Value & Momentum Breakouts - 1st YTD Report For 2017

Summary

• A look at the 2017 performance YTD for the four primary Value & Momentum Breakout portfolios.
• An explanation of how each of the breakouts and value selections are measured and obtained.
• Charts and graphs of key performance indicators through Week 38
• The inevitable pitch to consider becoming a subscriber now that the forward testing project has moved to a subscription based service.

Here are the YTD results of the four models I offer regularly and in their entirety to subscribers.  These are each very different models, that serve very different purposes over different investment time periods. 

• The Momentum Breakout Portfolio is a weekly replenishment model for short-term high volatility stock returns. 
• The Forensic (Negative/Positive) Portfolios are experimental value portfolios based on two top bankruptcy models and one earnings manipulation model from the field of financial forensics.
• The Piotroski Enhanced Value Portfolios are monthly, buy/hold, medium to long term portfolios based on documented abnormal returns in financial literature.
  
• The Russell 3000 Anomaly Stock Portfolio is another experimental portfolio based on documented abnormal returns from the annual reconstitution of the Russell 3000 stock index.

Model Returns YTD

Model (Value / Breakout)	Returns YTD	Number of Periods*
Weekly Momentum Breakouts	40.39%	28 Weeks
Forensic Negative Selections
July (-) Forensic Portfolio	33.97%	3 Months
Aug (-) Forensic Portfolio	23.02%	2 Months
Sep (-) Forensic Portfolio	9.42%	1 Month
Forensic Positive Selections
July (+) Forensic Portfolio	11.75%	3 Months
Aug (+) Forensic Portfolio	7.43%	2 Months
Sep (+) Forensic Portfolio	0.78%	1 Month
Piotroski Enhanced Value
August Portfolio	4.77%	2 Months
September Portfolio	4.74%	1 Month
Russell 3000 Anomaly Stocks	13.72%	2 Months

^{* All YTD return results are shown through Week 38 - this creates partial returns for the September month portfolio results.}

Now before any critics descend upon these results, unleash nasty projectiles, or equate all the methodologies to monkeys throwing darts at the stock pages, let's look at the different models including their flaws and strengths. I've included a couple foundational investing principles worth reading that are applicable to the stock selection models I offer:

1. "All Models Are Wrong, Some Are Useful."  Applying models in the most appropriate way is the key to judging utility.
2. "I've never seen a bad backtest." (.pdf) ~ Unnamed Equity Derivatives Professional.  If you're following a system with "proven success" based only on backtesting, you're wasting your time and probably your money too.  Forward testing remains the best way to assess reliability and utility.  You also have my permission to slap me if I ever annualize out my future returns in any of my marketing.
   
3. Even if something works, if it doesn't help you, don't use it.  I'm not making the case that any of the models presented here are invaluable or even applicable to your investing goals.  You need to carefully and independently evaluate that for yourself.

If you want to join me in applying these models that have personally provided me value, then please come aboard and let's use these stock selections in the best way possible. 

Weekly Momentum Breakouts

This model was borne out of my doctoral dissertation building on the methods of Altman (1964) and Taffler (1984) who applied predictive discriminant analysis to classify firms at risk of bankruptcy.  I extended their approach by initially evaluating 24 variables simultaneously from behavioral, fundamental, and technical theories to classify different conditions of stock price momentum.  I have since greatly expanded the original study to more than 40 variables over significantly longer periods time, but nothing beats the raw application of actual forward selection.  So a desire to forge a forward testing track record is what prompted me to start this momentum breakout study on the Seeking Alpha platform at the start of 2017. My selections and results are all documented herein.  The full set of current selections and performance results are naturally provided to subscribers as part of the value package.

What is this Breakout method trying to accomplish?

These Breakout selections are focused on a very small population of stocks in the market capable of +/- 10% daily and weekly moves.  For example, on a typical day fewer than 40 stocks out of more than 4,300 stocks (non-OTC, ex-funds, share price > $2) gain more than 10% in a day.  That represents a segment of less than 1% of the stocks on the NYSE, NYSE Mkt, and NASDAQ.  This method is trying to improve that frequency of occurrence using the strongest variables that emerged from the multiple discriminant analysis in my ongoing research.  

