Friday, June 22, 2007

Trade Ideas: Data analysis

Regular readers will note when I publish the list of stocks drawn up by the Trade-Ideas scanner I also note the time required to select a "top-8" of stocks, and eight "20-appearance" or "20-hit" stocks (the latter set of stocks are what you see listed each day). The "top-8" picks are the eight most recent stocks selected from the close of the day, counting back. I exclude repeat counts for the same stock. I use the end-of-day as a baseline to keep the timing consistent across days. The "20-appearance" list are the eight stocks you see published; these are stocks which make 20 or more appearances on the scan (with Trade Ideas, if a stock meets the parameters of your scan it will be 'printed' in real-time, so stocks repeatedly meeting the criteria will be receive a fresh print on the scan. It is a great way to see which stocks are gathering the most buying momentum during the day). Some days enjoy broad buying which means numerous stocks are been printed, on other days there could be few, if any, stocks meeting the criteria set. I also note the time it takes to select eight stocks using this method.

But what, if any, is this data of help to us. Below are two graphs showing the raw data for the "top-8":

and the "20-appearance list" (I have less data here because I started doing this later):

For the "top-8" what is clear from the graph are the numerous spikes (of weakness) inter spaced with periods of 'speedy' buying. The spikes of weakness are usually associated with down periods in the market (which makes sense) - although the spike in November 2006 did not look to be associated with a major correction in the market.

A 10-day moving average for the data smooths the effect a little:

And removing the November 360 spike cleans the data up a little:

Can one draw a conclusion from the data?

For bulls, it is a little easier. If the scan shows strong buying into the close - i.e. a "top-8" list which is selected inside 10 minutes - you should consider buying stocks. It is perhaps best to use a moving average filter to protect against whipsaw common during the formation of a bottom. For example, May 15th 2006 saw the 10-day MA drop below 10 (=6.8 minutes) which would have been a false signal given it rose to above 11 minutes the next day. Broadening the filter to two consecutive days below 10 minutes would have given a signal for September 6th (=4.7 minutes) as the 10-day MA stayed below 10 minutes until September 21st - a decent time to buy. Even with that, when the Nasdaq was stumbling around the 2,000 level in July and August the Trade-Ideas scan was showing strong buying into the close; for the 14th, 24th, 25th, and 26th of July when the Nasdaq closed at 2,037, 2,062, 2,074, 2,070 - the Trade Ideas scan had selected 8 stocks in 5 minutes, 1 minute, 4 minutes, and 1 minute. The July 14th close was 17 points away from the 2006 low of 2,020 made on July 21st (and which I have not data to show what the level of buying was - so it may have been strong then too).

Once your in - how long should you hold? The 10-day MA helps again, although the data is not so clear as to when a top is likely to occur. When the 10-day MA rises above 30-minutes and/or individual day spikes climb above 100 minutes then things are perhaps getting a little hot and one should consider tightening stops or taking profits. The November 2006 spike to 360 may have been a mixed blessing as selling at that time would not have been a bad idea given it would have skewed the 10-day MA into a sell signal. Excluding the spike from the data would have meant a 10-day MA sell trigger on January 5th (perhaps a 1% gain over the November signal).

Unfortunately, the problem with the 10-day MA filter as a sell signal becomes apparent on the next signal: one would have re-entered the market after February 13th when there was two consecutive days the 10-MA was below 10 (6.8 and 6.7 minutes) and exited after March 7th when the 10-day MA got to 34 minutes - but this would have meant selling at the very low the market made during this correction. The next entry signal would be after April 2nd (5.8 and 3.7 minutes) with an exit after June 20th (31.4 minutes, although there was an individual day spike to 110 minutes on June 12th).

For the "20-appearance" timed data is harder to draw any conclusions. For one, I don't have a lot of data, and the data I have doesn't appear to show any clear evidence for trading one way or the other. The 10-day MA has an interesting spike low for June 4th (71 minutes) which should be bullish. Anything around the 120 minute mark for the 10-day MA looks to be a relatively good time to buy. When to sell? Hard to tell.

One avenue to look for in this regard (to sell) may be the period of time between days when the scan took the entire day to come up with 8 or fewer stocks. The narrower these spikes run together - the greater the chance for a top.

An easier use for the "20-appearance" hit list is to buy those stocks when the "top-8" list generates a 'buy'. The table below shows the performance of the stocks traded according to the 10-day MA signals derived from the "top-8" list.


[1] Buy stock(s) from the "20-appearance" list at the next days open, for the day the 10-day MA of the "top-8" list is below 10 minutes for the second consecutive day
[2] Sell at the next day's open when the 10-day MA for the "top-8" list is over 30 minutes.

Unfortunately, the overall returns from this system were low. In part, because of the buying and selling for the February-March leg, but also because of a disappointing under performance of picks relative to buying the ETFs of the key indices. It is easy to blame the sell-side signal given the buy signal will be favorably biased in a bull market, but even allowing for that it may be necessary to adopt a different MA (perhaps an exponential moving average) to help generate the sell signals - going back to the late March 'sell' signal, the timing of the "top-8" pick list had started to indicate weakness on February 22nd when it went from a prior 6 days worth of sub-10 minute readings up to 23 minutes, and was followed by subsequent days of 14 and 16 minutes until the February 27th crash and the 52 minutes for that day. But it took until March 8th before this data influenced the 10-day MA enough to issue a 'sell', which by then was to late. What is interesting about the February 22nd reading was it followed the day the Nasdaq made its absolute high for the month:

In summary, the rate of buying as measured by the "top-8" list (and perhaps the "20-appearance" list to a lesser degree) is a relatively good indicator to use when entering the market using 10-day MA approach outlined above, and may have use in a 'buy-and-hold' strategy when one is looking to buy on dips as bottoms are quickly established by an increase in the rate of buying. It is a poor indicator for marking absolute tops and should not be used as a method to exit stocks, although a rise in the length of time (i.e. a slow down) required to select 8 stocks is a fair warning of an imminent reaction top approaching.

If you would like to try a full version of the Trade-Ideas software, follow this link for a free 7-day trial.