The key is the date that you own the stock. The dividend was awarded on April 3, but only to shareholders of record on March 27. In the backtest you shared, the purchase was on April 1, so you don't actually get the dividend. If you move the purchase date earlier, then you get the dividend.

With the backtest I shared above I was trying to show how I believe Quantopian handled this data is incorrect. The Google split was a little unconventional and maybe the way you have to track it is through a dividend like you mention. I guess that question is what should happen when the stock is purchased after March 27 and before April 3 and what "price" is paid for the shares. I suppose paying half the price during this period would fix this?

I was also just able to pick out the trades from my algo that were able to capitalize (accidentally) on this split, giving an immediate 50% return that wouldn't have actually existed.

Yeah, I was thinking about this one all day. There's something not right there. That Google split is the gift that keeps on giving, from a platform perspective. It was a one-of-a-kind event and requires a lot of custom code to handle.

It's funny because most of the writeups you see of the event on the internet refer to the dividend being payable to owners of record on March 27, which is how our code is handling it. But clearly the price of the stock shows that if you bought the stock on April 2nd, you still get the payout on April 3rd. Somehow that date of March 27th because irrelevant in practice.

I'm going to write a bug on this one, but I don't plan on putting the "urgent" tag on it. Just to set expectations =)

It also looks like the history() function doesn't handle this well, it returns a stair-step in the prices as you move across the split date.

OMG, I have tears in my eyes, thank you Trevor for bringing this year old issue back to light, I have not laughed so hard in days. The GOOG split and how it has not been handled actually ties together many issues that I have seen on the Quantopian platform regarding returns and is actually quite funny to anyone with any knowledge of actual equity trading.

You see, most of Quantopian's staff have not worked in the investment industry and so they have blind spots a mile wide for a great many common practices. I could tell you, and the Q staff exactly what they did wrong with the GOOG split and how to generically handle it, along with solving their return discrepancies.

Donate two bitcoins to Khan Academy, post the transaction hash here and I'll tell you what you are doing wrong and teach you a valuable lesson.

In the meantime, anyone who uses Q's platform to trade or even backtest ... well ... sometimes the best things in life are not free.