Suggested IWM/GDX Allocation Levels

A discussion in [URL="http://www.effectivevolume.com/showthread.php?3585-Discretionary-trading"]this thread[/URL] triggered the idea to update [B][I]this[/I][/B] thread weekly with "modeled" or "idealistic" minimum risk allocations for the IWM and GDX robot.

Suggested IWM/GDX allocations for [I]minimum portfolio volatility[/I]:

[B]IWM: 73%
GDX: 27%[/B]

This is a walk-forward test to determine the variability of this allocation. Use at your own peril...

The next update will be the weekend of June 25th.

===================

Correlation Matrix: [ATTACH=CONFIG]8948[/ATTACH]

In attachement a picture of a test I did in Amibroker based on the robot trades. It shows (when not taking into account strong/weak signals) that the optimal allocation (highest return for the least risk) is :

[B]IWM: 68%
GDX: 32%[/B]

Of course like all backtests this is purely theoretical. But it kind of matches Paul's findings based on the ETF data alone. And it also matches Pascal's tests that show the combination of a double leverage Russel2K ETF combined with GDX works best.

If we'd be making a cocktail the recipe would be ... one part GDX, two parts IWM and a dash of lime.

Rembert,

Could you tell us what the maximum DD is intra-trade, not closed trade with this combination?

Adam

[QUOTE=adam ali;14225]Rembert,

Could you tell us what the maximum DD is intra-trade, not closed trade with this combination?

Adam[/QUOTE]

Hi Adam,

I can't because the source data for this test is based on closed trade data.

The test scenario is as follows :

- generate equity curve data in Excel based on gain/loss of each trade
- load that data into Amibroker as a ticker
- repeat for the next system
- run the optimization on the 'virtual' tickers

On the heels of Ray's analysis, I thought I'd update the values to see if there is correlation. I did this day-over-day from my last posting, just to see the progression and if daily volatility really moved the values.

Here's the latest correlation matrix:

[ATTACH=CONFIG]9148[/ATTACH]

Compared to two weeks ago, which you can view lower in this thread (see entry on 6/21), we've had a slight change but nothing major, as we would expect.

On 6/21 I calculated that the "optimal" levels were 73/27 in favor of IWM; this week the levels have moved a bit more in favor of IWM and the result is

[B]IWM: 78%
GDX: 22%
[/B]

In sliding-window backtests of where I use just the dates where the IWM or GDX was long (I haven't added short to the mix), I've seen variations as high as 100% IWM and as low as 62% IWM.

The methods of calculation are very different from Ray's but are in the ballpark, so the 1 part GDX to 2-parts IWM and a dash of lime still holds well.

I'll continue to watch and post these numbers, but based on my backtests, which use 100% invested, as well as just the long side of GDX and IWM, I think the 2:1 IWM:GDX is a good overall mix for "being in the ballpark".

[QUOTE=Rembert;14224]In attachement a picture of a test I did in Amibroker based on the robot trades. It shows (when not taking into account strong/weak signals) that the optimal allocation (highest return for the least risk) is :

[B]IWM: 68%
GDX: 32%[/B]

Of course like all backtests this is purely theoretical. But it kind of matches Paul's findings based on the ETF data alone. And it also matches Pascal's tests that show the combination of a double leverage Russel2K ETF combined with GDX works best.

If we'd be making a cocktail the recipe would be ... one part GDX, two parts IWM and a dash of lime.[/QUOTE]

Awesome stuff, Rembert.

Thank you sharing.

This morning I was thinking have I done this right. Because I've build the equity curves without re-investing the profits. So tonight I'll do another test with re-investing of profits. We'll see if that gives different results.

As promised in attachement an update of the test I did on the robot allocation levels.

[QUOTE=Rembert;14304]As promised in attachement an update of the test I did on the robot allocation levels.[/QUOTE]

This is excellent work. It confirms what I think some of us had gathered, as well.

Thank you again, Rembert.

[QUOTE=Rembert;14304]As promised in attachement an update of the test I did on the robot allocation levels.[/QUOTE]

Thank you Rembert. This is very instructive and helpful.
From our own research, optimal allocations can vary widely depending on the strength of the robot signals. For example, when GDX issues a very strong buy or sell signal, it is better to allocate 100% into GDX and 0% into IWM, whatever the strength of the IWM signal.
But under the same circumstances, with double leveraged IWM, it is best to allocate 50% to TWM/UWM and only 50% to GDX.
So, the use or lack of leverage is of major importance for correct allocation.
This research is still in progress and it tells us that the optimal allocation rules are far from straightforward when the systems like the robots can graduate leverage and the strength of the signals in presence.
Pascal and I will advise in this forum when we clearly see allocation edges based on the research.
Billy

Thanks. Looking forward to your findings.

For a next test I might build equity curves with position sizing built into them based on strong/weak signals and see what results that gives in an optimization test.

With all the great discussion ongoing regarding position sizing, I wonder if it's time to update this thread using recent Robot returns to determine optimum allocation percentages?