June 23, 2007: Visualizing a market strategy
Visualization recap A chart of a market strategy. How are we doing? Current ETF rankings. Next column teaser.
Visualization recapIn the last column we looked at simple data visualization and demonstrated its power. If you missed that column, here's a link to it.
If I had to summarize that column in one paragraph, I would put it this way:Data visualization is a powerful tool to help gain insight.
Insight begets conviction.
conviction begets action
And only action can (hopefully) lead to out-performance.
There's a hidden assumption in there: when making investment decisions, one must focus on raw-data grounded in facts and reality, and ignore herd mentality, prevailing sentiment, rumors, newsbytes , or anything else that might distract from the underlying facts. One should try to examine as many relevant facts (data points) as possible, and nothing but the facts.
Since the stock market and economics in general, are a source of so much data, I keep wondering why the financial mass media keeps showing the same old simplistic charts: A single company stock price over time, or a single histogram of some economic indicator? Where's the big picture? There must be better ways to look at all the markets in one nice chart showing all sectors, all countries, many asset classes all in one, by adding color and animating ETF movements over time.
Visualizing a simple market strategy
Visualizing recent market preferences like we've seen last time, has been useful. We now step-up to visualizing market strategies. Doing this can provide insights into why a strategy is working or not. Perhaps it can even convince us that a strategy is sound and likely to keep working. Again, a chart is worth a 1000 words.
A few weeks back, I did some strategy back-testing, with Weka (Waikato Environment for Knowledge Analysis).
Weka is a flightless New Zealand bird. It is also the name a nicely integrated and comprehensive suite of data-mining and machine-learning tools from the University of Waikato, New Zealand. Weka contains a large number of components to import data, filter, clean and transform data, build models, select best features, estimate errors, and even some visualization aids. Weka is free (both as in speech and in beer) and released under the Free Software Foundation's GNU GPL (General Public License), just like Linux.
Below is a snapshot of Weka visualizing a data-set. This data-set has two new ETF columns (features) I've never shown here before. Together, they represent an investment strategy:
- The X axis represents an ETF ranking-function This is the ranking function, or model, we use to target better future returns.
- The Y axis represents the future returns.
How can you possibly visualize a ranking-function? You ask. Well, since our ranking ranks ETFs in order of attractiveness, it essentially assigns a numeric attractiveness score to each ETF data point. This attractiveness score is a number that can be plotted along the X axis in a chart.
Aha, that's obvious now. But how can one plot future performance? Well, one can't, or we wouldn't be here, but one can get close. It is called back-testing. If you go back 3 months, and run the ranking functions on the data as it existed at that point in time, you can also apply hindsight, and look at the future 3-month return, from that point in time till today.
Of course, there's no guarantee that what worked in the past 3-months will work in the next 3-months, but:
- That's about the best we can do when building models
- As we've seen in several previous columns, strong market preferences can last for a very long time. If we can figure out what works, it is reasonable to assume, although not with 100% certainty, that it will continue to work, at least for the near future.
Moreover, we can repeat this experiment for longer time frames, or start from different points in time, or look at other strategies, for comparison, and so on. In short, we have a very powerful tool in our hands. And this tool can be applied to many, and diverse, future cases.
Let's take a look at this Weka visualization screenshot:
Each point represents some ETF at a random point in time in the past. Let's go over the exact coordinates of each point:
- The X axis represents the score given to the ETF at some point in time by the MMVR (Moderate momentum, Value, Risk) ETF picking strategy. Higher scores mean the ETF was found as attractive buy by the strategy ranking function at that time.
- The Y axis represents the future 3-month percent return of the ETF after the point in time mentioned in X. The points at the upper half of the chart, especially those orange points in the middle, have appreciated significantly in the 3-month period following their being ranked by the MMVR strategy.
- The size of the X in each point, represents the technical risk (sigma, beta) in the year prior to the pick. The smaller the 'x', the more predictable, and less volatile the ETF was at the time of the ranking. Big X's mean performance has been all over the place, up and down with seemingly random direction changes. In other words: the big X points may have done great in the "future 3-month" period, but have been much less predictable over the course of the preceding 1-year period.
Note that the number of points here is larger than the number of ETFs because the sampling was done for many different points in time to ensure randomness. IOW: each ETFs appears multiple times in the chart representing different points in time. At different times, an ETF can have somewhat different score it was given by the ranking function of the strategy, a different technical risk, and of course, different future 3-month return.
The most salient point here is that MMVR, by design, is not trying to maximize future returns, but risk-adjusted future returns. This is why it seeks moderate momentum rather than high momentum, and this is why it is trying to avoid volatile ETFs.
To borrow a concept from artillery. MMVR is seeking long range trajectories, as opposed to high angle (> 45 °) trajectories which gain altitude fast, but then abruptly change direction, and lose altitude just as fast.
The assumption is that by limiting risk, we limit potentially unpredictable downsides. It makes the product of (Gain x Chances_Of_Gain) more uniform and hopefully higher on average, over many pick instances, in the long term.
