Replicating Meb Faber's Relative Strength Strategies for Investing


While trying to decide if I was going to remain on P123 or leave (see What now? 12345), I decided to try to replicate Mebane Faber’s results in his research paper titled “Relative Strength Strategies for Investing” (SSRN-id 1585517.pdf. See White Papers - Meb Faber Research - Stock Market and Investing Blog ). In this paper, he showed how using momentum (relative strength) measurements and sector rotation can result in better performance with lower drawdown, compared to the S&P 500 as a whole. (It was in preparing for this test that I posted I wanted to see if the results Mr. Faber got up to 2010 still applied in the years 2011-2022.

When I ran the results, I was unable to show ANY outperformance over the S&P 500, and was able at best to match the S&P 500 in performance, even for years prior to 2010. (See " Cary_Sector_Relative_Performance" and “Cary_Meb_Faber_Sector_Rotation” for the ranking and trading systems.)

Has it been the experiences of the folks here on the forum who have tried replicating published strategies like these to fail to perform as well as the published results?


Mebanes own fund GMOM has underperformed even a conservative benchmark since inception.

I see that someone has tried to make the case using that paper and some updated ones for a TAA model going back to 1970. Extrategic Dashboard - Faber Tactical Asset Allocation w/ Relative Strength (Updated)

But they bench it against the S&P 500 and not a 60/40 portfolio or something more relevant.

I think the core idea is pretty basic.

I think it’s safe to say that replicating published studies will almost never perform as well as the published results. Sometimes they will outperform the benchmark but not as strongly. Sometimes they won’t outperform at all. One might think that the easier the study is to replicate, the less likely it is to outperform, due to the nature of investor arbitrage. However, factors that are rooted in human behavior or mathematical laws are much more resistant to being arbitraged. So, for example, there are many studies that show how value and/or momentum strategies have worked in the past, and because those factors are so deeply rooted in investor behavior, I wouldn’t be surprised if they kept working for a long time even if they’re relatively easy to replicate.

You might want to read the following blog posts, which discuss this in different ways:

Thanks to both of you for your comments.

Thank you, Mr. @hemmerling, for the links. The web site there didn’t have the S&P 500 sector rotation strategy that I studied, but it does have other strategies of Faber’s that I had considered studying, such as GTAA13. The performance graphs for those strategies did not appear to have any drop-off in performance in the later years.

Mr. @yuvaltaylor, thank you so much for those links. I have read everyone of those posts you provided and wondered about if the diminishment of returns were due to the market starting to anticipate those selections, akin to price-to-sales or price-to-book having lost their edge, or if the factors they devised were statistical artifacts (curve fitting). I will definitely go back and reread those posts (and all of the other posts on the blog).

What I tried to express with my original post was that what I expected to see was perhaps a diminishment of returns over the years since 2010, the date of Faber’s paper. Instead what I found was under-performance in ALL the years I started, including 2002-2009, years covered by Faber’s paper. I figured I had done the ranking or trading wrong, but was wondering if other had similar experiences.

So, thank you both for your responses. I will study those links again.


I completely agree, over the last 10 years, only a few asset allocation strategies have outperformed the SPY 500. Partly, this is due to 15 incredible years in the stock market, and bonds have not acted as a “hedge” as they historically have. Many asset allocation strategies rely on the negative correlation between stocks and bonds to work.

By the way, you can find some good overviews here:

I think that if you use the theory the period of relative strength needs to be shortened to 3 months now. People or computers are just faster to respond now. Plus, this only works if you are slight quicker than everyone else, right? It will probably continue to speed up until it no longer works at all. That would be the game theory prediction anyway.

Also each asset has to perform at about the same level or you need to use some leverage to equalize the expected returns.

You should consider modern ideas on how to weight the assets. Maybe minimum-variance. I would recommend against using historical returns to maximize the Sharpe Ratio but you could shrink the returns using Bayesian statistics and try to maximize the Sharpe Ratio with those Bayesian numbers.

Bayesian statistics can be a lot of hype. But Yuval keeps linking to this paper (see Zoo of returns above). For good reason I think (although this paper hypes a lot too): Is There a Replication Crisis in Finance?

I think you will find that you get better risk adjusted returns if not much improvement in the absolute returns.


1 Like

Thanks to you both for your comments and observations.

