Saturday, September 26, 2009

Mrkt_Rwnd: Home of "ETFRanks"

Admittedly, the ETF Rewind service's greatest strength is also its weakness:  a tremendous amount of information in a very compact presentation.  Okay, maybe too much information for some.

To address this issue, earlier this month I began incorporating a proprietary composite scoring system for all 200 or so tracked ETFs using many of the nightly spreadsheets' key heuristics as presented in the service's introductory video series. I call the resulting scores 'ETFRanks'.  As discussed below, the ranks appear to be strong selection criteria for active investors using an intermediate time-frame.

Rank Construction

Here is a brief overview of the factors included in the calculation of the ranks:
  1. Long-Term Trend
  2. Intermediate-Term Trend
  3. Relative Strength
  4. Risk-Reward Profile
  5. Short-Term Oversold/ Overbought Oscillator State
  6. Relative Volume/ Liquidity
To achieve a rank of 10, the most 'bullish' score, an ETF would have to be strongly and consistently upward trending on both an absolute and relative basis, and be at least slightly oversold on a very short-term basis.

Cross-Sectional View of Market Status

Each night I create a histogram of the resulting ranks for all the equity-based indices, like so:

The modality of this chart provides me with a strong sense of overall market status.  Note how the markets' recent pullback within an environment of upward strength has pushed many of the ETFs into the highest scoring quintile.  Also, it is abundantly clear with the strong rightward skew how preternaturally strong this market has been.  Lastly, it is also a good touchstone against which to compare any individual ETFs of interest.  In the future, I will make note of any major changes to the modality of this chart in my weekly ETF Rewind summary.

Out-of-Sample Performance

As a first-pass test of the out-of-sample efficacy of the ranks, I ran the scores for all the ETFs as of twenty-days ago, and then again as of sixty-days ago. The upper panels of the chart below show the backward looking/ in-sample performance of the ranks as of the date of those respective runs for the two periods (left  = 20-days; right = 60-days):

The forward, out-of-sample performance for the same ETFs as ranked in the upper half during the ensuing twenty- and sixty-day periods is then shown in the lower half.  As you can see, the relative out-performance/dispersion of the higher ETFRank'ed securities held up admirably.

This has been a very trendy period within a relatively stable beta seeking regime, so I will grant that I need to conduct the same type of testing over a far greater sampling for a thorough evaluation.  However, I strongly suspect the ETFRanks will hold up well. 

[See also Home of Mrkt Metrics]

As a postscript, this is my personal attempt at an inclusive ranking system.  Yes, it's built into the ETFR, and yes I'm keeping it 'proprietary', but I promise that it isn't rocket science and there are no 'magic formulas' involved. Rather,  I hope between the article and the comments below that you find enough raw material and inspiration to attempt your own.


Jim Kane said...

Have you given any thought to turning your ranking system into a trading plan like buy equal $ amout of all 10 rated etfs rebalance weekly, x% stop loss. or something completely else....

David Varadi said...

very impressed with the linearity and stability of performance by rank in and out of sample buckets.
excellent work.

Tamás said...

what do you mean by risk-reward profile?

Damian said...

Somewhat similar, in terms of approach, to the Fund-X approach. Having worked on these type of rankings for a long time, the toughest part is dealing with the change from "bull" to "bear" - meaning that these systems do very well in bull markets and not very well in bear markets. I think it was David Varadi (above) who actually proposed using adaptive momentum time-period analysis.

jgpietsch said...

@ Jim - A very fine idea indeed.

@ Tamas - Risk-reward may be measured through Sharpe/ Semi-Sharpe/ Sortino/ Omega and the like. Essentially, returns relative to the volatility risk encountered in generating those returns.

@ Damian - To deal with this very issue, the rankings contain both absolute and relative components. Scores under 5, for instance, are bearish irrespective to how one ticker compares to another. Also, looking to multiple time-frames simultaneously can help to better pivot the metrics to timely respond to changing conditions. Nevertheless, I understand this challenge well and it's an insightful comment.