GPS: The Goodman Price/Volume Stock Selection System
Monday, January 3, 2005
Fred Goodman

At last -- Fred develops a technical trading method for individual stocks.
HOME -->
Fred Goodman's Key Technical Indicators for Investment Success
Latest Goodman
Goodman archives

Be sure to check the Frequently Asked Questions page for the latest insights on GPS.

I'm excited to announce a breakthrough that extends my technical methods of market timing into new ground -- the selection of individual stocks. It's based on a modern adaptation of the nearly forgotten technique of Price/Volume Charting -- originated a half century ago by Benjamin Crocker (for background, you can read more about Price/Volume Charting by clicking here).

Price/Volume Charting is a very effective method of generating short-term market timing signals, yet the technique has been passed by in the computer age because it requires labor-intensive personal interpretation of visually complicated charts. It's hard enough to apply it to a handful of market indices every day (believe me, I know -- I've been doing it for years). So applying it to hundreds or even thousands of individual stocks has always seemed like an impossible dream. Until now.

After years of research I am now ready to introduce the Goodman Price/Volume Stock Selection System -- or GPS for short. Starting immediately, I will be using it for my investment management clients. I will also report on GPS in every edition of my online newsletter, Technical Excellence. I have started three model portfolios as of the last market day of 2004 -- one based on the stocks in the S&P 100 Index, another based on the stocks in the NASDAQ 100 Index, and a third that combines the first two portfolios with my market timing strategies. As time goes by, I may decide to make GPS the focus of a separate publication with a separate subscription fee.

PERFORMANCE   I have a great deal of confidence in GPS. I have been successfully "paper-trading" it for the last six months, and have rigorously backtested it back to mid-2001 to observe how it behaves in both bull markets and bear markets (for more information on my backtesting methodology, click here). Note that the backtested performance figures given here are based on trading at the closing price on the day trading signals are given, assume no commissions or other transaction costs, and we ignore the receipt of dividends. Past performance and backtests are not guarantees of future returns.

  • Based on my backtest, selecting stocks to buy and sell from among the 100 that make up the S&P 100 Index (OEX), GPS beat the benchmark index overall and in every individual period, with generally slightly lower risk (annualized volatility).
  • Since inception on July 2, 2002, GPS returned 33% cumulatively (not including dividends), compared to a loss of 9% by the index. In the bear market period through the bottom of October 9, 2002, GPS barely edged out the benchmark, losing 37% cumulatively compared to a loss of 38%. In the bull market period starting on the October 9, 2002 bottom, GPS far outpaced the benchmark, returning 109% cumulatively compared to 46%.

  • Similar overall results were obtained by buying and selling the 100 stocks that make up the NASDAQ 100 Index (NDX). Again, GPS beat the benchmark index overall and in every individual period, with generally slightly lower risk.
  • Since inception on July 2, 2002, GPS returned 42% cumulatively (not including dividends), compared to a loss of 11% by the index. In the bear market period through the bottom of October 7, 2002, GPS beat the benchmark, losing 41% cumulatively compared to a loss of 56%. In the bull market period starting on the October 7, 2002 bottom, GPS outpaced the benchmark, returning 140% cumulatively compared to 101%.

HOW DOES GPS WORK?   GPS is based on Price/Volume Charting, which is a systematic discipline for identifying bullish and bearish trading patterns. Most traders already have intuitions about how prices and volume interact. For example, we have all experienced "wash-outs" or "blow-offs" where markets reverse on high volume. And we've seen "fake-outs" where markets make seemingly significant moves but on suspiciously low volume. And when markets don't move at all, but volume is high, we call it "churning." Benjamin Crocker built these intuitions into a formal market timing methodology by tracking how price/volume relationships evolve through time. By plotting price on one chart axis and volume on the other -- and then connecting the plot-points through time -- Crocker learned that clockwise patterns were bullish and counter-clockwise patterns were bearish.

