I could mention that my own theory is that it's because of this:

All securities have a unique security id (SID) in our system. Since symbols may be reused among exchanges, this prevents any confusion and verifies you are calling the desired security. You can use our sid method to look up a security by its id, symbol, or name.

But the documentation only mentions the problem, without offering a solution - how do I set exchange? (NYSE in this case, in Mathematica I would write "NYSE:DAL" as an example.)

Tickers are reused on exchanges as companies go private, bankrupt, merge etc. In a simple algo, if you try to backtest using "symbol('AI')", you get an error because the symbol has been used by two companies. The backtester requires you to specify which stock you want to trade.

To fix the error, add the following to initialize() to trade Arlington Asset Investment Corp.

 set_symbol_lookup_date('2004-01-01')  

No, that won't fix the error. From the documentation:

Fetcher maps the symbol to the Quantopian security id (sid).

It's this mapping that fails, and it seems to me like that internal mapping isn't affected by set_symbol_lookup_date. I provide an example of this where I have replaced two stocks from Quantopian's example with EDS and AI, who suffer from this problem.

The log output shows:

1970-01-01preview:18INFO symbol start date stock_score
0 AI 2/13/12 11.7
1 EDS 2/13/12 15.8
2 FDX 2/14/12 12.1
3 M 2/16/12 14.3
1970-01-01preview:18INFO stock_score sid
dt
2012-02-13 00:00:00+00:00 11.7 0
2012-02-13 00:00:00+00:00 15.8 0
2012-02-14 00:00:00+00:00 12.1 Security(2765 [FDX])
2012-02-16 00:00:00+00:00 14.3 Security(2754 [M])

As can be seen, the mapping returns 0 for both EDS and AI instead of the right SID.

EDIT: In case someone has the same problem and finds this topic, my workaround is to add the following in the post_func, it removes tickers that weren't mapped properly:

  df = df[df.sid != 0]  

Digging into this some more, those tickers are failing to map to your algo because of a bug in our backend. If a symbol is re-used (like 'AI') then it doesn't get mapped properly and is filtered out from your algo universe. If the ticker is unique, then it will work normally, as expected.

In the meantime, you have a good workaround in the post_func.

Ok, good to get that confirmed :)

I found a temporary fix for this. Use this as your pre_func

def pre_func(df):  
    df['sid'] = df['symbol']  
    return df  

It works for most symbols, FB included. Some symbols like TWTR still doesn't work for unknown reason.