa.
Can you automatically detect the distinct peaks and dips automatically (and fit trendlines)?
View attachment 16051




You can also get data from yahoo.finance, in R there is quantmod package that does it automatically (in particular, the chart above was generated with it).
You likely also heard about quantopian (if not, google!)


In either case it is a big project. Start with a task I have formulated above. If you cope with it, we find the next.

Excellent. Thanks for the suggestion. This should be easily doable. I found this paper which discusses multiple methods to automatically detect peaks in time series data. It's a good and easy read.

Simple Algorithms for Peak Detection in Time-Series

This will make a decent project for a start. Could you please point me to exactly how you got the data in the graph?

To answer your first question : I'm passionate about applied mathematics and programming. As a software engineer (my previous jobs) you don't get to actually do any applied maths at all. I don't find the regular software engineering positions to be intellectually stimulating enough (minus the competitive programming for interviews). There are certain positions like Data Scientists in the tech industry that use a lot of statistics/ML and expect similar skills but it only makes sense if the domain it's applied in interests me.. I'm very interested in the financial markets and it's what I'd like to apply my skills in. I have no interest in predicting Amazon clicks or Netflix recommendations. That being said, I'm aware of the jobs in Computer Vision and AI ( Deep Learning ) that are very interesting and I'm definitely considering those as an option.

 

[SRQuant]

What kind of ML are you doing in your PhD?

My PhD is in computer vision, which nowadays is a very applied area of Machine Learning. Some of the ML methods I use on a regular basis would be Hidden Markov models, Markov chains, Gaussian Processes for regression and latent models, SVMs, Linear/Logistic regression, Markov Random fields ( Bayesian Networks ), Gaussian mixture models, Probabilistic PCA, Time series analysis and optimization techniques ( SGD, CGD ) etc.

I understand the models and methods might vary, but I would think the underlying mathematics ( probability, optimization, calculus, linear algebra, statistics ) should apply even in the quant domain. Please correct me if I'm wrong.

 

[Philip_Mitterrand]

The skills are highly transferrable and you should think about building a portfolio of projects to bridge the gap. The intutive way would be look at SSRN, seek out the trending papers and implement/model some of the research. You will learn a great deal about markets new toolkits and know the buoyant areas of research. It will open a new space to speak while interviewing and make yourself more agreeable as a new hire.

My PhD is in computer vision, which nowadays is a very applied area of Machine Learning. Some of the ML methods I use on a regular basis would be Hidden Markov models, Markov chains, Gaussian Processes for regression and latent models, SVMs, Linear/Logistic regression, Markov Random fields ( Bayesian Networks ), Gaussian mixture models, Probabilistic PCA, Time series analysis and optimization techniques ( SGD, CGD ) etc.

I understand the models and methods might vary, but I would think the underlying mathematics ( probability, optimization, calculus, linear algebra, statistics ) should apply even in the quant domain. Please correct me if I'm wrong.

 

[Anupriya]

Hi Lynx

See if these projects/strategies are of any help:
Project Work EPAT

These are trading strategies project summaries (with codes uploaded on github) that includes different steps involving ideation, backtesting and execution conducted by our students.

Great to see your interest in Algo/Quant domain! Keep rocking!

-Anupriya

 

[SRQuant]

Hi Lynx

See if these projects/strategies are of any help:
Project Work EPAT

These are trading strategies project summaries (with codes uploaded on github) that includes different steps involving ideation, backtesting and execution conducted by our students.

Great to see your interest in Algo/Quant domain! Keep rocking!

-Anupriya

Thanks a lot. That looks like a useful site.

 

[yetanotherquant]

This will make a decent project for a start. Could you please point me to exactly how you got the data in the graph?

R-code:
library(quantmod)
startDate = '2007-01-02'
endDate = '2016-09-28'
getSymbols("LHA.DE", from=startDate, to=endDate) #Lufthansa
chartSeries(LHA.DE, subset="last 6 years")

 

[Phi-quant]

The skills are highly transferrable and you should think about building a portfolio of projects to bridge the gap. The intutive way would be look at SSRN, seek out the trending papers and implement/model some of the research. You will learn a great deal about markets new toolkits and know the buoyant areas of research. It will open a new space to speak while interviewing and make yourself more agreeable as a new hire.

Sorry but I'm not familiar with SSRN. What do I do to find the trending papers in there?