Hello Taibo,

Unless something has changed, the mavg transform cannot be called with a variable; you need to specify a fixed number of days. To Quantopian Support, I suggest adding guidance to your help page, under the "Simple Transforms" section.

To code a weighted moving average, I suggest considering the batch transform and a numpy average with weights. If you write one, please post it here--I'd be interested in seeing your example.

Grant

I do feel the error message could be better, since this is going to be a fairly common "gotchya"

def initialize(context):  
    pass

def handle_data(context, data):  
    n = 3  
    f = data[sid(24)].mavg(n)  # "There was a runtime error"  

Grant is right. The various transforms are created at the beginning of a backtest, and if you try to change the size of the window part way through, it barfs.

I agree we need a better error. I'll work on that and the docs.

@Grant: Thanks for your suggestions! Here's what I did using batch transform and numpy average. We can tweak the weights to make it closer to exponential average or any other kind of average.

Taibo,

Yes, that's the idea. I'll take a look at your code when I get the chance. Glad it worked out.

Grant

Hello Taibo,

If you have access to a copy of Python for Data Analysis by Wes McKinney, there is a section Moving Window Functions in Chapter 10, Time Series. Code and examples are available there, including a discussion of exponentially-weighted functions.

One could also consider an algorithm that adaptively adjusts the weighting of the moving average, to optimize the portfolio performance as market conditions vary. Thomas Wiecki may have some ideas (for a general outline of walk-forward optimization, see his blog post http://blog.quantopian.com/parameter-optimization/).

Grant