November 21, 2015

Bring More Data

Several months ago we posted an article called “Bring Data” where we showed the importance of having abundant data for system development and validation. This was further reinforced to us recently when someone actually brought us additional U.S. stock sector data. Previously, we only had Morningstar sector data that went back to 1992, which we used to construct our Dual Momentum Sector Rotation (DMSR) model. (S&P sector data also goes back to only the early 1990s.) DMSR was shown in my book as one example of other ways you might use dual momentum.

When we were given equivalent Thompson Reuters U.S. stock sector data back to 1973, we immediately extended our DMSR back test to include this additional data. After incorporating the new data, DMSR still looked considerably more attractive than buying and holding the S&P 500 index. But one could argue that the performance of our dual momentum models using broad-based equity indexes, such as Global Equities Momentum (GEM), now look better than DMSR. Here are the comparative performance figures from January 1974 through October 2015:

S&P 500
Average Annual Return
Standard Deviation
Sharpe Ratio
Maximum Drawdown

Results are hypothetical, are NOT an indicator of future results, and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. Please see our website's Performance and Disclaimer pages for more information.

Because the monthly correlation between GEM and DMSR is only 0.59, sector rotation can still have a useful but modest role to play in a diversified equities-oriented portfolio. But DMSR is not the best choice as a core portfolio holding. Sector rotation programs that use data no further back than the early 1990s to develop their models may be in for a rude awakening someday if future drawdowns are higher and returns are lower than they expect based on back testing with a limited amount of data.

Along the same lines, there are also momentum-based portfolios popping up on the internet all the time now, some even labeled as “dual momentum,” that are modeled on the basis of only 10 or 15 years of ETF data. Momentum may be robust enough that future results won’t suffer much because of this. But those who think they are constructing optimal models this way are just fooling themselves. Overfitting modest amounts of data is one of the most pernicious problems in the development of investment models. Those who do this may argue that the markets change over time, so the best model parameters from years ago may not be as relevant as today’s best parameters. This may be true. However, what is also true is that today’s parameter values are also likely to be sub-optimal when moving forward in time. The following chart from my book, Dual Momentum Investing, shows what I mean:

 Chart courtesy of Tony Cooper

The S&P 500 is highlighted in different colors for each 15 year period. You can see that the latest period, 1999-2013, looks different from the preceding period, 1984-1998. 1999-2013, in fact, looks more like the earlier 1969-1983 period. 1984-1998 is also different from its preceding period, 1969-1983 and similar to the earlier years 1954-1968. If you had used each 15-year period to develop your model, you would have had something unsuited for each of the next 15-year periods. You would likely be better off using all four periods to formulate a model rather than just the last 15-year period. The more data you use, the more likely you are to have a robust model that will hold up reasonably well in the future, even though it isn’t the best fit to any one particular period.

The 12-month look back parameter we use for our GEM and ESGM dual momentum models was found to work well in 1937 by Cowles & Jones. It has been used extensively in momentum research since then and has held up well out-of-sample. But there is a lot more history than that to help give us more confidence in momentum. Let's take a look at some of that now.

We focus on stocks as our core asset since they have historically offered the highest risk premium to investors. U.S. stocks, in particular, have given investors the best long-run returns. Other assets can create a drag on long-run portfolio performance. They also lose some importance as diversifiers once you use a trend following overlay like absolute momentum to help attenuate your downside risk exposure.

The longest back test on stock market momentum is by Geczy and Samonov (G&S). Their 2013 paper called “212 Years of Price Momentum: The World’s Longest Back Test 1801-2012” compared the top one-third to the bottom one-third of U.S. stocks sorted monthly by relative momentum. Over this entire sample period, the top equally weighted momentum stocks outperformed the bottom ones by 0.4% per month with a highly significant t-stat of 5.7. Prior to this study, momentum outperformance on U.S. stocks had been found significant back to 1926. G&S showed that stock momentum was also positive and statistically significant from 1801 to 1926.

G&S also found that stock market momentum was remarkably consistent. In only 2 of the 21 decades from 1801 through 2012 did long-only momentum under perform buy-and- hold, and these were by just -1.2% and -0.7% annually. In all the other 19 decades, momentum outperformed buy-and-hold by an average of 3.8% annually.

