In the good old days before the computer revolution, stocks were generally priced by traders using pencil and paper. Analysts would read news, study corporate financials, and then give ideas to traders. Trading happened in a physical trading pit. When a large player (Goldman Sachs, J.P. Morgan) wanted to …

Given the commodification and decline of high frequency trading, I was a bit surprised to see that Michael Lewis wrote a book on the topic. Not only that, but based on the reviews (I haven't read the actual book), it sounds like a scary "tell-all" book revealing how HFT rips …

A Fervent defense of Frequentist Statistics. Eliezer Yudkowsky's comments are also well worth reading. Most notably:

Who are these mysterious straw Bayesians who refuse to use algorithms that work well and could easily turn out to have a good explanation later? Bayes is epistemological background not a toolbox of algorithms …

Bryan Caplan recently asked for one good reason why basic eye exams can't be done by machine. I.e., why a machine can't ask "better like this, or like that?"

Computerized eye testing is probably a little different than what Caplan imageined. The shopkeeper (not a medical doctor) is still …

For most data scientists, a toolkit like scipy/matplotlib or R becomes so familiar that it becomes almost an extension of their own mind. When fleshing out an idea, we will just open up ipython, write a few lines of code, and display the plot on the screen to get …

I've been hacking on Breeze lately, which is a scientific library for Scala which aims to replace Scipy. It's got a long way to go, but it's definitely a great library and one that I make significant use of. One of the great things about using the same language that …

Last week I gave a talk at Helpshift about Bayesian analysis of conversion rates, A/B testing and Bandit Algorithms. The slides are hopefully self contained and might be useful for others. You can view them here.

I was recently introduced to the scalaz-stream library by the excellent blog post Actors are Overly Nondeterministic. I decided to take a look at them, and I realize they have a lot to offer.

One of the cools things I noticed could be built with them is Clojure-style Agents. But …

In the past couple of weeks, the excellent article A Fervent Defense of Frequentist Statistics has been making the rounds. The article makes some strong arguments in favor of using frequentist tools, since many of them are more computationally efficient than Bayesian methods.

However, the author makes a fundamental mistake …

A lot of people believe that Bitcoin and other cryptocurrencies are a speculative bubble. Scott Sumner is a notable person who disagrees, and believes that Bitcoin may be rationally priced based on the low probability of high expected rewards.

Most of the arguments claiming BTC is a bubble are nonsensical …