Please read the disclaimer at the bottom of this page

Thursday, May 31, 2007

New hedge

All positions have been adjusted to match the new portfolio weights. I also updated the hedge. I closed the previous DIA put (Sep '07 strike 123). The previous SPY put is pretty much worthless. The new hedge consists of some SPY Sep '07 puts with strike 144. If the SPY would drop by 6%, the delta should be app. equal to my portfolio beta of 1.6. Thus I'd expect to be about market-neutral below any 6% drop in the SPY. The reason for choosing 6% is that this is the largest drawdown for the SPY over the 1-year period for which I have data for all 3 systems in my portfolio.

Wednesday, May 30, 2007

Tomorrow: portfolio update

Tomorrow is the last day of the month, a day on which I will implement the new portfolio weights. Because extreme-os positions are always open for one or two days at most, I don't have to change any of the actual positions. I can just change the auto trading settings on C2, such that the next trades are scaled properly.

For Weekend Trader and Trend Plays #1 it's a little different. They currently have open positions, which will probably stay open for quite a while. What I then do is simply adjust those positions such that they will reflect the new portfolio weights.

As I wrote earlier this week, the new weights will be:
extreme-os 127%
Weekend Trader 53%
Trend Plays #1 20%

Monday, May 28, 2007

New Download Section

I added a download section on the right-hand side. The first file available for download contains my real-life fills since March this year for extreme-os.

What do Weekend Trader's and Trend Plays#1 Sharpe ratios say?

Taking the analysis in my previous post one step further, we can repeat it with the current performance of systems in the portfolio as inputs.
For example, Weekend Trader has a history of 404 trading days, and a Sharpe ratio of 1.68. What I want to know is: What happened to systems with an equal or longer history and a Sharpe ratio of at least 1.68 after they'd been around for 404 trading days?

It turns out 3 systems qualify: Mutual Fund Trader, extreme-os, and ARS.

These are their Sharpe ratios for the period starting after their first 404 trading days:
- Mutual Fund Trader 1.01
- extreme-os 4.01
- ARS 0.87

These odds are not really bad! (But remember that slippage and transaction cost are not accounted for)

We can repeat the same for Trend Plays #1. This system currently has a history of 241 trading days, and a Sharpe ratio of 2.40. Applying the same criteria to all other systems (including those that are terminated, of course), the following qualify: Mutual Fund Trader, extreme-os, vicinoo! trading - Daytrading, and Tango. Each of these had a Sharpe ratio > 2.40 over its first 241 trading days. How did they do afterwards?
- Mutual Fund Trader 1.01
- extreme-os 3.74
- vicinoo! trading - Daytrading 1.92
- Tango 2.14

Not bad either!

There's one big caveat to this kind of analysis: The underlying trend of the market was mostly upwards. If tomorrow a 3-year bear market would start, Sharpe ratios might not look very good for the next 3 years. One way around this would be to look at alphas instead of Sharpe ratios, which could be a good topic for future analysis.

What does a high Sharpe ratio say?

The Sharpe ratio is often used as a measure of reward/risk and might be interesting as a criterion to select a system on C2. However, does past performance on the Sharpe ratio predict future performance? I'm not ready yet to show a rigorous analysis, but some preliminary findings might be interesting.

What I did a few weeks ago was to select all stock systems with at least a 1-year history (on C2). I then selected only those with a Sharpe ratio of at least 2 in the first year (250 trading days) since they appeared on C2. Finally, I calculated the Sharpe ratio over the period after the first year (not taking into account slippage and transaction cost). The underlying idea is: Suppose we would pick systems with Sharpe ratios > 2 and at least a 1 year history, how would they do afterwards.

The results are very interesting. I found 6 systems trading stocks with Sharpe ratio's >2 in their first year:
Mutual Fund Trader 2.25
extreme-os 4.53
vicinoo! trading - Daytrading 2.74
Tango 3.64
MBN-1 2.03
Weekend Trader 2.28

These are the Sharpe ratios I calculated for the period after the first year (note the length of this period is different per system, with the shortest being 89 trading days for MBN-1 and the longest being 399 days for Mutual Fund Trader):
Mutual Fund Trader 1.09
extreme-os 3.87
vicinoo! trading - Daytrading 2.48
Tango 2.03
MBN-1 -0.18
Weekend Trader 1.19

Interestingly, all Sharpe ratios in the second period are lower than in the first, but still very acceptable, except for one system: MBN-1. The ranking of systems is almost the same in both periods (except for Tango).

This seems to suggest that the Sharpe ratio has been quite informative for stock systems till so far. Of course, this is no guarantee that it will stay this informative in the future.

Saturday, May 26, 2007

New portfolio weights

As I wrote yesterday, at the end of this month I will update the portfolio weights. The new weights will be:
extreme-os 127%
Weekend Trader 53%
Trend Plays #1 20%

Because Trend Plays has a shorter history and I just started trading this system, I am scaling in slowly. If it keeps performing well, I might increase to 30% by the end of June. There's now also a better balance between extreme-os and Weekend Trader.

What we gain from this change in portfolio weights is higher Sharpe and Sortino ratios.
Sharpe = 2.91 [0.91 , 4.88]
Sortino = 4.39 [1.22 , 8.35]

The downside is a portfolio with a higher beta and higher correlation with the general market.
Alpha (annualized) = 49%
Beta (SPY) = 1.29
Beta (DIA) = 0.35
R-squared = 0.38
N = 241 days

PC maintenance

During the weekend I always run the following programs to keep my PC clean:
virus: AVG
spyware: Ad-Aware, AVG
registry: regvac
cache: ccleaner

McAfee and Defender are running on a daily basis, so in a way AVG should be redundant.