Thanks, Mr. @Whycliffes, for the articles as found on AllocateSmartly. I first learned of AllocateSmartly and Portfolio123 on the financial blog of Mr. @paulnovell located at I really enjoyed reading his blog which I discovered in 2018, as he started out in 2011 using AAII’s Stock Investor Pro, and then progressed in his learning, eventually writing about Tactical Asset Allocation (TAA) and Portfolio 123. Based on his writing I joined on 2018 Feb 24 and signed up to follow 2 different TAA strategies: Adaptive Asset Allocation (AAA) and Vigilant Asset Allocation (VAA). I followed both of them regularly for 1 year. Both performed badly. The next year I dropped the AAA and devoted all of my funds to following VAA, this time using semi-monthly tranches to reduce timing luck. I did terribly the second year with that strategy. Finally, after 2 1/2 years, I cancelled my membership, since the performances weren’t anything like they had been prior to my joining.

Recently, as I was considering leaving P123 because of my lack of progress in building a decently performing ranking and trading system and thinking of alternatives, I considered going back to, thinking that perhaps I was not patient enough with them. But when I visited the site last month, I noticed that their example TAAs with performance figures all ended at the end of 2016 (see What We Do - Allocate Smartly). While that may have been fine back when I first signed up in early 2018, I don’t think it is fine in 2023. It made me suspicious that TAA performances have dropped.

Then I decided that P123 had the tools needed to be able to determine how the different TAAs performed. While I wasn’t certain that I could locate the rules for AAA or VAA, I knew that I had Meb Faber’s papers for different TAAs, along with their associated rules, so I decided to try to implement the market sector rotation TAA. When I was unable to reproduce the results as presented in the paper, then I decided to create this thread and get your opinions.

Mr. @Jrinne, sir! Thank you for your reply. I have a feeling that I need to go study probability and statistics in-depth, including Bayesian math. I had a course that addressed those topics, but that was almost 50 years ago and I haven’t had much need to use that knowledge since, having gotten my degree in computer science. But I can certainly see it being useful in financial analysis.

Thanks again to all.


Yes, I agree. I use my asset class strategy more as a hedge against my stock portfolio. However, it is true that since 2016, my strategy, which consists of three different, uncorrelated strategies from allocatesmartly, has not outperformed the S&P 500, but I am only slightly below it and have much less total drawdown and volatility, which is a point in itself for me. At the same time, the comparison is a bit unfair because the S&P 500 has performed exceptionally well since 2010.

I believe that the returns we have seen in the stock market from 2010-2022 will not be the case for the next 10 years, so allocation strategies like these may perform better and be better suited going forward than during the period where the market has performed incredibly well.

1 Like


I do not want to leave you—or anyone—with the wrong impression. I love this machine learning and statistics stuff. And I use it. I come at a problem from multiple different directions and I guess some of those directions can seem a little esoteric. (I guess, but it is either mathematically correct or not: one simple logical step at a time and I miss any apparent complexity sometimes).

At the end of the day, the algorithms I actually use are VERY simple by any definition. To be clear Python can make some difficult things simple. But for the most part Python is making it simple for me in that—back in the day—I used to do this with pencil and paper which would probably be impossible with all of the data P123 provides. BTW, everybody who has ever used Python is at least quicker with Python than I am: I am not a programmer. But that is okay because the ideas I use with my money are very simple: Consistent with Occam’s Razor. Okay that I do not program well because what I do does not require a lot of “data wrangling” which I find tedious at best.

Don’t make it too complex whatever you use, and don’t let me make you think that you need to. You have the math skills already—especially considering you are sandbagging us a little in the forum, I believe :slight_smile:. Your post reflect an excellent working knowledge of everything you need, IMHO. Not that your way will end up being anything like mine.


Thank you both!

A few years ago, I wrote about the three main problems with Tactical Asset Allocation (I learned these from this article, which is now behind a paywall: Tactical asset allocation: When diversification is a better bet than timing | Special Report | IPE):

First, assessing where we are in the business cycle in real time can be challenging, to say the least, mainly because economic data is always lagging real-time developments and because different sets of economic data usually contradict each other or are inconclusive when looked at as a whole.

Second, markets move much more rapidly than economic data. As Alfred Kassam, the head of equity applied research at MSCI, observes, “It is very hard to predict fast-moving financial markets using slow-moving economic data.”