This year G&S came out with a new study called, “215 Years of Global Multi-Asset Momentum: 1800-2014: Equities, Sectors, Currencies, Bonds, Commodities, and Stocks.” Here G&S expanded their momentum study to cover six different asset classes, including bonds, stock sectors, and equity indices, which are the ones we use in our momentum models. [1] G&S demonstrated the outperformance of momentum inside and across all asset classes except commodities. Here is a chart from their paper showing the log cumulative equally weighted average of the 6 asset classes plus the cross asset momentum excess returns.
The strongest momentum effect is in country equity indices, which had a long-only monthly excess return over buy-and-hold of 0.52% with a highly significant t-stat of 11.7, compared to 0.29% with a t-stat of 6.4 for individual U.S. stocks and 0.24% with a t-stat of 15.5 for all assets. G&S also show that long-only absolute (time series) momentum outperformed buy-and-hold by 0.15% per month with a t-stat of 11.2.

For those who want to further their momentum education, I suggest you first read the seminal paper by Jegadeesh and Titman (1993) that started the modern momentum renaissance. Next, learn about absolute momentum from Moskowitz et al (2012) or Antonacci (2013). Then follow up with Geczy and Samonov (2015) to satisfy yourself as to the efficacy and robustness of momentum investing based on 215 years of empirical evidence.

[1] Equity indexes are equally as good as individual stocks (or better, according to G&S) in capturing the momentum effect. Indexes are much easier to use and avoid the enormously high transaction costs associated with rebalancing momentum-based stock portfolios.

October 18, 2015

Multi-Factor Investing

Multi-factor investing that combines value, momentum, quality (profitability), or low volatility factors is today’s hot new investment approach. There has been an explosion of multi-factor ETFs recently with eleven of the sixteen existing U.S. multi-factor funds coming to market this year and five of them showing up within the past 60 days.

Multi-factor funds may be a good thing, since single factor funds can have some serious drawbacks.  However, multi-factor funds can also have their own quirks and issues. If the large variety of factors is thought of as the “factor zoo,” then multi-factor approaches may be the “factor circus” with its own collection of silly clowns, dangerous acrobats, and amusing jugglers.

Factor Investing Issues

With factor investing in general there are three potential problem areas: tractability, scalability, and volatility. With respect to tractability, it is well-known that value investing can have long periods of serious under performance. This happened in the late 1990s and also somewhat during the past two years. Not all value investors may be willing to watch this happen without losing patience and giving up on their factor portfolios. To a lesser degree, momentum and other factors are also subject to sustained tracking error.

Scalability has to do with too much money chasing after too few stocks. Factors perform best when you can focus on those stocks having the strongest factor characteristics. For example, van Oord (2015) showed that from 1926 through 2014, only the top decile of U.S. momentum stocks outperformed the market. Stocks below the top decile added nothing to strategy results.

Yet just two out of the twelve large cap U.S. equities single factor ETFs only include stocks that are within the top decile of their factor rankings. For example, the oldest and largest single factor value ETFs are iShares S&P 500 Value (IVE), iShares Russell 1000 Value (IWD), and Vanguard Value (VTV). They hold 72%, 69%, and 50% respectively of the stocks that are in their investable universes. This makes them, to a great extent, closet broad index funds with higher fees.

Their large sizes ($8.3 billion, $23.5 billion, and $34.6 billion, respectively) may impede them from  focusing on just fifty (the top decile of S&P 500 stocks) or one-hundred (the top decile of Russell 1000) value stocks. The same is true with respect to momentum. The largest momentum fund, with over $1 billion in assets, is the AQR Large Cap Momentum Style mutual fund with an expense ratio of 0.45. It holds 532 out of an investable universe of 1000 stocks.  This is a far cry from the top decile of momentum stocks. Large amounts of investment capital may make it difficult for single factor funds to focus exclusively on the relatively small number of stocks that appear in their top factor deciles.

The third problem for single factor portfolios is increased volatility and high bear market drawdowns that accompany value, momentum, and small cap factors. Trend following filters, such as absolute momentum, can help reduce downside exposure with respect to long-term bear markets, but it does little to alleviate uncomfortable short-term volatility. Trend following is also less effective when applied to value factors than when applied to other factors like momentum.

