A deep dive into the strange new world of initial Coin Offerings
Wexboy with a great round up of all listed stocks that have some relationship with Cryptocurrencies
Why machine learning investment strategies fail (so far)
Don’t miss: Today is the last day to sign up for the free “Venture Deals” online course.
It seems that shareholder activism is slowly coming to Japan
A very interesting attempt to value Teva
The interesting case of Home Capital shareholders rejecting a (second) capital injection from Berkshire
Thanks for the link mmi!
since you are currently interested in bitcoin, this podcast might be for you
By the way I can also recommend Patrick O’Shaughnessy’s other podcasts!
thanks for the comment. But I think the lesson of the machine learning article is clearly that KI itself is not the solution for great returns but that you have to think hard about data input.
A lot of highly hyped KI funds have relatively quickly shut down, for instance the Catana fund. I guess it was because of such issues.
Training a “KI” on fundamental issues sounds interesting but will be even more challenging in my opinion, especially if there are structural changes.
I think this is a typical case where the reality and public perception of artificial intelligence diverge strongly. At the moment all “Machine Learning” and AI algorithms are more or less improved statistical models, even neural networks are just a version of a complicated statistical model with multiple layers (and so complex even the researchers see them as blackbox). I work as a Data Scientist so I clearly know what I am talking about.
It is natural that due to a bunch of observable anomalies this is the exactly right approach for complex arbitrage trading. The short term approach can identify also the effectiveness of correlations between stocks (for example pair trading between companies in the same sector) and trade them. But yes, the problem is that one has to find out when changes in the price are justified by “soft” information like politics or when they are just random and thus probable to rebound. And this is where the human thinking comes in. Since stock markets are far from perfect and stock prices often irrational you have no good training data if you only take the stock price movement as target. And with fundamentals there is not that much training data. So you have to be extremely careful to avoid overfitting, an almost impossible task if your data consists of only 10000 companies each having many dependent “dimensions” like Revenue, Earnings, Capital … The statistical models fed with this data overfit or learn vague things every human knows like low PE is better.
But in the long term different models will come up and it might get ever more difficult to outperform computers as a human.
Hey, Thanks again for these really interesting collection, especially the Teva case 🙂
But I have one comment regarding the machine learning strategies. I think the link title is misleading, because what is described there is what kind of mistakes make you fail when implementing automatic trading algorithms based on machine learning. All it talks about is how to sample from price data the right way, how important it is to not leak the training data and so on, so it is looks not like what I would call long term investing. Secondly this post admits that there are a few firms that continue to make huge amounts of money, so in principle the approach works. It is just a bit more difficult than some people might think. Especially the remark that returns of stocks are not iid indep. id. distributed is an important point to think about.
But I think one of the biggest problems is that it is trading (based on charts) and not investing (based on fundamentals). Due to lack of time and big datasets I never tried, but I think it could be promising to train models predicting not stock returns but the fundamentals of a company over the next years. I guess if someone solves this (at least approximately) he could make great long term success…