One system, two portfolios, 24000 parameter settings
Posted: Wed Oct 20, 2010 10:25 pm
For the past 5 years or so, I've consistently noticed a trend in my research results: diversifying the portfolio of traded futures markets, beyond the boundaries of the United States, has been very profitable. Both in research (backtesting) and in real life, real money trading.
Here is one small example. I've chosen to define nine "sectors" or "groups" of futures markets, and selected a US-only portfolio by choosing five markets from each sector. (However, some sectors contain fewer than five US markets, so I just took all available US markets in that sector. This happened in meats, STIRs, LTIRs, and Energies.)
There are 38 total markets in this US-only portfolio.
I chose another portfolio of 38 markets, but this time included non-USA markets in my choices. The second portfolio had the same number of markets in each sector as the first portfolio, as you can see:
I ran the two portfolios on the same system, the Blox presupplied "Triple Moving Average" system. I swept the parameters of this system over a wide range, shown in Figure 0. This is a total of 16 x 10 x 15 x 10 = 24,000 different parameter settings. Results are shown in Figure 1 and Figure 2. These are "Sharpe Efficient Frontier" plots as invented (HERE).
Each of the 24,000 simulated parameter variations is plotted as a dot on this scatterplot. The X-coordinate shows the average trade duration for that particular parameter set, and the Y-coordinate shows the Robust Sharpe Ratio.
Figure 3 overlays the Frontiers (the upper boundaries) of the USA-38 portfolio results, and the INTERNATIONAL-38 portfolio results, on the same graph. As you can see, diversifying from USA to INTERNATIONAL has increased profitability, at least in this particular case of this one system and these two portfolios. However, I have seen confirmations of this general trend, again and again: the more I diversify globally, the better the results. This is merely one example.
Please feel free to perform similar experiments yourself, possibly using different systems, different portfolio construction rules, different portfolio sizes, and so forth; whatever interests YOU. Run some tests! Collect some data! Analyze the results and, occasionally, have a Eureka! moment.
EDIT: added another paragraph
EDIT2: fixed spelling mistake
EDIT3: fixed incorrect extra mkt in International list
Here is one small example. I've chosen to define nine "sectors" or "groups" of futures markets, and selected a US-only portfolio by choosing five markets from each sector. (However, some sectors contain fewer than five US markets, so I just took all available US markets in that sector. This happened in meats, STIRs, LTIRs, and Energies.)
Code: Select all
PORTFOLIO: USA-38
Grains Corn, Oats, Rough Rice, Soybeans, Wheat
Metals Gold, Copper, Palladium, Platinum, Silver
Meats Milk, Feeder Cattle, Live Cattle, Hogs
Softs/Tropicals Cocoa, Cotton, Coffee, OJuice, Sugar
Stock Indices eMini S&P, eMini NASDAQ, miniDow, miniMidcap
Short Term Interest Rates Eurodollars, FedFunds
Long Term Interest Rates 5Year, 2Year, 10Year, Bonds
Energies Crude, HeatingOil, NatGas, RBOB
Currencies Aussie, British, Canadian, Euro, Yen
I chose another portfolio of 38 markets, but this time included non-USA markets in my choices. The second portfolio had the same number of markets in each sector as the first portfolio, as you can see:
Code: Select all
PORTFOLIO: INTERNATIONAL-38
Grains Corn, ParisRapeseed, RoughRice, Wheat, KC Wheat
Metals Gold, Silver, LMECopper, LMENickel, TokyoPlatinum
Meats Milk, Feeder Cattle, Live Cattle, Hogs
Softs/Tropicals LondonCoffee, LondonSugar, TokyoRubber, Cotton, MalayPalmOil
Stock Indices HangSeng, Singapore, Topix, CAC-40
Short Term Interest Rates FedFunds, EURIBOR
Long Term Interest Rates 5Year, 2Year, Japanese, Swiss
Energies LondonCrude, LondonGasoil, NatGas, RBOB
Currencies Euro/Yen cross, NZDollar, Yen, Euro, DollarIndex
Each of the 24,000 simulated parameter variations is plotted as a dot on this scatterplot. The X-coordinate shows the average trade duration for that particular parameter set, and the Y-coordinate shows the Robust Sharpe Ratio.
Figure 3 overlays the Frontiers (the upper boundaries) of the USA-38 portfolio results, and the INTERNATIONAL-38 portfolio results, on the same graph. As you can see, diversifying from USA to INTERNATIONAL has increased profitability, at least in this particular case of this one system and these two portfolios. However, I have seen confirmations of this general trend, again and again: the more I diversify globally, the better the results. This is merely one example.
Please feel free to perform similar experiments yourself, possibly using different systems, different portfolio construction rules, different portfolio sizes, and so forth; whatever interests YOU. Run some tests! Collect some data! Analyze the results and, occasionally, have a Eureka! moment.
EDIT: added another paragraph
EDIT2: fixed spelling mistake
EDIT3: fixed incorrect extra mkt in International list