What should I do with my model?

Quote from galilean:

Acutally, it only does 5 to 10 trades a day. Ideally, those trades should be done a few minutes before market close because my model is based on the daily close prices. You only hold up to 7 stocks. I created a software for it. It can download the latest quote, do some analysis and displays which stocks triggered buy or sell signals. It could even link to your trading account to place the orders if you want. By the way, in my backtesting, I only consider stocks greater than $5.00 and average volume > 500K.
More...

Based upon your comments about trading fractal, trade frequency, streams of capital running, the price range, and time of trades, I have some recommendations on the pair of indicators you are using, etc..

By comingling the market analysis and the instruments' characterisitics, you must use the pair of indicators to que your preference of crossover candidates to go from one instrument to another in each of your seven streams. The originator of MACD uses absolute values for doing a go/no go test but you limit yourself to gating by observing a contrast in relative values at a time that is most remote (stale data) from the time of data taking.
You sneak a peak as you say, to deal with this.

For EOD trading it is true that the two segments of the index profile are opposites over long times (RTH is net negative and non RTH is net positive) but for short term trading (your average hold is a day) even longer than you do, the short term cycle net well overrules the long term trend of markets. This means that you have relatively poor entries because of timing and relatively poor exits as well. I assume you exit to have capital for a better trade that is available.

I admit my fav type of trade is one that frontruns opening gaps by taking the trade along with the bottom fishers and after the day traders leaving for the day. It is in contrast to what you do sort of since the hold for frontrunning a gap trade is like your hold. We use differing fractals to achieve the same hold and I do use MACD differently than you do and not as a trigger. So I recommend that you do a MACD multi-fractal assessment rather than a MACD pair on the EOD fractal.

Doubling every 2 to 3 months is not the ballbark for one day holds (buying near close and selling near close on day plus 1) on multi-streams. I recommend making the market analysis and trading strategy each autonomous to begin to look into this. Your instrument criteria is almost absent. Were you to sharpen it, you would get the benefits of "they all look alike to me" and the high Beta. Orijinal, obviously does not trade with high Beta stocks among other things, for example.

The trading strategy, unhitched from the markets, would allow you to see the opportunity of having timing working for you. Using a one day hold time on average as a stategy means you have to to pick the portion of the instrument cycle that has the highest money velocity. Opening gaps are best, but you do not select your trades to do that. The alternative available with MACD pairs (either on one fractal or on two fractals) is the breakout portion of a cycle which occurs and completes during RTH of one day. So you would have to revise the data analysis time in your strategy and use a faster fractal to get the signal.

Doubling money involves doing a set of cycles. The one day hold is the cycle time so doubling comes down to profit per cycle. There is no way your approach can take 40 to 60 days to double. It is not possible to make so little money per cycle to take that long. Taking 40 to 60 days is like trading noise or some other betting approach. Probably a portion of a month may be involved; for high Beta it is somewhat over a week using the one day hold you use.
 
Jack, actually, I'm not closing my positions at the end of the day. And I mostly hold stocks for 1 or 2 weeks. Also, the MACD I'm using is quite different from the one you usually find in other software package or text books, and I'm not simply using crossover as triggers.
 
I wouldn't trade that system. Looking at the equity curve alone, there are two 1 year periods before new equity highs, and some half year periods before new equity highs. Plus some big drawdowns to boot (50%+). I don't trade ones with greater than 15-20%, and on to of that your biggest drawdown is yet to come.

Judging by the results and your number of trades, your trading efficieny is probably quite low. I bet its less than 0.05 per share. Include slippage and commissions and it probably becomes a negative strategy. What profit factor does it have? Most people will tell you less than 1.5 will probably not work real-time (though I find < 1.5 can be profitable with a high trade frequency and where hypothetical results match real-time results closely)

Also, is any of that out of sample data? Or did you just over optimize it for 7 years? You need to walk forward your strategy for it to have ANY validity. Cause I could easily turn 10K into 1M in a year using optimized parameters in backtesting but I assure you it would not work in the future.

And how many shares are being traded in the later stage of your equity curve? If you're trading 500K at the end of the day in a stock that does 3m vol a day... well, I think you know. It's probably not as scalable as you think.

Plus, you lost me when I heard MACD. But its okay, just keep trying. We were all at that stage at one time.
 
