Ivy Porfolio de Mebane Faber : allocation d'actifs tactique quantitative

Bonjour,
J’ai lu une étude qui s’intéresse à l’utilisation de la moyenne mobile sur 10 mois des actions comme signal de vente/d’achat (comme expliqué dans le 1er message de ce sujet).

J’ai extrait quelques passages mais je vous conseille si cela vous intéresse d’aller consulter directement l’étude qui est forcément plus exhaustive :
The Real-Life Performance of Market Timing with Moving Average and Time-Series Momentum Rules - 2013

Valeriy Zakamulin a écrit :

« In this paper we argue that the reported performance of market timing strategies usually contains a substantial data-mining bias. In addition, in a study of a technical trading rule one usually ignores important market frictions such as, for example, trading costs.
The data-mining fallacy that has been repeated over and over again in many studies consists in the following. Using a full historical data sample one tests many k-month moving average trading rules and picks up a rule that performs best. One then reports the performance of the best trading rule in a back test and either explicitly or implicitly assumes that the expected future performance of this rule will be the same as the past performance. Yet, as a matter of fact, one should expect a much poorer performance in the future than the reported performance in the past. »

« As in Faber (2007) and many other papers, when we examine the performance of the best trading rule in a back test, we do not take into account the transaction costs in order to show the best possible performance. The results are reported in Table 2. Observe that the SMA(10) rule, advocated by Faber and many others, is the best trading rule when the passive benchmark is the Standard and Poor’s Composite index »

« The main goal of this subsection is to demonstrate that the out-of-sample performance of a trading rule is sensitive to the choice of a split point between the initial in-sample and outof- sample subsets. What is more crucial is that depending on the choice of a split point the out-of-sample performance of a trading rule might by either superior or inferior as compared to that of the passive counterpart. »

« Before turning to the discussion of performance dependence on the split point, we would like to draw the reader’s attention to the fact that in every historical period the best trading rule in a back test outperforms the passive benchmark with a solid margin. This demonstrates that, when many trading rules are back tested, one can virtually always find a rule that performs better than the passive benchmark. 

Consider in details the performance of the market timing strategies versus the performance of the passive strategy when the split point between the initial in-sample and out-of-sample subsets is chosen to be December 1929. In this case the initial in-sample period is January 1926 to December 1929, whereas the initial out-of-sample period is January 1930 to December 2012.
With this choice, the Sharpe ratio of either of the best trading rules in a back test amounts to 0.16 which is about 50% higher than the Sharpe ratio of the buy-and-hold strategy. In the out-of-sample tests, the Sharpe ratio of both the trading rules amounts to 0.12 which is about 10% higher than the Sharpe ratio of the buy-and-hold strategy. Even though the reallife performance11 of the trading rules turns out to be much poorer than the performance of the best trading rules in a back test, still, judging by the Sharpe ratio criterion, we have to conclude that either of the market timing strategies delivers a better performance than that of the buy-and-hold strategy.

Note, however, that if the split point is chosen to be December 1931 (which seems to be a relatively negligible relocation of the original split point) then we arrive at the opposite conclusion: the real-life performance of the market timing is either worse than or equal to the performance of the buy-and-hold strategy. »

« Our analysis reveals that the superior performance of market timing strategies is confined to relatively short historical episodes. In particular, over our total sample 1926 to 2012 there have been only four relatively short historical periods that contributed most to the superior performance of the market timing strategies. Specifically, they are the periods of severe bear markets of 1930-31, 1973-74, 2001-02, and 2007-08. As a result, the performance of market timing strategies, relative to the performance of the passive benchmark, depends largely on whether the split point is located before or after a major stock market downturn. »

« The proponents of the market timing with moving averages and momentum rules often advocate that such a strategy allows investors both to reduce risk and enhance returns. We have to acknowledge that there is indeed a small chance of this happening. For instance, if the S&P Composite index is used as the passive benchmark, there have been 4 relatively short periods of severe market downturns when the real-life market timing strategy provided higher returns with lower risk than the buy-and-hold strategy. It happened that 2 out of 4 such periods occurred in the decade of 2000s.
As a result, this decade was an incredibly successful decade for the market timing strategies. Yet this is not a typical performance of the real-life market timing strategy. Our findings reveal that over a long run the market timing strategy is indeed less risky, but the reduction of risk always comes at the expense of reduction of returns. »

« Recall that every market timing strategy prescribes moving to cash when a Sell signal is generated. The investor, who employs the market timing strategy, hopes that during a Sell period the Treasury bills provide better return than the stocks. We define by “failure” an event in which the Treasury bills fail to deliver a higher return than the stocks during a Sell period. So how often a real-life stock market timing strategy generates false signals?
Our analysis reveals that the average failure rate amounts to about 80%. In words, this means that when the market timing strategy generates a Sell signal, in approximately 80% of cases at the end of a Sell period the market timing strategy will provide a lower return than the buyand- hold strategy.12 Again, this result demonstrates that, when the passive benchmark is a stock price index, the superior performance of the market timing strategy is confined to some relatively short particular episodes.

That means that in order the market timing strategy delivers a superior performance as compared to that of the passive counterpart, investors sometimes have to wait a very long time and experience painful emotions because their active portfolios consistently lag the benchmark. For example, the real-life performance of market timing was inferior during a 25-year period from 1975 to 1999 regardless of the choice of a passive stock market index. »

Dernière modification par NicolasV (20/04/2015 13h03)