The period settings are confusing for the Rotation models. In the Compose panel, one selects the Rebalance Frequency, and in the Settings panel, one selects the Data Scale periodicity. This would seem to mean we set Rebalancing dates in the Compose panel and the scale of graphing in the Settings panel. If the same value (say Monthly) is selected for both of these, everything seems to make sense, like the Positions backtest panel shows that all the entries were near the first of each month, etc.
But if we change Settings (Data Scale) to Weekly, leaving Compose (Rebalance Frequency) at Monthly, the new backtest is different completely, and the Positions show that purchases (rebalances) were still monthly (although after the first week, not at the month's end). Change Data Scale to daily (shouldn't this just change the graph?), and now there seems to be rebalancing monthly, but a day or so into the new month.
However, the returns for the three choices are totally different, as are the number of Positions taken. In the model I was testing (with Rebalancing set to Monthly for all), there were 27 positions (+5 NSF) when Data Scale is set to Monthly, 72 (+19 NSF) positions with Data Scale set to Weekly, and 405 (+7 NSF) positions when set to Daily.
What's going on? Oh wait, I guess maybe that "Data Scale" sets not only the graphics for output (so that one can better visualize portfolio fluctuations between rebalancings), but also the input periodicities for the indicator chosen in Compose. I'm going to test that idea: using SMA(Close,24) as an indicator, we can see that Monthly for Data Scale gives us a 2 year lag, Weekly gives us about 6 months, and Daily barely lags a month. So, my guess was right, I believe.
I think that the periodicity of the indicator should be selectable in the Compose Model section, and it should be separate from the Rebalancing Frequency (call it Indicator Period, and default it to Daily). The Data Scale parameter in the Settings should just be for the Graphing. All of these suggestions apply only to the Rotation models, I think. I haven't run into this issue in other models yet, or haven't noticed them.