Multi-Factor Solutions

All three of these problem areas for single factor investing – tractability, scalability, and volatility – can be significantly reduced by using intelligently constructed multi-factor portfolios. Multiple factors can obviously reduce tracking error, since it is unlikely that several factors will substantially under perform at the same time. As for scalability, if a fund uses four factors instead of just one, it can handle four times the investment capital without eroding its ability to enter and exit the markets. Finally, the volatility and large bear market drawdown associated with value and momentum factors can be reduced by combining these factors with less volatile ones, such as quality and low volatility.

However, I intentionally included the words “intelligently constructed” when I referred to the potential benefits of multi-factor portfolios. There is some controversy regarding the impact of size on risk adjusted returns [1]. Small cap stocks, while giving higher returns, may add little on a risk-adjusted basis because of their high volatility. When combined with value or with value and momentum, which is what  six of these multi-factor funds do, small cap may be undesirable, since it may aggravate already  high portfolio volatility and potential bear market exposure. Small cap stocks are also more costly and difficult to trade..

It is also surprising that the strongest anomaly by far, price momentum, is included in only twelve of the sixteen U.S. multi-factor funds. Considerable research has shown that momentum is the most powerful factor for generating positive risk-adjusted returns.

The final issue associated with multi-factor funds is their average annual expense ratio of 41 basis points for what are enhanced index funds. This is higher than the Morningstar US ETF Large Blend Strategic Beta expense ratio of 38 basis points and the Morningstar US ETF Large Blend Index expense ratio of 36 basis points. Until just recently, an investor who wanted multi-factor exposure would have been better off creating it herself by combining the single factor iShares MSCI USA Value Factor, USA Momentum Factor, USA Quality Factor, and USA Minimum Volatility ETFs, since these all have expense ratios of only 15 basis points.

New Solution

This situation changed dramatically last month when Goldman Sachs entered the ETF business with an offering called Goldman Sachs Active Beta U.S. Large Cap Equity (GSLC). GSLC is the only multi-factor fund having what I consider an optimal mix of factors: value, momentum, quality, and low volatility. Here is a description of how they determine these factors:

•      Value: The value measurement is a composite of three valuation measures, which consist of book value-to-price, sales-to-price and free cash flow-to-price (earnings-to-price ratios are used for financial stocks or where free cash flow data are not available).

•      Momentum: The momentum measurement is based on beta- and volatility-adjusted daily returns over an 11-month period ending one month prior to the rebalance date.

•      Quality: The quality measurement is gross profit divided by total assets or return on equity (ROE) for financial stocks or when gross profit is not available.

•      Low Volatility: The volatility measurement is defined as the inverse of the standard deviation of past 12-month daily total stock returns.

Even though the fund holds 432 stocks out of an investable universe of 500, it uses a weighting scheme (most multi-factor funds with a large number of holdings do the same) that allocates substantially more of its capital to stocks with high factor ratings. GSLC rebalances positions quarterly and uses a turnover minimization technique (especially useful for momentum stocks) of buffer zones to reduce the number of portfolio transactions. I use a similar buffer zone technique myself with some of my  more active momentum models.

What is especially appealing about GSLC is its low cost structure. The fund came into existence because some of Goldman’s largest clients wanted to invest this way using an ETF wrapper to minimize their tax consequences. Because of this sponsorship, the fund was set up with an annual expense ratio of only 9 basis points.[2] This is the same expense ratio as the biggest and most popular ETF in the world, the SPDR S&P 500 ETF Trust (SPY). GSLC already has $78 million invested in it since coming to market one month ago.

GSLC is not an ideal investment from our point of view, since it doesn’t have a trend following filter like absolute momentum to help it avoid severe bear market drawdown. GSLC is also unable to benefit from international diversification during those times when international stocks show greater relative strength than U.S. stocks. Even though our testing has shown that value and individual stock momentum do not  add  any value to any of our dual momentum models, GSLC might be worth considering as a portfolio addition if it continues to have a low expense ratio, high liquidity, and good relative performance.
Multi Factor Funds

Symbol Factors Assets Stocks Exp Ratio
4 Factor

Goldman Sachs Active Beta U.S. Large Cap  GSLC Value, Mom, Quality, LoVolty $78 m 432 0.09
ETFS Diversified Factor U.S. Large Cap SBUS Value, Mom, Size, LowVolty $17 m 492 0.40
iShares Factor Select MSCI USA LRGF Value, Quality, Mom, Size $5 m 135 0.35
Global X Scientific U.S. SCIU Value, Size, Low Volty, Mom $2 m 489 0.35
3 Factor
Lattice U.S. Equity Strategy  