While variables were selected and parameters calculated for each of seven different momentum conditions both daily and weekly in the original study, the Breakout Model offered here is primarily focused on Segment 6 (Positive Momentum Acceleration) as shown in the graph below. 
The bar chart below shows the mean % of price performance of the original classification study baseline used to identify the best stock selection criteria.  Again, compared to the majority of stocks in the market that move less than +/- 5% or 10% (segments 3, 4, and 5) in a week or month, the more volatile segments (1, 2, 6, and 7) represents a very small sample (less than 5%) of the stocks in the US exchanges.

Additionally, as I run the selection screens I try to limit selections to stocks with prices greater than $2/share and preferably greater than $5/share. I understand that this higher share price range is more attractive to most investors and avoids penny-stock speculation and higher risk and volatility.

However, reducing risk and volatility with a $2/share price limit also greatly reduces the available sample of stocks capable of producing greater than 10% short term returns. For example, when a $2/share limit is applied the number of 10%+ gainers is reduced by 44% on a typical day. At a $5/share cut-off as much as 68% the 10%+ gainers are no longer available for selection. Because I don't want to make this a penny-stock chasing model, this is the challenging trade-off that occurs in this segment.  As challenging as the effort is to find positive +10% moves in a week, the sample of available stocks with -10% moves in a week is even smaller on average.  Keep that in mind as you leverage the selections.

First Phase of Testing

In the first phase of forward testing from fiscal week 2 through fiscal week 17, I selected 8 long positions, 4 short positions, and 4 dividend positions with a high potential for accelerated momentum.  Some astute readers observed that the data source used for dividend stocks was not accurate on several dividend statements and I dropped the dividend selections from the study for phase 2.  Now with additions and improvements in my data sources I am still considering a return to the dividend selections for future testing.  However, the very short-term weekly replenishment model of these breakout picks is really not intended or well-suited for the long term approach of dividend investment strategies.  I will have to consider the best way to present future results.

Second Phase of Testing

In the second phase from fiscal week 25 through fiscal week 35, I modified some of the parameters, removed the dividend study, and improved some formatting to more easily conduct Chi-square testing. For example one of the parameters I changed was to avoid selecting stocks with a share price under $2, even though penny-stocks represent a high proportion of all the double-digit percentage gainers on any given day.  Despite some of these limitations and as a result of some improvements, the results continue to have statistically significantly higher breakouts than expected market outcomes.  You can review the statistical tests on the links to the different test phases, all of which have remained fairly consistent with the frequencies obtained in the more robust statistical testing in my doctoral study. 

Breakout Frequency Table

The frequency table below shows the number of intraweek percentage gains of all the long positions for Phase I, Phase II, and the current period by each category of price movement 5%+, 10%+, 15%+, 20%+, and 30%+.  After Week 35, I accepted an invitation to become a marketplace subscription based author and subsequently had to change the content offered in my public articles.  The list of 8 long selections continues every week, but is only available in its entirety along with short positions to subscribers as one of the carrots offered for risking a membership here with me.

So I think the best way to measure the significance of the breakout model's results is to track the number of stock price movements greater than 10% during each week and compare it to the market rate of 10% gains in the same period among the same population of stocks in the NYSE, NYSE MKT, and NASDAQ.  Or in this case, compare the frequencies of greater than 5%, 10%, 15%, 20%, and 30% price moves to see if each % category is occurring at frequencies in excess of expected market performance over the same periods. 

Some weeks had fewer than 5 trading days and that affected the results as well.  In week 27 for example, I skipped making stock selections as there were only 3 trading days to test the model.  I applied the non-parametric statistical test, Chi-Square, because stock returns are neither independent nor normally distributed and parametric tests are not adequate for reliable results with such data distributions.  In all breakout categories (5%+ to 30%+) the frequency of stock selection was statistically significantly higher in the breakout model sample population than in the same population of stocks in the US exchanges.

This is illustrated in the frequency chart below by category through Week 38 using the data in the frequency table above.  So far through Week 38, more than half of all portfolio selections (52.2%) have gained at least 5%+ intraweek compared with an expected 14.45% of the broader stock market.  The frequency of occurrence in the breakout model of 10%+, 15%+, 20%+ and 30%+ compared to the broader stock market of all the same type of stocks (ex-funds, non-OTC, share price > $2) is higher with statistically significant differences:

One SA author explains that if you can capture fifty consecutive 10% gains you can turn $10k into $1,000,000. The odds are quite long, but so far I've collected a smattering of 62 stock moves greater than 10% in 28 weekly selections.  So please just avoid any downturns and let me know if that strategy to $1,000,000 ever works out for you!

