My only problem with this logic is the arbitrary choice of gold as a "diversification" tool. Why not find other assets that may not be as liquid, but offer better real returns?
I used gold because poster said "uncorrelated instrument with dubious -ve real return". Gold fits that category. What other asset would you use for higher inflation hedge. Now days one can use TIPS, but does not have longer history. Commodities falls under the same category. Only other asset class which has +ve inflation beta is US farmland. Atleast from limited data, GLD is better diversifier than TIPS. It has less market correlation and higher SD. So small allocation of GLD is much better than TIPS http://bit.ly/2z4ywAw
That assumes the conclusion. If there were magical free S&P puts available (ie, a tail hedge with non-negative Sharpe), every investor would sign up.
Fair point. I admit to choosing numbers that made the result starker. The problem is that a classic Markowitz framework treats this kind of strategy very badly, so if you like my number massaging was justified. GAT
I agree that there are challenges regarding the data... I recall one of the asset allocation papers arguing that property is better than gold, but I can't be certain that my memory isn't failing me. My point, broadly, is that there's an obvious trade-off that one needs to make when judging the merits of a given diversification instrument. Given that the whole purpose of diversification is to enable you to hold on to a portfolio long-term, under a whole variety of scenarios, I find the "liquidity fetish" a bit puzzling. IMHO, if you make an effort and relax your liquidity constraints somewhat, you should be able to do better than gold. I believe that's something that Buffett has actually done quite well.
Yes. Once investors decide asset allocation, they should hold for long periods other than rebalancing. Gold and Real estate both holds store of value and slightly illiquid. But that's where comparison stops. Both behaves differently to stocks and bonds and both has places in portfolio.
This is a beauty! http://www.managedfutures.com/program_performance.aspx?fundtype=mf&productId=48059 Cantab's SR is 0.28 according to the above. Where do you get 0.5? And 10 years, in your view, is not a "short period of time"? A technical analyst, fundamental analyst, and a quant try to predict the stock market. The TA says "Our rate of change indicators among others should help us get the job done". The FA replies: "That's all wrong don't you know all the technicals are in the price?. Let's analyze this Fed data over here". The quant, insulted by such uneducated and non-rigorous thinking pipes up: "What nonsense! First, let's assume that this market is uncorrelated with this other market...."
No, 10 years isn't sufficient to say very much. I'll give you some more maths, since you love it so much and I'm some kind of masochist who enjoys trying to educate someone who is basically putting their hands over their ears and saying "la la la I can't hear you". The variance of SR estimates can be shown to be (1+ 0.5SR^2)/T*. Assume the SR estimate is 0.28** (by the way the data shown finishes about a year ago but let's use that number) with 10 years of returns. The variance is 0.104. The standard deviation is the square root of that is 0.322. So we can be 95% confident that the true historical SR was in the range 0.28 - (0.322*2) to 0.28 + (0.322*2) i.e. in the range -0.36 to 0.92. If we take Winton for example with roughly 20 years of data and a SR of about 0.6 the range is 0.11 to 1.09. Twenty years of data is barely enough to show that there is statistical evidence that Wintons returns are positive. However they could plausibly be just above zero or over one. Ten years of data tells us basically nothing. So actually it's a moot point whether the estimated SR of Cantab was 0.28, 0.5 (in which case the range would be -0.17 to 1.17) or something else. Seriously having a better intuitive understanding of statistics will make you a much better trader, even if you don't want to start wearing glasses, developing poor hygiene and calling yourself a "data scientist" or "machine learning guru". GAT * this relies on certain assumptions but for CTA type strategies*** with typically negative return autocorrelation this probably overstates the variance, i.e. we are being conservative. See Lo, 2002 http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.460.3450&rep=rep1&type=pdf ** Not that it matters but I hand calculated the SR for Cantab in my head and obviously got it wrong, for the time period shown it is 0.28. However there is about a year of missing data so I don't actually know what the latest estimate for Cantabs SR is. *** Lo doesn't consider the effect of skew and I need to do the maths on this myself since I can't find anyone else who has looked at it, but my intuition is that large positive or negative skew makes the variance of the SR wider. EDIT: I've just found this paper https://cran.r-project.org/web/packages/SharpeR/vignettes/SharpeRatio.pdf and page 13 is relevant
The linked page from your post shows: Annualized Return 7.50% Annualized Std. Deviation 17.92% Which, using rf== 0%, would be 0.42 Sharpe, close to GAT's estimate. (The linked page uses rf== 2½%, which has a big downward impact on small Sharpes)