Value, Mom, Quality

$23 m


SPDR MSCI USA Quality Mix QUS Quality, Value, LowVolty $6 m 624 0.15
JP Morgan Diversified Return U.S. Equity JPUS Value, Mom, Quality $11 m 561 0.29
John Hancock Multifactor Large Cap JHML Value, Mom, Profit $79 m 772 0.35
AQR Large Cap Multi-Style (non-ETF) QCELX Value, Mom, Profit $1.2 b 338 0.45
iShares Enhanced U.S. Large Cap IELG Value, Quality, Size $71 m 109 0.18
PowerShares Dynamic Large Cap Value PWV Value, Mom, Quality $927 m 50 0.58
FlexShares U.S. Quality Large Cap Index QLC Quality, Value, Mom $3 m 120 0.32
Gerstein Fisher Multi-Factor Growth Equity (non-ETF) GFMGX Size, Value, Mom $227 m 298 1.03
2 Factor

ValueShares Quantitative Value QVAL Value, Quality $47 m 41 0.79
FlexShares Morningstar U.S. Market Factor Tilt  TILT Value, Size $740 m 2249 0.27
Cambria Value and Momentum VAMO Value, Mom $3 m 100 0.59

Nothing contained herein should be interpreted as personalized investment advice.  Under no circumstances does this information represent a recommendation to buy, sell or hold any security. Users should be aware that all investments carry risk and may lose value. Users of these sites are urged to consult their own independent financial advisors with respect to any investment.

[1] See Israel and Moskowitz (2012), for example, who say that the alpha from size are statistically weak, and AlphaArchitect (2014), who explores the empirical side of small cap stock performance.
[2]  GSLC has an annual fee waiver of 15 basis points until September 14, 2016, after which time it has the option to raise its expense ratio up to 0.24. Whether or not fees are raised often depends on asset growth.

September 8, 2015

Book Review: DIY Financial Advisor: A Simple Solution to Build and Protect Your Wealth

I have always looked favorably upon do-it-yourself investing (DIY). It was a prominent feature of my own book. So I’ve been looking forward to DIY Financial Advisor: A Simple Solution to Build and Protect Your Wealth by Wes Gray, Jack Vogel, and David Foulke (GVF), the managing members of Alpha Architect.

GVF took on an ambitious project in that they cover a broad range of subjects from the theoretical ideas of models versus experts, proper investment evaluation, and behavioral finance to the practical applications of value and momentum investing, market timing, and asset allocation.

Part 1 of the book is called “Why You Can Beat the Experts.” Beginning with the Preface, GVF says it is best for investors to maintain direct control over their own accounts. They point out the misalignment of incentives and objectives between owners of capital and investment managers. Even though more concentrated portfolios usually perform better over the long run, managers prefer to hold broader portfolios that have less of a chance of deviating from benchmarks in the short run. Outperforming benchmarks gives managers little or no reward, while under performing can get one fired.
In Chapter 1, GVF presents evidence that model-based decision making gives better results than discretionary decision making. Simple is better than complex, yet “experts” often favor complexity so they can charge higher fees.

GVF points out that the decision making process has three components: R&D, implementation, and assessment. Experts may be useful for model R&D and assessment, but are not usually necessary for implementation. Experts’ use of qualitative information and additional data is often of little or no value. Experts can also suffer from the illusion of skill and a failure to recognize randomness.
In Chapter 2, GVF shows with additional studies and meta-studies that models are consistently better than experts. Models are even better than experts using the same models with discretion. Why is this? Models, unlike experts, are not subject to behavioral biases such as overconfidence, anchoring, framing, and loss aversion.

In Chapter 3, GVF goes into more detail about behavioral biases and shows just how strong they can be.  Chapter 4 explains how experts tell stories in order to make some approaches, such as buy-and-hold or value investing, seem like the only ones that make sense. Investors buy into these stories because they want to feel that expert opinions and judgements matter – when, in fact, they often do not.  In the course of their discussion, GVF debunks a few other investment myths, such as quality enhances value and economic growth leads to higher stock returns.

