Ah, so they are not split-adjusting historical dividends?

Hi Simon, interesting that I didn't notice the dividend records from Yahoo is adjusted by split. But here's what I get from Nasdaq without split adjusting: http://www.nasdaq.com/symbol/aapl/dividend-history
8/7/2014 Cash 0.47 7/22/2014 8/11/2014 8/14/2014
5/8/2014 Cash 3.29 4/23/2014 5/12/2014 5/15/2014

Let's take a look at the cash dividend distributed at 5/15/2014: 3.29 * 1326 = 4362, which looks correct for the dividend cash calculated in my original post. But let's think about the real case:

1. On the first day 2014-03-01 of BT, I've put my whole 100000 into market, I can ONLY hold 100000 / 527.76 = 190 shares. And then hold to 5/15/2014 I've got my first dividend cash: 3.29 * 190 = 625.1, it's much less than 4362 in Q's calculation.

2. Then at 2014/06/09, split happens, I've now got 190 * 7 = 1330 shares.

3. Finally at 2014/08/15, a new dividend 0.47 happens, I've got 0.47 * 1330 = 625, only the last dividend calculation in Q is correct.

Or... I am wrong about data somewhere?

OK, a simple way to think is:
Before split the dividend is 3.29, after split dividend is becoming 0.47 (got /7). And shares also enlarged by 7 times, so the dividend cash received by these two dividend events should be the SAME instead of 4361 vs 623 if you check in the "Daily Positions & Gains".

I agree with you. I ran into the very same problem in my system, hopefully they'll get it fixed.

Thanks Simon for your reply.

Hi Q guys, will this get fixed? Otherwise all the returns and risk related calculation is NOT ACCURATE.

This helps show the issue:

def handle_data(context, data):  
    for stock in data:  
        PrintPosition(context, data, stock, data[stock])  
    if not context.fired:  
        order_percent(context.aapl,1)  
        context.fired = True

def PrintPosition(context, data, stock, stockData):  
    print(  
        "{0:>15} {1:>10.2f}".format("Positions value", context.account.total_positions_value) +  
        "{0:>15} {1:>8.2f}" .format("Portfolio P&L", context.portfolio.pnl) +  
        "{0:>15} {1:>8.2f}" .format("Portfolio cash", context.portfolio.cash) +  
        "{0:>10} {1:>5}"    .format("Symbol", stock.symbol) +  
        "{0:>15} {1:>8.2f}" .format("Close price", stockData.close_price) +  
        "{0:>15} {1:>8.2f}" .format("Quantity", context.portfolio.positions[stock].amount) +  
        "{0:>15} {1:>8.2f}" .format("Cost basis", context.portfolio.positions[stock].cost_basis))  

I think I just ran into this myself backtesting a strategy on NUGT - quantopian misinterprets a $1 dividend in 2011 as a $50 dividend, since it's since been split 1:50.

That does look like a bug. I'll write it up and look into it. Thank you for reporting it.

This is fixed!

If you re-run the backtest above, you'll find that the total return is lower because the dividend returned is lower (and now correct).

Similarly, there seems to be issues with the dividend values in the Morningstar data.

Is there another, more reliable, source for dividend data?

Here it reports the last dividend per share paid by AAPL to be 3.0, but in reality it was 0.77 according to Yahoo.

Ah ok the FactSet data seems to be accurate.

Fundamentals can be tricky. The timeframes aren't always obvious especially with the Morningstar data. For the most part, the Mornngstar data is quarterly and represents the numbers from the latest 10Q filing. However, in the case of cash_dividends_paid, the value is for the total trailing twelve months (TTM) dividends. The Factset and Yahoo data are simply the most recent dividend.

In this case, the value 3.0 on 2019-9-3 for AAPL is correct. It represents the sum of the TTM dividends:

8/9/2019    $0.77  
5/10/2019   $0.77  
2/8/2019    $0.73  
11/8/2018   $0.73

It's excellent you are looking at the fields before passively using them. There can of course be issues with the data but, more often, the issue is in what the data represents.

Oh, thank you Dan!