The point to be made is that the Momentum Breakout model was designed to increase the frequency, i.e. the rate over time, for selecting stocks that make greater than 10% moves.  I've talked a bit about the increased rates of 5%+, 10%+ and 15%+ etc. moves compared to expected market rates, but I want to touch on the component of time.  Time in this model is arbitrarily set to one week.  We know from Jegadeesh & Titman (1993) as well as Fama & French (2008) that price momentum is based on the observed phenomenon,

where stocks with low returns over the last year tend to have low returns for the next few months and stocks with high past returns tend to have high future returns (Fama & French, 2008, p. 1653)

The question that still lingers is what is the optimal period of time for the best results? 1 year? 1 month? 1 week? Some people have looked to use the Breakout Forecast for buy/sell points or to produce a pass/fail percentage of picks that were missed each week to assign a score to the model that way.  As a datahead I can get caught up in all those measurement schemes too.  It's just not how the model was intended to be applied or appropriately measured, though I am open to suggestions to make performance results as accurate and meaningful as possible.

Evaluating Momentum Holding Periods and Buy/Hold Strategy

Since I still don't know what period of time is the optimum holding period for the highest frequency of greater than 10% gains, I considered another perspective. I know that when using the arbitrary period of 1 week (4 or 5 trading days) this model is consistently outperforming the market at more than 4 times the expected market frequency.  So what if I take a look at longer momentum survivors? Can we see decay in performance among the top stock selections?  Can I simulate a buy/hold approach knowing we would all agree to the dump the worst stocks as early as possible?

So at the completion of testing Phases I and II earlier this year, I decided to take a look at momentum of the top 25% (top 2 of 8 stocks) each week.  This approach gave me a glimpse that momentum among the top sustained breakout stocks does appear to continue and quite significantly.  Yes, it is only the top 2 of 8 stocks from each of the 25 weeks.  Yes, this is only 25 intervals of weeks in an arbitrarily selected and relatively good year.  However,  these kind of returns from such a small sample of 8 stocks appears significant, especially considering different durations of weekly returns represented from the two testing periods in the table below. 

Still, I'm not advocating a blind buy/hold approach to these picks and strongly caution against that due to the high volatility inherent in these selections.  The table below shows the top 25% of each week from the start of the breakout study through Week 35:

Perhaps the best application of these volatile breakout selections is to set percentage gain targets over weekly (or longer) periods keeping past probabilities in mind and see if there is a more optimal profitable return period than one week.   Determining what the optimum holding period should be for these stocks will definitely be a profitable undertaking for a future study, limited only by time and resources.

Indexing the Breakout Momentum Returns

Meanwhile, I continue to advance the breakout model as originally designed in my dissertation, as a system that generates more than 4 to 5 times the expected market frequency of weekly breakout returns.  However, most people don't think in terms of frequencies and really prefer to assess the selection risk with a visual form of indexing against a benchmark like the S&P 500.  Which leads to accommodating another form of measurement:

So after I piece together the end-of-week results for each of the momentum breakout selection weeks while eliminating all the intervening gaps and recording the appropriate S&P 500 returns for each week, I get the resulting performance chart above.  The average weekly breakout return is 1.44% with the following sample standard deviation and variance:

It does not take into account the transaction costs of over 200 trades (8 deliberately different stocks every week), entry/exit slippage, taxes, fees and other factors.  Nor does it permit any dumping of obvious intraweek losers, nor allow the selling of any 20% gainers before the end of the week -- but it does give a view of performance against the S&P 500 that some profitability may be achievable from week to week. 

The Forensic Portfolio Selections

The next set of models I offer are the Forensic Model value selections.  The positive and negative scoring forensic models arose out of curiosity about whether the combined application of two top bankruptcy models and one earnings manipulation model could produce abnormal returns in short-term forward testing.  The selections for both the positive and adverse scores are updated every month using the highest and lowing scoring stocks across all three forensic models (Altman Z-score, Ohlson O-score, and the Beneish M-score).  Unlike the Breakout Model the Forensic Model allows for continuation of the same qualifying stocks from one period to the next.

