How does numer.ai make predictions about the future? Numer.ai is a crowd sourced hedge fund that uses the individual classifiers of its users to predict future asset prices.
They themselves do not provide a lot of information on how it works. There is a couple of articles out there (Article 1, Article 2, reddit ). The way I understand it is as follows:
They take propriety real world financial data and encrypt it with homomorphic encryption. Homomorphic encryption is a mapping method that makes it possible, from analysis performed on the encrypted data, to derive conclusions on the original data. They split the data into a training data set and a test data set.
The training data's shape is (96320, 22) with $21$ normalised features columns, no index and a binary label target. The test data's shape is (36072, 21) with $21$ normalised features and unsorted index t_id.
The idea is to develop a classifier based on the training data and then to submit your prediction for the test set. If your classification performs well on the test set and your algorithm is original (diversifying), they pay you in Bitcoins. Unlike Kaggle you do not upload your code.
Throwing a couple of standard algorithms at the problem I came up with a log loss of $ \approx 0.6930$ which doesn't get you anywhere close to the top ranks.
My question is: How are they making money out of this?
If encrypted data is based on historical stock prices, financial fundamental data or any other real world time series data (which I assume you will need for any kind of quantitative trading), how can you predict future asset prices without having access to my classifier?  Can you infer it from my classification of the training data set or the test data set (closest distance)? That seems unlikely given the dimensionality of the data set. You would have to look at the most recent data points and put them into the classifier. I assume this is not a Ponzi scheme because they have sophisticated investors with a quantitative trading background.
 A: If you sort the t_id column in the test dataset you will see that it goes from 1-36000. I would assume that it refers to "trade id". The way financial time series forecasting works is the you usually take lagged values of features at time t-1 and use them to predict the target value at time t, thus I would assume that all the features from 1-21 are lagged values from the previous week and the target variable must be the price increase/decrease for a particular trade id. 
If you look at the probabilities that your algo outputs you can see that they are usually in a range of 0.45-0.55 or something like this, thus it's not very precise, however you still get a slight edge over random results. This explains the large testing set of 36000, in order to squeeze that small edge you need to make a lot of trades. Here Renaissance Technologies (arguably the best quant fund) mentions that they have a very slight edge in forecasting the prices, but they exploit it via a large number of trades + they are using a lot of leverage to enhance these returns. I would assume Numerai is doing something similar.
A: You give them your predictions and not your classifier.  
They're lowering the barrier to entry in one area, and not all of them.  Sure, you could start up a fund of your own or sell your classifier to some other firm, but, (or so I think) goes their thought process and business model, the competition that they're providing via incentivisation will lead to them them coming out ahead (or at least tied with the best) as everyone will be incentivized to share their future predictons with them.  