Quote from galilean:

The problem is that I don't have enough capital to start with. From my study, I need at least $200K to get it started. I also don't have time for that.
More...

can you explain why the model was basically dead for 3 years?
 
Quote from galilean:

Jack, actually, I'm not closing my positions at the end of the day. And I mostly hold stocks for 1 or 2 weeks. Also, the MACD I'm using is quite different from the one you usually find in other software package or text books, and I'm not simply using crossover as triggers.
More...

I missunderstood your relationship of trades per day (5-10) and streams you run (7). I saw this as a one to one ratio instead of a 1 to 2 week hold (5 to 10 days).

Your comment here implies that your use of MACD is more of a filter than a signal or rather a combination filter and signal. I feel that indicators are good for timing (signals) rather than filtering. I now better under stand the level of your equity curve.

I ws not clear in my use of the word crossover. I was referring to a trading technique rather than an indicator signal. At some pint to improve the annual turnovers, people consider crossover trading. Usually this occurs at the 80 plus turnover breakpoint. You are still in the money velocity range of 30 to 40 yielding a compounded result of 40 to 50% a year. a specific example is where a trader compresses a 3 day hold to 2 and 1/2 days. this takes the 80 turn overs per year to 100 which is a nice effect when profits per cycle happen to be maintained.

Specifically, using an exit signal around 11:00 am and then reentering with the capital in the late pm (for an opening gap the next day) gives an overlap day.

The most significant increase in equity curves comes from doing shorter and shorter cycles without sacrificing cycle profits. I use a nominal cycle profit of 10% as a criteria.

Moving to a restricted universe is a powerful tool too. A 3 Beta minimum is easy to obtain and this also narrows the chosen price range of trading which is vey volatile (meaning profitable in short term trades).

I agree, no one really uses the text version of MACD to do high velocity trading; the text versions and their signals "bridge" trading opportunities and put a person in the Intermediate Term Trend trading ballpark.

Making 10% evey two or three days is a smooth and convenient way to trade equities, in my opinion.

thanks for responding. I see I drew a few humorous posts; enjoy them.
 
Quote from jack hershey:

Moving to a restricted universe is a powerful tool too. A 3 Beta minimum is easy to obtain and this also narrows the chosen price range of trading which is vey volatile (meaning profitable in short term trades).
More...

LOL Hershey you are completely mad

you are a gonner hershey boy

if I was a man of god

I would pray for you :( :(
 
Which is why your only documented performance is -24%?
Quote from jack hershey:

Making 10% evey two or three days is a smooth and convenient way to trade equities, in my opinion.
More...
 
Thanks for all the comments.

My model is based on MACD which is, yeah, lagging in its nature. Also, it does long only (no shorts), so if the market keeps going down, it will not make money. This is the situation around 2003. I know many people were losing big money around that time, so I'm glad with my model. I think if I could include shorts (that's I plan to do next), it will perform better.

Why the model was basically dead for 3 years? That's not true. That was just the scaling issue. I should probably use log vertical scale to filter out the high growth in later years.

In my backtesting, I already considered a commission fee of $5.00 per transacton (this is the flat rate at TradeKing.com) - I was told you could get much better rate if you could associate with a brokerage house (right?). I don't think slippage is an issue for me because I'm using the daily closing price in backtesting. Some brokerages allows you to place trades using closing price, but the only conflict with my model is that the trade must be places 20 min. before market closes and once you placed the order, you can't cancel it. Trading on other day time may result better, but I don't have historical data to back test it.

For each parameter set, my backtest runs 6 times with different starting point (Jan. 1 of each year starting from 2000) until end of 2006.The curve attached to this forum is just one of the cases whose parameters seem to be a good compromise. With this parameter set, I could confirm the similar return in year 2007. Yes, I have seen cases with much higher but inconsistent returns, which I consider to be too optimized and thus is not selected for use.

I use MACD as filter and as signal. First, I scan all US stocks (about 1200 stocks with price > 5 and volume > 500K) to rank the stocks. This scaning is done once per month. I pick 100 top performer from this rank list and on each day I run the (more intensive) analysis against those 100 stocks for trading (buy or sell) signals.
 
There is no special handling for survisorship bias. But it's unlikely that a company will be picked up by my model when it's going to be out of business next day.
 
You must log in or register to reply here.
Back
Top