Tuesday, April 17, 2007

Trade of the Day - ACADIA Pharmaceuticals Inc. (Public, NASDAQ:ACAD)

ACAD carved out a NR7 bar yesterday and based on yesterday's late day price action, I indicated in last night's watch list that I thought it would break to the upside. It did and this is how I traded it. Very choppy, I might add. A better NR7 play would have been SMSI, but I did not pick up on it in last night's review.

JCOM was a continuation play from Sunday's watch list. I think it needs to pull back here because it has had an extended run.
AKAM and GSIC did not trigger an entry. NVDA took out its pivot point in pre-market and I tried to pick some up on a pullback, but did not get a fill. I did, however, buy on a break of the pre-market high for a quick 50 cent scalp on the open. NVDA was not low risk and I do not recommend that tactic to new traders. VRSN swing was stopped out.

7 comments:

Anonymous said...

Great play on ACAD. I was watching it on your recommendation, but I didn't trade it.

Did you buy based on it breaking your upper triangle line, or was it just a break of yesterday's high? On either, did you just buy market at the price plus two ticks? What was your initial stop? If it was the prior day's low, then this was only a little more than 1R...

I've been trying to put together an NR7 breakout system, but my thoughts are very scattered on it. I don't have an intuition for what's important to screen candidates in or out. :(

TJ said...

Hi Prospectus,

The ACAD trade was a buy stop entry on the edge of the triangle. My preliminary stop was just pennies below the OR low. These NR7 price expansion trades generally take off swiftly and are often very choppy as this one was so I understand why you didn't step in.

I noticed SMSI and TZOO had nice NR7 rallies today. I tried to write a scan to pre-screen NR7 candidates but I'm not strong enough in math to figure it out. All I can come up with is inside bars. This usually results in a huge list and then I have to eye ball each chart to find suitable candidates. So usually I don't bother. I had traded ACAD when it gapped up a few weeks back and had kept it in my database, so I found it haphazardly.

I've thought about keeping a folder of breakaway gappers in stockcharts and running the scan on the gappers in order to filter out a lot of the extra noise.

Any ideas on how to write a better scan would be appreciated.

Prospectus said...

Well, I'm great in math, but not strong enough in trading, so maybe we can help each other out.

Here's how I find NR7's on daily charts:

I use an awesome tool called StockFetcher. It's free, but you only get some limits on features. If you subscribe there's tons of stuff is can do, including backtesting. The scans can be constructed out of phrases in english, so it's actually very easy to use, though still powerful.

I use the StockFetcher 2.0 interface, found here. I wrote my own scan for finding NR7's. You can paste this into a scan and save it:

Show stocks where close is above 1.00
and average volume(30) is above 250000
and volume is above 50000
and close has been above ema(20) for the past 7 days
and day range is less than day range 1 days ago
and day range is less than day range 2 days ago
and day range is less than day range 3 days ago
and day range is less than day range 4 days ago
and day range is less than day range 5 days ago
and day range is less than day range 6 days ago
and average day range(20) is greater than 6%
and do not draw day range
and add column high
and add column low
and date offset is 0


Most of those things are self explanatory, but the day range is just the range the stock had over a day (high - low). Average day range is just an average of the day ranges. Date offset lets you see scan results from the past--an offset of 2 gives results from 2 days ago. Also, once the market is open, it uses the current day's prices (delayed 15 mins), so to get your results from yesterday's close, you'd need to change the offset to 1, or you'd get all of the NR7's that are still forming today, many of which could not be NR7's at the end of the day.

I could use some help in tweaking the parameters. The volume numbers are just pulled out of the air. I put a constraint of a > 6% average day range over the last 20 days to try to get things that really move, but I don't know if that's the best way to do it. Also, I put close above ema(20) for the last 7 days to find things that are in a bullish mode instead of NR7's that are chopping around or in a weak phase. Once again, is this a good idea?

You could write scans that look for stocks that made a new 52 week high in the last 2 weeks and then print an NR7, or for stocks that gapped away and then print an NR7 within a week, or whatever you want. If you've got the trading idea, I can write a StockFetcher scan for it.

Prospectus said...

As an example, here's the results my scan came up with this morning:

DNDN, NUVO, BPG, SIGA, OPBL, THLD, LJPC, TGX, CXTI, PCYC, SWEB

TJ said...

Congrats Prospectus, your scan is much more sophisticated than mine.

DNDN had a true NR7 on April 4th, as did LJPC on April 4&5. Both of these were followed by price expansion. Most of the rest of these are the narrowest bars over the last 7 days, however, many do not have the requisite volume contraction.

Do you think that you can add an element to scan for volume contraction as well?

Thanks!

Prospectus said...

Sure, I'll add a volume contraction in there. A couple of questions: Do you just want a net volume decrease, from day 1 to day 7, or an orderly decrease each day? Should volume decrease by 50%, or less or more?

What's the difference between a true NR7 and one that's not true? Just the volume expansion? Any comment on the other scan elements?

TJ said...

That's right, a true NR7 contracts in both price and volume.

We only need a significant volume contraction on the NR7 day.

If we use DNDN April 4th as an example of a true NR7, volume contracted 42.5% from April 3rd to April 4th.

I don't want to make the scan too rigid. Perhaps if we say a minimum 20% volume contraction on the 7th day, we could probably eliminate a lot of the less interesting scan results. Does this make sense from a mathematical modelling standpoint?