Olivers Weekly Journal

[Anonymous]

Morning Said,

Here are some prelim stats, trading stocks from the S&P500, taking 10 trades at 10% of equity. This run uses a 150 day look back period, and removes any stocks with a 10% gap up or more over the past 2 months. There is a 200 day SMA index filter. Portfolio goes to 100% cash if S&P500 is below the 200 day on the 1st of the month when the re-balance takes place.

20 year test up until Dec 2015

CAR – 20.19%
RAR – 27.80%
Win% – 58%
Profit factor – 2.22
Max DD – 31.23%

Attached is the monthly return table and rolling 12 months returns, stepped monthly. Of the 230 12 month windows, 83% positive, 10% negative, 7% flat.

Fairly large range in the 12 months returns that you could expect.

Regards

[Anonymous]

Here is a look at the CAR% for the different look back periods over the past 20 years and past 10 years.

Over the past 20 years, there is a bit of a hump around the 150 day area, then it drops off, only to improve with a look back period in the 225 to 250 day range.

Over the past 10 years, again there is a bit of a hump around the 150 day area, but then it drops off quite considerably.

I would much prefer to see a more stable plateau for the look back period.

[Nick Radge]

The initial chart is not that bad – the variance of returns is quite small. I really don’t think many will ‘feel’ the difference between 17% and 18% when trading.

The second chart is more concerning because there is considerable variance.

[SaidBitar]

thanks for the statistics

have you tried with 20 positions using the same position sizing as in stocks on the move.

i believe drawdowns will be lower

[Anonymous]

Hi Said,

I have tried a couple of different position sizing methods. Results attached.

1. 10 positions @ 10% of equity
2. 20 positions @ 5% of equity
3. Risk parity position sizing using a factor of 0.1
4. Risk parity position sizing using a factor of 0.15

Just note, for the 10 positions at 10%, the slight difference between the results attached and the ones I posted yesterday, is that I just excluded all stocks under US$5 a share.

Risk adjusted position sizing indeed drops the max draw down.

[Anonymous]

One thing that I realised with using risk adjusted position sizing is that you can’t also stipulate a fixed number of holdings. As Clenow says in his book, when buying, you would start at the top of the list working your way down until you run out of cash.

The charts below shows the exposures when using both 0.1 and 0.15 as the risk factor.

[Anonymous]

I wonder if looking at just the largest draw down throughout the history can be a little misleading. The below 2 charts are of the max draw down from firstly the 10 positions @ 10% of equity, and also using the 0.15 risk adjusted.

While the risk adjusted method dose reduce the maximum draw down from 31.46% to 24.32%, the next 3-4 largest seem closer in size.

Does anyone know how to get stats on say the 5 largest draw downs??

Regards

Oliver

[TrentRothall]

Have you calculated Max Wait of the systems to see how llong the DDs last? i guess with trading monthly some drawdowns can last years if you go to cash for ages.

PS i think if you press insert on your attachments they should be visible in the post

[Nick Radge]

Quote:

Does anyone know how to get stats on say the 5 largest draw downs??

Use this in the Custom Backtester:

Code:

VarSet(“DD2”, 100);

eq = bo.EquityArray();
drPerc = -100*(eq/Highest(eq) – 1);

dd = 0;
bslh = HighestBars(eq);
bslh[BarCount-1] = 0; // need to catch last DD
dd = IIf(bslh == 0, IIf(Ref(bslh, -1)==0, 0, HHV(drPerc, Ref(bslh, -1)+1)), 0);

ddLast = LastValue(Highest(dd));

VarSet(“DD1”, ddLast);
bo.AddCustomMetric(“DD1”, ddLast);

for (nDD = 2; nDD <= 5; nDD++) { ddNew = LastValue(Highest(IIf(dd < ddLast, dd, 0))); VarSet("DD" + nDD, ddNew); ddLast = ddNew; bo.AddCustomMetric("DD"+nDD, ddLast); } bo.AddCustomMetric("MDD Cur", LastValue(drPerc));

[Anonymous]

Thanks Nick.

[Anonymous]

I have changed tact a little over the past couple of weeks. Initially I was looking at using exponential regression (ExReg) as a way to rank stocks, however, I decided to look at ROC after discovering that ExReg didn’t produce as consistent results over different look back lengths.

I also found that using a combination of 2 look back periods as opposed to just 1 seems to work quite well.

When using 2 periods, I just used an equal weighting of the annualised ROC over the 2 periods to determine the final score to rank with. I found that a 11-12 month look back period combined with a 5-6 month look back period worked quite well. I have attached a chart which shows the top 10% or so of different look back combinations between 100 and 250 days.

I’ll post the back tested results in a bit.

The model uses a 200 day MA index filter, and re-balances the portfolio on the 1st of each month. The S&P500 has just rallied back above the 200 day MA, so if it stays above come 1 April, the model will be buying.

Regards

Oliver

[Anonymous]

Below are a few stats for those interested, over the past 15 years. The first using 10 positions at equal weighting and the second using 0.15 Risk adjusted position sizing. Exposure is only 66%, because there has been 2 bear markets over that time, whereby the model has been in cash.

Is it fair to use RAR% as the best performance measure for a model like this, considering that it is either fully invested or fully in cash? When in cash, capital can be deployed elsewhere.

The second chart shows the rolling 12 month returns using the 10 positions @ 10% of equity. Of the 169 12 month periods, 142 (84%) are positive, 14 (8.3%) are negative, and 13 (7.70%) are zero.

A model like this can spend plenty of time out of the market. As an example, it was in cash for all of 2008 and the first 6 months of 2009. There is also a large variance in the 12 month returns. But considering it is fully invested, it is a huge improvement over buy and hold.

All in all, while there is a lot to like about a momentum model, I feel to follow it with ease, you need a longish time horizon, and other models to compliment both the volatility and times out of the market.

From next week, I am going to work on mean reversion systems for both AU and US.

Regards

Oliver

[Nick Radge]

Quote:

Is it fair to use RAR% as the best performance measure for a model like this, considering that it is either fully invested or fully in cash?

Not sure why that would be useful.

[Anonymous]

I have been testing out different mean reversion systems over the past couple of weeks. Apologies for not posting much on my progress, just been busy.

I am away for the next 2 weeks on a holiday, so will get back into it when I return.

Regards

Oliver

[Nick Radge]

You’re famous!

[video width=425 height=344 type=youtube]k9XS0vWkkIg[/video]

[Author]

Posts