As can be seen in the chart, most of the points with the highest MMVR score are concentrated around a "latitude" of about +5% to +11% next 3-month (future) return. This (very) roughly translates to about 30%/year. Of course, this model, and its data cover a period within a bull market. A +30%/year return should not be considered a long term, sustainable figure. Still, 30%/year, even during pure bull markets, is a very respectable and desirable goal.
MMVR avoids both:
Since the chart is representing different points in time, some of the biggest winners happen to be among the biggest losers sometime later. MMVR looks for the most stable and straight lines, which is the source of some of its strength.
- High flyers (some of which appreciated close to +30% in a 3-month period). These often have a tendency to crash and burn afterwards, and
- The big losers (some of which lost over 8% in 3-months in the midst of a bull run)
What would have happened if we picked a strategy that is willing to take more risk and buy more high-flyers? Or maybe some crash-and-burners, perhaps after they crashed? What if we picked more components or less components to make up our portfolio mix? What are the error rates of the model when trying to use it as a predictor on data that was not used to build it? These are all excellent questions which I will try to partially answer in future columns.
How are we doing?Our model portfolio SZDM continues to outperform the 3 major US indexes since its inception, although the gap has narrowed due to record performance of the Dow and the S&P500 last spring. In the past couple of weeks, including in the recent down week, we are back outperforming both the S&P500 and the DJI. Earlier in June we've also hit a milestone of being up +10% since our inception less than 5 months ago (January 28, 2007). The recent down week has reduced our total gain to +8.47%. Our annualized (extrapolated) returns stood at 28.32% on June 15th. Some recent figures can be found here.
Current ETF rankingsHere are last Friday's (June 22, 2007) rankings (top 40 ETFs among over 400):Using mmvr ranking method on 20070622 1 3.0973 EWY iShares MSCI South Korea Index 2 3.0742 TTH Telecom HOLDRs 3 3.0014 VAW Vanguard Materials VIPERs 4 2.8606 EWG iShares MSCI Germany Index 5 2.8470 ADRU BLDRS Europe 100 ADR Index 6 2.7763 PRFE PowerShares FTSE RAFI Energy 7 2.6968 ADRD BLDRS Developed Markets 100 ADR Index 8 2.6138 RPV Rydex S&P 500 Pure Value 9 2.6126 FEZ streetTRACKS Dow Jones Euro STOXX 50 10 2.5769 EZU iShares MSCI EMU Index 11 2.5693 PKB PowerShares Dynamic Building & Construction 12 2.5430 PXE PowerShares Dynamic Energy Exploration 13 2.5411 RZV Rydex S&P Smallcap 600 Pure Value 14 2.5397 RYE Rydex S&P EqWght Energy 15 2.5167 XLE Energy Select Sector SPDR 16 2.5163 PWP PowerShares Dynamic Mid Cap Value 17 2.4888 DLS WisdomTree Intl SmallCap Dividend 18 2.4810 EWT iShares MSCI Taiwan Index 19 2.4678 DKA WisdomTree Intl Energy 20 2.4631 DEW WisdomTree Europe High-Yielding Equity 21 2.4392 DEB WisdomTree Europe Total Dividend 22 2.4339 DFE WisdomTree Europe SmallCap Dividend 23 2.4171 VOX Vanguard Telecom Services VIPERs 24 2.4121 IEO iShares Dow Jones US Oil & Gas Ex Index 25 2.3862 IXP iShares S&P Global Telecommunications 26 2.3791 DBN WisdomTree Intl Basic Materials 27 2.3601 EWQ iShares MSCI France Index 28 2.3445 RFV Rydex S&P Midcap 400 Pure Value 29 2.3424 IEV iShares S&P Europe 350 Index 30 2.3344 IYZ iShares Dow Jones US Telecom 31 2.3050 EWK iShares MSCI Belgium Index 32 2.2910 VDE Vanguard Energy VIPERs 33 2.2765 PPA PowerShares Aerospace & Defense 34 2.2705 EWN iShares MSCI Netherlands Index 35 2.2540 PRF PowerShares FTSE RAFI US 1000 36 2.2306 IXC iShares S&P Global Energy Sector 37 2.2248 EWC iShares MSCI Canada Index 38 2.2035 PTE PowerShares Dynamic Telecom & Wireless 39 2.1966 DIM WisdomTree Intl MidCap Dividend 40 2.1961 XOP SPDR Oil & Gas Exploration & Production
Next column: ETF strategies Battle RoyaleNow that we saw one strategy "in action", we're ready to look at a large number of strategies. In the next column, I'll present an experiment I did by back-testing a large number of different simple investing strategies to pick ETFs. Which one did best? By which criteria? Should we employ more than one strategy in different circumstances? Can we do better than MMVR based on more data and models?
Some surprising and exciting results are coming. Stay tuned.
As always, I hope you found this column useful. Feedback, good or bad, is always more than welcome.