Assignment 5 BUSFIN 7234 – FinTech

Submission by – Darshan Kataria, Vinay Menon, Vaibhav Modi,

Part 1: ML exercise

B) Ans: Using the rank analysis of the features for the given data, we find the following five features (shown in image below)
are having highest explanatory power for default. However, recoveries and collection recovery fees are highly correlated with
default rate because of their value being available at time of actual default only and represents data leakage problem. Hence
we will exclude there two features from our model.

C) Ans: We stated in previous part, we exclude only two variables : recoveries and collection recovery fees from our model using
select column widget and the updated result using rank widget shows the top 5 features (shown in image below) that are having
highest explanatory power for default.

D) Ans: As we are using many variables, we ran three different models Logistic Regression (Ridge Regularization), Logistic
Regression (Lasso Regularization), and Random Forest. The model is shown in below diagram.
E) Ans: Evaluation of the three Models
Of the three, Logistic Regression Lasso Model and Logistic Regression Ridge Model are pretty close in the ROC analysis as can
been seen in the overlap between them in diagram below, Lasso is marginally ahead of Ridge as can be seen in confusion matrix.
Criteria Logistic Regression Lasso Logistic Regression Ridge Random Forest
Confusion
Matrix

ROC
Analysis
F) Ans: Supplementing the above analysis with F1 and AUC score for all the models (as indicated below), we believe Logistic
Regression Ridge model is the best for predicting default rates as it has the the highest F1 and AUC score and have the highest
Precision and Recall ratio.

Part 2
A) Quantopian is a web-based platform that allows professional and amateur quantitative traders to develop, test and
execute their strategies. It follows a classic freemium model, wherein it offers most of its services such as backtesting and
access to many databases for free and charges subscription fees for premium databases such as CEO Changes and Earnings
Surprise. The company has a two-sided business model -
1. The developers/algorithmic traders who develop, test, trade, and execute the algorithms and the strategies for
Quantopian’s investors.
2. The institutional investors who invest in Quantopian’s crowdsourced hedge fund
The fee structure is like most other hedge funds such as an annual fee of 1% of AUM plus a performance-based fee.

Even though hedge funds have been around for a long time, they barely beat equity indices such as the S&P 500 or DJIA overall
over a long period. Traditional hedge funds have higher level of discretion and flexibility over investments compared to
other investments such as mutual funds but have low levels of disclosure and high liquidity pressure. Also, with the rise of
internet and other sophisticated technologies, investors’ preference for quant hedge funds has consistently increased over time.
Seeing the weakness of the traditional hedge-fund model, Quantopian aims at revolutionizing the industry with a completely
different mindset. In 2015, it launched its first crowdsourced hedge fund. It does not hire quants for top universities or financial
institutions but crowdsources algorithm development from its community. Only some algorithms that ran through various tests
and measures by Quantopian were allowed to run on Quantopian’s platform and it also provided all administrative support. It
has developed a comprehensive scoring system with several performance metrics for different algorithms. Compared to the
conventional model, Quantopian’s new crowdsourced hedge funds solves many problems faced by the hedge fund industry. For
investors, algorithms and trading strategies are no longer a black box and the investment process is much more transparent,
which gives a huge boost to investor confidence. Instead of investing in a particular hedge fund manager, investors could invest
in a particular algorithm or strategy that has been backtested and reviewed by Quantopian’s detailed performance metrics.

This also reduces the operational risks as well as the costs of owning and operating a traditional hedge fund. The risk of
employing incompetent quants is greatly reduced since only proven and backtested algorithms are being invested in, which
reduces human capital costs. The fund also has a large algorithm pool, which allows diversification and reducing systematic
investment risk. Hence, we believe that Quantopian’s model has the potential to revolutionise the FinTech and the Hedge fund
industry.

B) Ans:
Numerai is an artificial intelligence hedge fund that uses Numeraire (NMR), a token on Ethereum, to incentivize data scientists
to improve the hedge fund through the improvement of its artificial intelligence. This fund’s trades are chosen by proprietary
artificial intelligence algorithms and encrypted public datasets. This is achieved by crowdsourced quantitative models from
more than twelve thousands of anonymous data scientists. The data scientists are rewarded with cryptocurrencies (NMR tokens
or Bitcoin in the past) for their effort through a collaborative reward mechanism. . Numerai tries to evaluate the models based
on how well they behave on new sets of data, improving the overall quality of the artificial intelligent hedge fund. 1
Evaluation of Numerai crowd wisdom model:

 Potential to emerge as a fully decentralized, or leaderless, software-powered hedge fund.

 Instead of relying on the genius of a single person or small team, the firm uses encrypted data sets to crowdsource stock
market prediction models built with artificial intelligence from tens of thousands of anonymous data scientists
worldwide trying to improve its investing strategy.

 Increased collaboration than competition within the data scientist community as value of their reward is linked with
value of cryptocurrency that they receive from Numerai.
However, two major problems with their models are capped payouts for winners, which in absolute amount is also very less
($54, 000 per annum) and problem of “Over-fit”. Numerai, which currently receives about 100 million predictions from its users
per day, then creates a meta-model from all submissions to make its investments. Problem is, because the models are built using
historical data, the data scientists sometimes “over-fit” it, so their model does better in the tournament and thus earns them
more bitcoin, but those predictions then perform less well in the live stock market. 2

We believe, between Numerai and Quantopian, Numerai has more potential to grow . This is because while Quantopian is
mainly democratizing the algo trading techniques and platforms and taking it to masses, they work more like a traditional hedge
fund in their offering and fee structure. However, Numerai is redefining the hedge fund industry in itself where data scientists
continuously try to improve their investment strategy and by linking their reward to value of common crypto currency, it has
smartly facilitated collaboration among them. Also, their rewards are linked to how the fund performs. Thus, it has designed
itself to be a win-win platform for all stakeholders.

1
https://medium.com/@luis.freitas/numerai-does-everyone-win-a9af15f58619
2
https://www.forbes.com/sites/laurashin/2017/02/21/this-is-the-worlds-first-cryptocurrency-issued-by-a-hedge-fund/