Third, the purpose of investing in alternative indexes is to beat the market over the long run, and switching in and out of them defeats that purpose altogether. If you’re buying value stocks because of the value premium, you deny yourself that premium if you switch in and out of that strategy.

For the entire discussion, see Market Timing, Tactical Asset Allocation, and Trading: A Dialogue - invest(igations)

Thank you, Mr, @yuvaltaylor, sir! Earlier I read your conversation between Moses and Celeste (as I have read every post on the blog) and very much loved it. I loved that you presented the good arguments of tactical asset allocation and market timing on the one hand versus buy-and-hope on the other hand. I know that I was a believer of buy-and-hope until the Great Recession of 2007-2009, and then became convinced that there has to be a better way. I have been trying to learn about those better ways ever since.

I am now going to re-read your blog post about the factor zoo and the replication crisis.

Thank you again, and thanks to all.


Reproducing the results of a study can be a complex process because there are many factors that can affect the results, including changes in the market and changes in the data used. In addition, there is a risk of over-learning when the model fits the past data too well and does not work as well on the new data. If you’re having trouble replicating the results of Mebein Faber’s study, it may be helpful to contact the author of the study or seek help from other experts in the investment field. They can help make sense of the problems you are facing and suggest possible solutions. It is also worth keeping in mind that investment strategies are not a guarantee of success and always involve the risk of losing capital. Therefore, it is important to approach investing prudently and with your personal goals and risk profiles in mind.

Hi, @FrancesWalsh, sir (ma;am?) ! Welcome to the forum! You are going to meet some really smart investing/trading gurus/geniuses here on this forum. You are also going to meet wannabes like me, for whom the Dunning Kruger effect is evident.

Thank you for your comments. It could be that I have misinterpreted the paper (SSRN 1585517, found here), or that the source of data gives different results than the data available to me in Portfolio 123. Faber used the data contained in the French Fama CRSP Library (see here). But the ideas in the paper were to use the 10 sectors in the US economy (now 11*) and rank each sector by relative strength (momentum) using the 1 month, 3 month, 6 month, 9 month, 12 month, and combination look back periods and see how the different sector rotations performed using the top 1, 2, 3, … 9, as well as equal weight. The performance of each portfolio over the period 1928-2009 is shown in exhibit 4.6, p. 9:

Mr. Faber then showed in Exhibit 6, p. 11, how using the top 1, 2, or 3 sectors performed versus B&H for the different decades:

As is seen in the table, the strategy outperformed every decade, more in some decades and less in others. Mr. Faber then goes on to describe how the returns can be improved by being fully invested when the market is going up (as indicated by the 10-month SMA of the S&P 500) and being in cash when the market is going down. The returns of the portfolios improved and the drawdowns lessened, as shown in exhibit 7, p. 12:

I then attempted to replicate the results. I was hoping that I could reproduce the results of the decade of the 2000s as reported in exhibit 6, and also was interested in seeing if there was a drop-off in performance of the strategy after the paper was published. I then ran the test for the period of 2002 Jan 01 to 2022 Jul 02. I used a 4 week rebalance period and the 200 day SMA of the S&P 500, since using monthly data is difficult in P123, as well as an ETF universe of 11 sector ETFs.

Imagine my surprise when I could not reproduce the results of his study for 2002-2009, with my version underperforming the S&P 500. So that is why I posted this thread, hoping to find out if (1) my test was invalid or otherwise incorrectly designed, (2) the use of a 4 week rebalancing period is sufficiently different from a monthly period that the comparison isn’t valid, (3) my ranking process of combining the 6 month (130 day), 9 month (195 day), and 12 month (251 day) relative strengths into a single rank was not valid.

So, Mr @FrancesWalsh, sir, I appreciate your comments and still ponder if my test was invalid or if the performance as discussed in Meb Faber’s paper cannot be duplicated in P123 even with a perfectly valid ranking ad trading system.


(^) now 11 sectors as shown in’s heat map from Friday, 2023 Mar 17 close:

I tried to see if anyone on had coded exactly what you were asking for.

I didn’t find anyone, but there are some other systems on Smartly that you can get for free on Additionally, you’ll get much more information about the system itself and an overview of the entire transaction history.

Here is Kellers, for example: Bold Asset Allocation (BAA) - Keller by Tristan F -