Re: time based moving average (and other newbie mathematica
- To: mathgroup at smc.vnet.net
- Subject: [mg92767] Re: time based moving average (and other newbie mathematica
- From: Mark Fisher <particlefilter at gmail.com>
- Date: Mon, 13 Oct 2008 06:15:47 -0400 (EDT)
- References: <gcn496$751$1@smc.vnet.net>
On Oct 10, 4:37 am, falcon <shahb... at gmail.com> wrote:
> Hi,
> I see that Mathematica provides a couple of moving average functions.
> As far as I can tell, they are based on the number of elements in an
> array. Is there a function for doing moving average based on time?
> In other words, if I pass in intra-day data containing prices, times
> (up to a millisecond) and some other fields, can I get Mathematica to
> to give me a 5 minute moving average rather than moving average of the
> last 100 trades? Obviously this five minute window may contain any
> number of elements.
>
> Secondly, along the same idea, I have a file with a large number of
> stocks. They are all mixed in (the file is in chronological order).
> Is there an sql type 'group by' command that lets me see moving
> averages for each stock?
>
> Third, if I'm able to get moving average for each stock in the list,
> can I plot all of them in one command (I guess this technique is
> called "small multiples" in charting vernacular). Obviously I will
> only have 20 or 30 stocks.
>
> Thanks
Interesting problem. Here's my take:
MovingTimeAverage::usage = "MovingTimeAverage[data, lag] takes a list
\
of {time, value} pairs and returns a list of {time, avg} pairs where \
avg is an average of the values computed from the window [time - lag,
time]. \
MovingTimeAverage assumes the data are sorted.";
MovingTimeAverage[data : {{_, _} ..}, lag_] :=
With[{lagindex = LagIndexCompiled[data[[All, 1]], lag]},
Table[{data[[i, 1]], Mean[data[[lagindex[[i]] ;; i, 2]]]},
{i, Length[data]}]
]
LagIndexCompiled =
Compile[{{time, _Real, 1}, {lag, _Real, 0}},
Module[{j = 1},
Table[While[time[[i]] - time[[j]] > lag, j++]; j,
{i, Length[time]}]
]];
ndraws = 10^4;
times = Accumulate[RandomReal[ExponentialDistribution[5], ndraws]];
values = Accumulate[RandomReal[NormalDistribution[], ndraws]];
data = Transpose[{times, values}];
ma = MovingTimeAverage[data, 5]; // Timing
ListLinePlot[{data, ma}]
On my laptop, it takes about .1 second to compute the moving average.
The key is to compute the position of where the window starts for each
observation (lagindex).
--Mark