After GVF establishes the fact that we can beat the experts, Part 2 of the book moves on to explain “How You Can Beat the Experts.” Chapter 5 shows that it is difficult to identify competent and trustworthy advisors. GVF then lays out their sensible FACTS framework for investment evaluation. FACTS stand for fees, access, complexity, taxes, and search. You want fees, complexity, and taxes to be low, while accessibility and search ability should be high. Some of this may be obvious, but not everyone may realize that complexity isn’t usually desirable.

 In Chapter 6, GVF shows that Markowitz mean-variance portfolio optimization (MVO) does not hold up well in the real world due to the instability of its inputs. Research shows that equal weighting of assets does a much better job than MVO or other optimization approaches. The remainder of chapter 6 is devoted to old paradigm asset allocation schemes, such as the Swensen, Bernstein, and Faber IVY 5 strategies. These aim at reducing portfolio volatility by diversifying with permanent allocations to different asset classes. Since equities have consistently provided the highest long-run returns, permanent multi-asset allocation schemes sacrifice some potential return in pursuit of volatility reduction.[1]  Someone who fully appreciates momentum investing might prefer instead to hold only the top performing asset classes instead of them all.

Recognizing the value of trend-following timing overlays, in chapter 7 GFV lays out their own risk management framework that they apply to the IVY 5 portfolio. Because all five asset classes are permanently represented in the IVY portfolio, a trend-following exit from one asset class puts that portion of the portfolio in Treasury bills, which can create a drag on portfolio performance.

GVF once used a 12-month moving average (MA) as a trend following filter. They show here that time series momentum (TMOM) beats out MA as a timing model in 4 out of 5 asset classes.[2]  Instead of just adopting TMOM themselves, GVF now combines 12-month absolute momentum (TMOM) with the12-month MA on a 50/50 basis to create a complicated approach that they call robust asset allocation (ROBUST). GVF says “…the evidence suggests that combining the two technical rules seems to be the strongest performer,” but results indicate TMOM and ROBUST are at least equal in their performance and that TMOM may actually be better. Table 7.8 shows the winner (marked by an X) between MA, TMOM, and ROBUST, according to GVF.

GVF declares a tie in two cases where there is only a 1 or 2 percentage difference in Sharpe and Sortino ratios, but they say that ROBUST is the winner in two other cases (SPX and REIT) where there is the same small difference in these ratios. The only case among the 5 assets where there is a clear cut winner is bonds (LTR), where TMOM comes out ahead.

GVF presents longer out-of-sample results for the U.S., German, and Japanese stock markets. The only clear winner there is also TMOM, when it is applied to the Japanese market. So both in-sample and out-of-sample, TMOM shows equal or better performance to the ROBUST method before transaction costs. After transaction costs, the case is stronger for TMOM over ROBUST.[3]  It is also simpler than ROBUST.
GVF finishes up chapter 7 by giving a good reason, in addition to the usual behavioral finance ones, of why trend following works. This has to do with what GVF calls dynamic risk aversion. As prices drop, investors become more risk averse and do not want to step up to support prices. This accentuates trends as markets extend beyond their fair value. GVF does not mention it, but the same logic could apply to the upside. Investors may become less risk averse as markets rise, causing trends to overextend on the upside as well.

In Chapter 8, GVF discusses security selection using value and momentum. Their Alpha Architect website maintains an updated list of the top 100 value and momentum stocks based on the simple screening criteria that GFV presents here.

GVF’s value selection criterion is earnings before interest and taxes (EBIT) divided by total enterprise value (TEV). This is covered in greater detail in the earlier book Quantitative Value by Gray and Carlisle. GVF shows that the top decile of value stocks from 1927 through 2014 outperformed the market with respect to returns, but that value had much higher volatility, making the Sharpe ratios of value and the market the same. Both value and market also had high worst drawdowns of -91.7% and -84.6%, respectively.

A risk factor that GVF did not mention is the idea of a value trap. Some of the cheapest value stocks may be depressed because they deserve to be, based on very poor fundamentals. These stocks may remain permanently depressed or become even more depressed because they are on the verge of bankruptcy. For this reason, it may be better to invest in more than just the very cheapest value stocks.

