A key question that amateurs and professionals must always ask is "do the winners perform better than the losers?". The question is of interest to all who like growth or contrary strategies, who like to back the good or the bad, and who have to choose which stocks to buy or sell from a portfolio. It's also of great interest to my colleagues and me as we are trying to relearn from scratch the sources of performance in individual stocks. The academics have performed numerous studies on this. Most of the famous names have looked at it one way or the other. One of the classic studies is by Griffin, Ji, and Martin 2003, Journal of Finance [33 page pdf]. They conclude that the best performers outperform the worst performers for subsequent periods up to a year in every world market. Before this there was Jegadeesh and Titman 1993 [28 page pdf], and a myriad of other studies.

One has been very wary however, of accepting academic findings throughout one's career, especially in an area like this. The problems are legion. Many of the worst performing stocks are very small. Let's say 100,000 in market value. The price could be $ 0.25. The returns in any period are highly variable. Indeed, the bid asked spread in the old days frequently averaged 50% and commissions and rebalancing could easily add another 50%. Another problem is that most academics don't take the trouble to properly take into account survivor bias in all its terrible manifestations. The most apparent one is that the worst performers that continued to perform badly go bankrupt and are not covered in the files. Furthermore the best performers in any continuous period often while great today were unknown yesterday and wouldn't have hit the files on a contemporaneous basis.

While academics sometimes address the problems of survivorship, bid asked spreads, impossibility of implementation, transactions costs, comovements between securities of different styles in a year so that what looks like 1000 independent observations is really one, selected starting and ending points (it's always easy to find a good time to start and end ), non-contemporaneousnous data (Shiller is the poster boy for this), retrospective multi comparison reporting of good results only kinds of problems, they never consider the problem of ever changing cycles.

Thus, it was great eagerness and pleasure that I learned that Dimson, Marsh and Staunton had made a thorough study of momentum. Their work is always of the highest standards. They get original data from the actual contemporaneous newspapers of the time. They examine many years of data, always bringing the results up to date. They present their work in a form suitable for both the academic and the layman to understand. And they always consider the profitability and commercial viability of their work. Between them, they are fully conversant with all literature in the field, they relate their work to every important academic model that has come down the pike. Furthermore, they can always be counted on to add a few embellishments of their own that raise important questions for further research. Their studies of momentum cover many universes of individual stocks from 1897 to 2010 for several English Markets, with updates for the last 55 years for all world markets. What more could you ask.  And yet … (to be continued)…

The naïve speculator starting from scratch has studied the issue for a universe of stocks that is actually big enough to matter and implement. The stocks are the OEX 100.. all the big companies are in there and there is turnover of less than 5% a year. Here are results. Let's consider the performance of companies in 2008 that were the best and worst performers in 2007. Consider the worst 20 performers in 2007 versus the best performers in 2007 and look at their performance in 2008.

   performance in next year

             of best performing 20              of worst performing 20
    2007         -32                                -25
    2008         +10                                +55
    2009          19                                 10
    2010          07                                 09
    2011          25                                 11

The standard deviations are so high to make all these differences totally random except for 2008. 

Momentum Continued:

The Dimson, Marsh, Staunton trio has outdone even their usual superb work in the extensiveness and depth of their momentum study. They study yearly momentum returns from 3 English markets, including all stocks, and the top 100 from 1900. They report international returns from 18 countries from 1950, to 2000, and then update these returns to 2007. They consider all leads and lags and skips in defining momentum, and then regress these returns on the standard though flawed measures of superior performance from the Fama French studies of 20 years ago. Their conclusion is that "momentum has been consistently profitable over the last 108 years, and is not subsumed by other factors. The momentum premium has been substantial across equities as a whole, but also within size and value based partitions".

Some highlights from the 9 figures and 8 tables that appear in their study.

1. A value weighted cumulative return for the winners from 1956 to 2007 us 5200% versus 30% for the losers. The winners beat the losers in38 years and lost in 13 years. An annual value weighted momentum portfolio of the winners from 1900 to 2007 returned 4.25 million % versus 111% for the losers. Of 17 international markets, in all but the US, the winners beat the losers from 1955 -2000 and 2000 to 2007. The average monthly superiority in return was approximately in both the earlier and later periods. DMS define a momentum strategy in terms of the base period for ranking the best and worst, the skip period for waiting to implement the trade, and the forward period. The returns are relatively invariant for all of the 3 parameters, and all breakdowns of where to define the cutoffs for the best and worst, e.g the top 10%, 20% or 30%. We will pass over the regression results showing the dependence of the return on value factors as there seems to be a statistical lapse in the measures of variance accounted for in their results, and the significance is vastly overstated. 

The main problem with the study is a combination of ever changing cycles and enormous losses that the strategy takes after a bad year for the S&P. In 2008, they report a loss of 75% for the best - worst strategy. Similar catastrophic results occurred in year, 2000, 2003, and 1973. Indeed the 2007 result was so horrid that the winners losers strategy did not work in the 2000-2010 period for us stocks. the results could be improved dramatically by taking account of the dependence of the winner -loser results on the market return during the preceding period. There is also an enormous negative serial correlation between the winner - loser returns. 

The second main problem with the study is that when results are widely disseminated they tend to be dissipated. This is the first part of the problem of ever changing cycles which Bacon and I have found so prevalent in all speculative activities. The public shoots down the odds on good systems, so that they aren't as attractive any more. And then regardless of the attractiveness, they perform worse than they should because the bigs can't makes as much by investing in them. It is no wonder that the momentum strategy has not worked in the last 10 years in the US markets. We would predict it would have similar lapses in the other big markets, just until they become big enough to lure big investments from funds and other slow moving participants searching for opportunities of superior performance by following slow moving, easily implemented systems. This is by no means a criticism of the fine work of the DMS trio. They are to be complimented on implementing one of the most thorough tests of an anomaly in the literature.

Momentum in Conclusion: 

In recent days, I have read a number of mopping up studies of momentum including Barroso,  Asness, (he of the no Fed model),  and Choi, and I have received a final note from the DMS trio. All of these papers are highly flawed in that they enter and close and faulty door. They try to improve on momentum, when it doesn't work in the first place. They try to improve it with better definitions of momentum, and multiple comparisons of momentum with the worthless FAMA french model. It is sad to see such wasted effort, dead weight effort. Splits of a series into seemingly alluring returns when the basic regularity is random. How many of us have wasted our time, knocking on a closed door in markets and life?





Speak your mind


Resources & Links