Real work done by Herb Haynes ?
In Part 5 of this series of posts, we started our report of LRPC back-test results by simply adding the Projection-Convolution parameters to the original Kipling momentum ranking system. This meant that we retained the requirement that an asset should be ranked higher than SHY for consideration to be included in the portfolio. We have always emphasized/promoted this as a mandatory requirement for previous momentum systems since back-tests showed this to be the primary strength of the system i.e., it acts as an effective filter to keep us out of bear markets and/or major pull-backs. However, now that we have new parameters (P and C), can this SHY filter requirement be relaxed/eliminated?
One parameter that we did not specify in Part 5 was the review date. In all these tests we use the closing price on the last trading day of the month – the End-Of-Month (EOM) date. We shall be addressing the issue of review date luck towards the end of this series of posts. For now, and for easy valid comparison purposes, we are using the fixed EOM dates.
In this post we run exactly the same series of tests that were run in the Part 5 report i.e., using exactly the same parameters, except that we relax the requirement for an asset to be ranked higher than SHY. The only mandatory required condition (also required in Part 5) is that the value of P minus C (P-C) must be positive – or that the Projection value/curve must lie above the Convolution value/curve (Part 3). Thus, a positive value of P-C replaces our ranking relative to SHY as our primary exclusion filter to avoid significant draw-downs.
As before, all tests are run from 12/31/2007 to 11/30/2017 (almost 10 years).
Let’s look first at the total returns:
In the above figure you will note that we have separated the table into two sections – as defined by the yellow stair-step line from the top left corner to the bottom right corner. The relevance of this is that the “system” is designed based on the assumption that recent changes in market conditions (“acceleration” – see Part 3) can be used to improve system performance. This implies that the Convolution look-back period should not be greater than the Projection look-back period – or that parameter variables should lie in the bottom left section of the table.
Of course, this does not mean that parameters out of this range (top right section) cannot be chosen to generate comparable performance – just that this is more likely to reflect data mining/curve fitting than to be a result of “robust” system design.
If we compare the returns reported in Part 5 (with SHY filter) with the returns in these tests we note two things:
- The top percentile “sweet spots” move.
- Total maximum returns in the “sweet spots” are higher than when SHY filtering is enabled.
If we convert to Compound Annual Growth Rate (CAGR) we get the following picture:
Note that the original L1=150, C1-110 “sweet spot” moves slightly to L1=140, C1=130 with very similar returns/CAGR (~11% CAGR). However, if we follow the values when L1=100, C1=90 we note a significant improvement in performance as compared to values when the SHY filter is enabled (12.9% CAGR vs 10.2% CAGR). This is a huge difference when compounded – total returns over ten years increase from 163% to 234% – 70% higher.
Now, let’s take a look at portfolio volatility:
Here, we see that the highlighted cells fall in the most preferred (green – low volatility) areas for the larger (~140 day) look-back values and in the “moderate” range at the shorter (~100 day) look-backs
So, let’s try to balance return/risk and look at the Sharpe Ratio:
Our highlighted cells are now showing a very appealing ratio of greater than 1 and this remains “robust” compared to values around them. In the battle of return vs risk return wins out suggesting better risk-adjusted returns.
For investors more concerned about draw-down (DD), here’s what the maximum draw-downs look like for the Rutherford portfolio in these tests:
Our highlighted cells remain in the top percentile green areas with DDs in the 11-15% range – lower than equivalent tests that included the SHY filter (15-19%, Part 5).
Let’s balance returns and maximum DD by looking at the MAR ratio:
Here we see that the highest (preferred) ratio occurs at the higher look-back periods (L1=140) rather than at the shorter (L1=100) periods identified by the Sharpe ratio.
So, the decision as to which parameters to use might be influenced by whether an investor was more worried about volatility or maximum DD.
The following HM shows the calculated number of trades in the above tests:
…. very similar to the tests using the SHY filter.
ITA members can make their own decisions as to whether the SHY filter should be turned on or off – this is retained in the LRPC workbook as a user-selectable option.
Based on the above results, all future tests reported in this series of posts will be run with the SHY filter turned off.
As usual, a downloadable PDF of this post can be found at https://www.dropbox.com/sh/ix9tocyjyf4ycl3/AACoRWSYilTa4Eu6sUi60GI3a?dl=0
In the next post we will take a look at the impact, on performance, of changing the Offset look-back parameter (O).
Herb and David