If you will use .bfill() you will get same results.

# SMA indicator

import talib

def initialize(context):  
    schedule_function(record_SMA, date_rules.every_day(), time_rules.market_close())

def record_SMA(context, data):  
    etf, period  = symbol('SSO'), 150

    price = data.current(etf, 'price')  
    prices = data.history(etf, 'price', period, '1d').bfill()  
    SMA_ta = talib.SMA(prices, period)  
    mavg = prices.mean()

    print(SMA_ta[-1],mavg)  
    record(price = price, mavg_ta = SMA_ta[-1], mavg = mavg)  

2006-07-03 09:59 PRINT (71.516933333333327, 71.51693333333333)
2006-07-05 12:59 PRINT (71.529820000000001, 71.52982)
2006-07-06 12:59 PRINT (71.545133333333325, 71.54513333333333)
2006-07-07 12:59 PRINT (71.55353333333332, 71.55353333333332)
2006-07-10 12:59 PRINT (71.563333333333318, 71.56333333333332)
2006-07-11 12:59 PRINT (71.577286666666666, 71.57728666666667)
2006-07-12 12:59 PRINT (71.579553333333322, 71.57955333333332)

Hi,
Thanks for the reply first.

But:
You use the talib.SAM and mean(). They will have the same values. But have you tried using the Built-in-Factor 'SimpleMovingAverage'? They are somewhat different.

Is it a correct way to use the bfill() here? The inception day of SSO was on Jun 19, 2006. There was no price 150 days before this date. The SMA should be NAN. If not, this means it is not really the SMA(150), right?

Hi @Thomas,

The built-in SimpleMovingAverage factor ignores NaN values when computing the mean. On the other hand, NaN values will cause talib.SMA to return an array filled with NaN values. As you noticed, your algorithm is passing NaN values to talib.SMA through stock_history. To solve this, you can drop NaN values from stock_history by using its .dropna() attribute method.

The above solution will introduce another problem. Talib's SMA function will return an array of NaN values if the timeperiod used is larger than the size of the input array. Since we are dropping NaN values from stock_history, we will need to set timeperiod to be the length of stock_history.

The attached backtest applies both fixes. You will notice in the log output that the values returned by both talib.SMA and SimpleMovingAverage are equal.

@Ernesto
Many thanks for the reply.

But this means the SMA at the beginning is not of SMA(150) but SMA(50) or even SMA(10), SMA(5) etc?

I prefer to do as follow:

    if len(stock_history) < context.sma_period:  
        stock_sma = talib.SMA(stock_history, timeperiod=len(stock_history))[-1]  
    else:  
        stock_sma = talib.SMA(stock_history, timeperiod=context.sma_period)[-1]