The working assumption is that the negatively scoring stock portfolio will have a correspondingly adverse stock return over a medium to long term investment period.  So far it is clear however that the portfolios with the adverse bankruptcy and earnings manipulation scores are substantially outperforming the positive portfolios expected to have the lowest risks of bankruptcy or earnings manipulation based on the forensic algorithms:

Value Model	Returns YTD	Number of Periods*
Forensic Negative Selections
July (-) Forensic Portfolio	33.97%	3 Months
Aug (-) Forensic Portfolio	23.02%	2 Months
Sep (-) Forensic Portfolio	9.42%	1 Month
Forensic Positive Selections
July (+) Forensic Portfolio	11.75%	3 Months
Aug (+) Forensic Portfolio	7.43%	2 Months
Sep (+) Forensic Portfolio	0.78%	1 Month

^{* All YTD return results are shown through Week 38 - this creates partial returns for the September month portfolio results.}

Public articles on the Forensic Value Selections for July, August and September are linked, however as you might expect, the full performance details and current selections are now only available to subscribers as part of the value offering.  It's a high priority to provide the most value to people willing to risk their own time and money on my work. 

The Piotroski Enhanced Value Portfolio

The next set of models I regularly offer are the Piotroski Enhanced Value Portfolio selections with a one-year investment target.  This Piotroski F-score model has emerged from the financial literature as one of the better value stock selections tested by financial scholars over the past 17 years.  I have added a few minor enhancements to avoid penny-stocks and adverse financial outliers.  Running multiple one-year portfolios from a new Piotroski selection each month may also provide a reasonable forward testing approach of the effectiveness of this standout algorithm. 

The early results so far show the potential for a solid low risk annualized return.  There is some slight overlap in the month to month selections and that may add to the credibility of the F-score selection process.  The YTD returns:

Value Model	Returns YTD	Number of Periods*
Piotroski Enhanced Value
August Portfolio	4.77%	2 Months
September Portfolio	4.74%	1 Month

^{* All YTD return results are shown through Week 38 - this creates partial returns for the September month portfolio results.}

The Russell 3000 Anomaly Stock Portfolio

The final model I currently offer on a regular basis is the Russell 3000 Anomaly Stock portfolio.  This is another well documented anomaly that generates abnormal returns that I recently came across in the process of identifying breakout stocks.  I decided to take a sample of the top 10 best performing stocks added to the Russell 3000 index in their annual reconstitution process every June on the basis that these might be among the best momentum performers throughout 2017. 

• Measurable price effects of the Russell 3000 annual index reconstitution have been documented for both stock additions and deletions (Chang, Hong, & Liskovich, 2013).

• Similarly, positive S&P 500 Index effects have been measured at around 5% to 7% in the month following the addition announcement with a large fraction of the gains remaining permanent.

• Stock additions to the Russell 3000 index leads to a "dramatic increase" in trading volume ratio in the month of June (Chang et al., 2013).

Historical returns of the Russell 3000 reconstitution stocks have been tracked closely since 2000 and are summarized in my initial study here.  So far the early results show some promising returns YTD:

Value Model	Returns YTD	Number of Periods*
Russell 3000 Anomaly Stocks	13.72%	2 Months

^{* All YTD return results are shown through Week 38 - this creates partial returns for the September month portfolio results.}

Concluding Thoughts

I am as interested in finding alpha as the next investor or financial researcher.  I continue to invest considerable time, resources, and energy in searching through different anomalies, models, and financial literature on the latest market discoveries.  All of my systems are applied in my own private equity fund trading and I have realized substantially better results from what I have learned over the years.  Behavioral, technical, and fundamental analysis are all of great interest to me as I am certain they each comprise critical pieces of the puzzle to answer the question of, what really matters in assessing the best potential return on investment of any asset?

My forward testing project has now merged into a subscription based Marketplace offering here on Seeking Alpha.  I'm still running financial experiments and thoroughly enjoy finding good ways to add value, especially for those of you willing to purchase a subscription membership and try to get every advantage in your market trading.  I would be glad to have you join me as I continue this journey wherever it leads!