Moving on to momentum, GVF gives behavioral explanations of why momentum works. They show that the spread between high and low momentum stocks is close to five times the spread between value and growth stocks (18.4% versus 3.7%) from 1927 through 2014. The top decile of stocks sorted by 12-month (skipping the last month) momentum and rebalanced monthly outperformed the market by 5.4% annually, after deducting 2.4% for annual transaction costs.

There is some controversy, however, about how high transaction costs might really be when momentum is applied to individual stocks. Gerstein Fisher (2015) estimates that a monthly rebalanced, long-only momentum portfolio can have an annual turnover of around 300%. Lesmond et al. (2004), whom GVF cites, report that bid/ask spreads are much higher for momentum stocks. They find strong evidence that trading costs for momentum stocks are at least 1.5% per trade using conservative assumptions and a battery of trading cost estimates. If you multiply the 300% turnover by at least 1.5% per trade, you get annual trading costs of at least 4.5% annually, rather than 2.4%. Perhaps transaction costs now are less than 4.5%. But they should still be substantial, since it is unlikely that the high bid/ask spreads of momentum stocks has changed all that much.

In Chapter 9, GVF gets to the heart of DIY investing by suggesting ways for investors to implement a DIY approach other than by buying and holding a balanced stock/bond portfolio. The first way is with a basic IVY 5 portfolio of generic ETFs filtered with a trend-following 12-month moving average. The five asset classes used are U.S. and international stocks, intermediate bonds, REITs, and commodities. GVF says, “Yahoo Finance charting allows you to run a monthly 12-month MA test for each asset class and with a yearly rebalance across assets you would be in DIY heaven.”  Unfortunately, this is not really possible with Yahoo finance, since it does not provide for monthly charts. Even if it did, Yahoo finance uses only price changes and not total returns. You could, however, use TMOM instead of an MA approach by inputting the appropriate ETFs into and selecting a one year SharpChart that you could bookmark and look at monthly.[4]

GVF next presents enhancements to this strategy by applying both an MA and a TMOM timing filter (which equals their ROBUST method) and by substituting individual U.S. and international value and momentum stocks in place of generic U.S. and international stock ETFs. GVF has a risk-conscious allocation scheme that varies the percentages allocated to equities for balanced, moderate, and aggressive investors to 40%, 60%, or 80%, respectively. However, since timing filters are meant to remove equities entirely from one’s portfolio during bear markets, the balanced and even the moderate allocations may be too conservative for most investors. Equities have provided the highest returns historically, and substantial allocations elsewhere may greatly diminish one’s accumulated wealth over a lifetime of investing.

It is not easy for most DIY investors to calculate TMOM and MA values. So to remedy this situation, GVF provides a free service on their website that applies the ROBUST filter to each asset class for balanced, moderate, and aggressive portfolios. Therefore, no one needs to pay for an asset allocation or timing overlay. They are available free, along with the top 100 value and momentum stocks, to anyone who signs up on the Alpha Architect website.

GVF recommends concentrated portfolios in order to maximize the benefits of value and momentum investing. Modern portfolio principles tell us that a well-diversified portfolio of at least 30 stocks can eliminate most idiosyncratic risk. But this may not be true for portfolios of stocks having high bankruptcy risk. Concentrated portfolios of 30-50 of the very cheapest value stocks may still be too risky.

The same issue applies to momentum stocks. Looking at the top momentum stocks according to the Alpha Architect screener, 18 of the top 25 and 31 of the top 50 stocks are in biotech/medical technology. High intra-industry covariance means that you would need to hold significantly more than the top 50 momentum stocks in order to have a well-diversified portfolio.

Another drawback to an individual stock momentum portfolio with only the strongest momentum stocks is that there is a good chance some of them may be takeover candidates or may have already been taken over. Takeover stocks usually do not perform as well in the future as stocks that showed less momentum but were subject to takeovers.

