Stanley,

Fortunately my test in IDE did not prove any significant difference in USEquityPricing.close.latest and data.current(symbol('ARNC'), 'close')

# Test pipe_prices and  data_prices

from quantopian.algorithm import attach_pipeline, pipeline_output  
from quantopian.pipeline.data.builtin import USEquityPricing  
from quantopian.pipeline import Pipeline, filters

def initialize(context):  
    set_symbol_lookup_date('2018-04-15')  
    price = USEquityPricing.close.latest  
    my_stock = filters.StaticAssets(symbols('ARNC'))  
    attach_pipeline(Pipeline(columns = {'price': price}, screen = my_stock), 'my_pipeline')         

def before_trading_start(context, data):  
    yest_pipe_price = pipeline_output('my_pipeline').price  
    yest_data_price = data.current(symbol('ARNC'), 'close')

    print(yest_pipe_price, yest_data_price)  

2015-04-15 05:45 PRINT (Equity(2 [ARNC]) 13.375
Name: price, dtype: float64, 13.375)
2015-04-16 05:45 PRINT (Equity(2 [ARNC]) 13.34
Name: price, dtype: float64, 13.34)
2015-04-17 05:45 PRINT (Equity(2 [ARNC]) 13.46
Name: price, dtype: float64, 13.46)
2015-04-20 05:45 PRINT (Equity(2 [ARNC]) 13.47
Name: price, dtype: float64, 13.47)
2015-04-21 05:45 PRINT (Equity(2 [ARNC]) 13.58
Name: price, dtype: float64, 13.58)
2015-04-22 05:45 PRINT (Equity(2 [ARNC]) 13.52
Name: price, dtype: float64, 13.52)
2015-04-23 05:45 PRINT (Equity(2 [ARNC]) 13.56
Name: price, dtype: float64, 13.56)
2015-04-24 05:45 PRINT (Equity(2 [ARNC]) 13.18
Name: price, dtype: float64, 13.18)
2015-04-27 05:45 PRINT (Equity(2 [ARNC]) 13.21
Name: price, dtype: float64, 13.21)
2015-04-28 05:45 PRINT (Equity(2 [ARNC]) 13.45
Name: price, dtype: float64, 13.450000000000001)
2015-04-29 05:45 PRINT (Equity(2 [ARNC]) 13.48
Name: price, dtype: float64, 13.48)
2015-04-30 05:45 PRINT (Equity(2 [ARNC]) 13.61
Name: price, dtype: float64, 13.61)
2015-05-01 05:45 PRINT (Equity(2 [ARNC]) 13.42
Name: price, dtype: float64, 13.42)
2015-05-04 05:45 PRINT (Equity(2 [ARNC]) 14.15
Name: price, dtype: float64, 14.15)
2015-05-05 05:45 PRINT (Equity(2 [ARNC]) 14.015
Name: price, dtype: float64, 14.015)
2015-05-06 05:45 PRINT (Equity(2 [ARNC]) 13.83
Name: price, dtype: float64, 13.83)
2015-05-07 05:45 PRINT (Equity(2 [ARNC]) 13.68
Name: price, dtype: float64, 13.680000000000001)
2015-05-08 05:45 PRINT (Equity(2 [ARNC]) 13.705
Name: price, dtype: float64, 13.705)
2015-05-11 05:45 PRINT (Equity(2 [ARNC]) 13.82
Name: price, dtype: float64, 13.82)

thanks, hmmm so is data.current(symbol('ARNC'), 'close') fetching from a different source than prices(assets=symbols('ARNC'))?

or is this bug perhaps only a problem on research notebooks, and is not a concern for the IDE?

I was struggling with getting returns for Alphalens that are not polluted by splits.

My impression is (but please correct me):

Prices retrieved through pipeline are what you were seeing at that point in time. This is what your algo should use.

As opposed to prices returned by the price function which are corporate action adjusted. This should be used for evaluation.

Maybe there's also a difference how the research and the algo evironment work for these two things.

Attila,

You are right that pricing data retrieved with Pipeline is adjusted point in time, while pricing data retrieved with get_pricing() is adjusted as of the end_date argument that you pass to it. The reason that get_pricing() is preferable for research / evaluation of alpha factors is that it is less computationally expensive, and thus you can iterate over ideas more quickly.

When you adjust prices to the present day, you introduce subtle lookahead bias. The workflow we encourage is to use Alphalens to quickly iterate over ideas, throw away alpha factors that clearly aren't worth anything, and use our more thorough (and slower) backtester once you settle on a set of alpha factors that you think is worthy of the full backtest.