I also clean and defragment the HD, and check for any TradeBullet and TWS updates. Just installed TradeBullet build 527 and TWS build 872.6.

Friday, May 25, 2007

Comments are welcome

I hope communication won't be too much of a one-way experience. Feel free to comment on my posts (either with your name, alias, or anonymous). I'd also appreciate your votes which system I should add next to the portfolio.

Sticking to the system

Big profits today, courtesy of extreme-os. This system didn't do as well as it was expected to do in the last few months. Some subscribers were asking for modifications to the timing of entries and exits. Because of the losses, some also appeared to be scaling down. One of the benefits of subscribing to systems with a longer history is that it is easier to maintain confidence in the system when it doesn't perform as well as you hope (which can always happen). I didn't have any reason to doubt the system, as most of its performance measures (e.g. rolling 50 day Sharpe ratio, drawdown etc.) did not really exceed their historical lows. So, I was sticking to the system. Today it paid off.

The ability to stick to the system is one of the most important reasons to trade a portfolio of systems. Ideally, the weight of each system in the portfolio should be small enough that a total blow up of one system would not destroy the entire portfolio. This is obviously not the case in my current portfolio, but we'll get there over time. At the end of this month I'm planning to increase the weight of Trend Plays #1 from 10 to 20 percent and make some changes to the weights of Weekend Trader and extreme-os as well.

Thursday, May 24, 2007

Portfolio up, S&P down

It's always nice to see a long-only portfolio go up when the broader market goes down. Today, the S&P500 lost close to 1%. Historically, days with losses of more than 1% in the S&P500 have occurred less than 10% of the time since 1950. The portfolio is up by ~0.3% today, mainly as a result of a profitable trade by extreme-os.

Tuesday, May 22, 2007

extreme-os execution (2)

Today the same problem as yesterday occurred, and I wasn't filled again on an extreme-os trade. I have decided to make a change to the TradeBullet autotrading settings, and checked the option "Convert limit into market orders once C2 reports a fill" with a delay of 5 seconds. That should take care of the problem, even though I can expect a fair amount of slippage on those trades that I would have missed otherwise.

extreme-os execution

In addition to the results of statistical analyses, general ideas and thoughts, I also keep track of execution issues. For example, I discovered that one of the extreme-os signals was not executed last Friday, and I missed a profitable trade. The vendor of extreme-os typically enters the entry order with a limit price above the current price. This is supposedly to prevent slippage. The lag between the time the order is issued by him and executed by TradeBullet can be in the order of seconds (TradeBullet seems to poll the C2 server only every 7 seconds, for example), and therefore it is possible that the price already moved beyond the limit price, and the trade will not be executed. Fortunately it doesn't happen very often, but when it happens and the trade is profitable it's unfortunate.

Monday, May 21, 2007

Weekend Trader

Today one of Weekend Trader's four positions was replaced by another one. I didn't have any problems replicating the C2 fills, but from the reviews on the C2 site, it seems some subscribers are afraid that the signals generated by the system have an impact on the price. If that's the case, subscribers should be able to match the C2 reported fills, but might experience lower profits compared to those reported in the past (when the number of subscribers was perhaps lower).

One way to test this, would be to calculate for each stock ever traded by Weekend Trader, the ratio:

Monday's Open / Preceding Friday's Close - 1

(i.e. the return over the weekend)

and estimate if a difference exists between the Monday the signal was issued and all other Mondays in a year.

Some hypotheses:
- We would expect a positive effect for buy signals, and negative effect for sell signals
- We would expect the effect to become larger over time (as the number of subscribers has been growing most likely)
- We would expect not to find a similar effect for other days of the week

Sunday, May 20, 2007

Alpha, Beta, Sharpe and Sortino

Some stats on the (backtested) portfolio:

2-index model, with DIA and SPY ETF as indices:
Alpha = 41% (annualized)
Beta (DIA) = 0.67
Beta (SPY) = 0.85
R-squared = 0.25
N = 236 days

Sharpe ratio [95% CI] = 2.40 [0.47 , 4.28]
Sortino ratio [95% CI] = 3.62 [0.57 , 7.81]

Correlations




Correlations between the returns of the 3 systems are small or negative.

Drawdowns


And these are the (backtested) drawdowns.

Backtested Portfolio Return

This is what the portfolio return (excl. options) would have looked like last year, assuming it was daily rebalanced. I assumed for extreme-os slippage per trade was 0.26% of the price (0.13% per side, which equals the average slippage in my own account).

Saturday, May 19, 2007

Infrastructure

I use Interactive Brokers as my broker and TradeBullet to "auto trade" extreme-os. Weekend Trader and Trend Plays #1 can be traded manually.

My internet provider is TimeWarner.

Hardware:
- Dell Dimension 4700, 2.5G memory, 80G HD
- Dell Dimension 1100 (backup)

Other tools:
- LogMeIn (for remote access)
- GoToMyPC (backup remote access)

Software for analyses:
- R
- Excel VBA

Current Portfolio

To kick it off, let's start with my current portfolio:
- extreme-os 178%
- Weekend Trader 12%
- Trend Plays #1 10%

The percentages refer to the equity allocated to each system. It sums to 200%, which means I am 2:1 leveraged. In addition to this portfolio I also hold option positions to limit my downside risk:
- DIA SEP07 123 P
- SPY JUN07 135 P

Welcome!

I used to keep a private notebook on my Collective2 trading. I thought it makes sense to continue this as a blog, as this is perhaps useful information to vendors and subscribers at C2.