For all the above reasons, investors may want to hold at least 100 momentum and 100 value stocks if they are selected using simple screens such as the ones presented here. How likely is it that a DIY investor would be willing to manage 100 value and 100 momentum stocks, or even 50 of each?
Periodic timing model signals would add to that complexity, since you would, at times, need to exit and re-enter your entire stock portfolio. If, on the other hand, you do not use a timing model overlay, then you should be prepared for some very large drawdowns and multi-year periods of significant benchmark under performance.[5]

Another potential problem with holding individual stocks is the issue of scalability. The identity of the top value and momentum stocks is readily available now through the use of screeners, such the one that Alpha Architect provides. Increasing amounts of capital buying the same limited number of stocks may mean diminished future returns.
GVF finishes up chapter 9 with their “Ultimate DIY Solution.”  Here, instead of a generic REIT allocation, they recommend a monthly rebalanced portfolio of the top one-third REITs ranked by momentum (excluding those below the 40th percentile in market capitalization).  In place of a generic allocation to commodities, GVF recommends a portfolio of commodity contracts selected on the basis of momentum and term structure.

Recognizing that DIY investors will never go about constructing all these different portfolios, GVF suggests that DIY investors hold all seven assets (U.S. value and momentum stocks, international value and momentum stocks, bonds, REITs, and commodities) in the form of exchange-traded products (ETFs or ETNs) and then apply their risk-management overlay to these funds each month. Here is a table of suggested investment vehicles:

There are currently no domestic or international momentum ETFs that hold fewer than 50 stocks, although some may be forthcoming. There are also no ETFs that hold fewer than 50 REITs or that use momentum to select REITs.

There is a just one commodity ETF or ETN with a focus on term-structure and momentum. It is the United States Commodity Index (USCI). However, USCI is required to maintain positions in six different commodity sectors at all times. This means some sectors may not always have positions that are in accordance with a favorable term-structure. USCI does use 12-month momentum to help select its long-only positions. According to Miffre and Rallis (2006), 12-month momentum applied to commodities is profitable, but mostly on the short side. However, Geczy and Samonov (2015) report that 12-month momentum (skipping two months) works in reverse for commodities. In other words, positive 12-month momentum is a negative factor. USCI is down 43% since its high in April 2011.

In their final chapter, GVF mentions some of the reasons investors might not want to do-it-themselves. These include being able to blame someone else if things don’t go well, inertia to change, unwillingness to let go of current relationships, bias in favor of experts over models, overconfidence in one’s own abilities, models not conforming to our preferences and beliefs, and the desire to be a hero. These psychological excuses are certainly worth thinking about. Perhaps then they will be less likely to influence us against adopting a sensible, model-based investment approach.

If I had been GVF, I might have said more about why investors who adopt DIY investing might abandon or modify it into something that is no longer purely model-based.  DIY investing should be simple, but it is not always easy. To be successful, DIY investing requires a good understanding of the principles underlying one’s model and the requisite discipline to stick with it under varying market conditions.
Here are my main conclusions about the second half of the book:

•    The IVY 5 permanent portfolio scheme does not take advantage of relative strength momentum.
•    The ROBUST timing model trades more, is more complicated, and is no better than TMOM.
•    Value trap risk and intra-industry co-variance risk may make value and momentum portfolios with 50 or fewer stocks too volatile, while larger value and momentum portfolios may be impractical for DIY investors.
•    Both value and momentum are subject to high worst drawdowns. The use of a timing model to reduce drawdown may be difficult with large portfolios of individual stocks.
•    Turnover is very high and transaction costs may offset all or most momentum profits for portfolios of individual stocks.    
•    The most practical approach is to apply a timing overlay model to asset class ETFs.  Investors can easily do this with TMOM and SharpCharts. Those who prefer to use ROBUST can access it for free on the Alpha Architect website.
The reason I can recommend this book is the good job GVF does in explaining models versus experts and behavioral biases. We can never be reminded too often of these important matters.

[1] A newer paradigm uses trend following methods to diversify among different assets on a temporal rather than a permanent basis.  This means investors can focus on equities for as long as equities remain strong and diversify into other assets when they are strongest. 
[2] GVF uses the term time-series momentum (TMOM) to mean absolute momentum. Both relative and absolute momentum are based on time series (asset returns), and absolute momentum is a better term to differentiate it from relative momentum. However, we will use time-series momentum (TMOM) here to avoid confusion.
[3]GVF estimates that the MA signal has a 20% higher turnover than TMOM, while our calculations show 30% more trades. MA and ROBUST should therefore have greater transaction costs and more whipsaws than TMOM.

[4] Alternatively, you could track a portfolio on ETF Screen or Morningstar that shows performance over the last year. 
[5] Timing models have a much more beneficial effect on momentum portfolios than on value portfolios.