This page describes why I want Equities Lab to be able to ignore data. So, in a Trading model, I have a perfectly normal, correct looking graph: GE. I’ve plotted the EPS_1Q. This graph is actually correct — GE Finance was split off and a large write down was taken.
However, when I do log scale, to see how the stock price and earnings are growing together, or not, I get:
Which is correct, given that I haven’t put a percent on eps. Time to put a percent on EPS.
What gives? It turns out, if you look closely at the top line that the eps doesn’t start until 1996 or so. Time to change the range and fix that.
Well, this helps, but the large negative bar at the end is confusing, and kills log scale (reasonably enough).
Now I can fix this one, because the outlier is at the end. All I do is adjust the scale to exclude 2015.
Success! But what if the outlier is in the middle, such as the outlier in jan 2000? It would be awfully handy to be able to just “nuke” it, especially if it’s throwing off the scale. Ideally nuked data wouldn’t show up in the report, so one could get an idea as to the average earnings increase, after nuking the skyscrapers there.
In backtest by time, we have the same desire to nuke data (again time intervals). Notive how the extremely tall (and low) bars on the end wash out the scale for the rest of the chart.
Notice how much clearer the chart gets once I cheat and nuke the bars at the end (by changing the dates)
Another possible use case would be looking at backtest returns. Consider Great Value Score top 10 (rebalanced monthly).
If we want to validate that that ginormous bar at 2009 isn’t skewing our results we can(as we do in the book, simply outlaw 2009. But this is decidedly less than ideal, because the benchmark moves whilst we stand still:
It also changes the way the report works, since we have 12 months of 0% performance, making our screen seem smoother than it is.
Future extensions of this idea might run to deleting poorly behaved scatter plot points, or stocks from the matches or positions tables, so as to be able to see good graphs in the results. This is where we’d need another field ExtraExclusions that is and-notted, or ExtraRequirements, that is anded.