It sounds simple -- but a typical Crocker Price/Volume Chart is a tangle of overlapping lines and loops, as readers of my daily reports know. Crocker developed numerous subtle rules for discerning price/volume patterns out of his tangled charts, and interpreting them as market timing signals. Crocker's rules take both time and artistry. With training, you can manage to do it for a handful of market indices every day. But there aren't enough hours in a day to apply the technique to hundreds of individual stocks, and it's too complex and too subtle to program a computer to do it for you.

So my challenge has been to simplify Crocker's complex and tangled charts into simple linear charts, and apply a set of rules so unambiguous that even a computer could determine buy and sell signals. Achieving this has depended on two techniques -- vector analysis and pattern recognition. While vector analysis has its origins in the work of mathematicians who lived nearly 400 years ago, pattern recognition is a relatively new technique, developed since computers have become fast enough to handle the large number of calculations required.

Vector analysis was essential to the first step -- converting Crocker charts into simple linear charts. Readers have already seen the results in the form of what I have called the Goodman Price/Volume Indicator, or GPVI -- which I have been applying to several stock indices since March in my daily reports. The Indicator takes the temporally restricted vectors used by Crocker in charting price and volume and creates a series of daily data points which may be charted without creating the crisscrossing lines that make Crocker charts so difficult to read.

As readers of my daily reports know, the charts of the Goodman Price/Volume Indicator are useful in and of themselves. But to create a system for automatically analyzing hundreds of stocks every day, we need a method by which a computer can analyze them without even physically drawing charts. And that's where pattern recognition comes in. I knew at the outset that I didn't want to study the price or volume of the stock in question directly, since those were already used in the calculation of the Goodman Price/Volume Indicator itself. Instead I focused on analyzing patterns in the indicator itself.

I started with the idea that I wanted to identify an "upside breakout" by the GPVI -- but what characteristics constitute such an event? One thought was to study the slope of the indicator to find accelerations in its rate of ascent. A change in slope can also be looked at as the net change over time -- the more the indicator advances in the least amount of time, the greater the slope will be. Another useful descriptor is the length of time spent in a trading range before a breakout occurs. Yet another important characteristic is the upward move required before the indicator can be considered to be in a breakout.

Using these ideas as the core, I set about to determine the parameters upon which decisions could be made concerning each of them. It turned out that four such parameters were required. With those parameters, I am able to transform the Goodman Price/Volume Indicator into an oscillator that swings back and forth between ever-changing upper and lower trigger levels. We buy a stock when the oscillator exceeds the upper trigger level and close it out when it drops below the lower trigger level.

The trigger levels move in response to parameters of time and distance mentioned above. Of course the trigger levels will vary from stock to stock depending upon the characteristics of its GPVI. And they will vary across time as the price/volume patterns of the market evolve. The determination and method for evolution of the parameters was achieved by optimizing against several years of past data. For more information on the parameters, see the section on backtesting methodology below. And for a detailed look at how GPS has adapted traditional Crocker Price/Volume Charting, review the case-history of Apple Computer, below, by clicking here.

TRADING STOCKS WITH GPS   There are any number of ways to put a stock selection system like GPS into action. My preference is for a simple, transparent approach that lets me any my readers easily evaluate the effectiveness of the system. So I have established three model portfolios as of the last market day of 2004, one based on the stocks in the S&P 100 Index, another based on the stocks in the NASDAQ 100 Index, and a third that combines the first two portfolios with my market timing strategies. When GPS gives a buy signal, we will buy. When it gives a sell signal we will close a position out (GPS does not currently support short-selling). On any day on which there is a GPS buy or sell signal for any stock in either portfolio, I will announce it the following morning in my report. Of course if I take a vacation day -- I typically take about 15 days off each year -- I'll cover any GPS signals in the next report.

Also, I'll reconstitute the portfolios every six months as I re-optimize the parameters of the GPS system. For more information on re-optimization, see the section on backtesting methodology below.

In addition, I will overlay a simple trading discipline on stocks once they are selected as buys by GPS. Any stock is automatically sold if its shows a loss after 30 days from the time it  drops back below the upper trigger.

One detail -- there may be times when the GPS system recommends a trade that I don't want to make. Sometimes when there is big news on a stock, there will be a big price/volume move that gets the system's attention -- but I still want to use my judgment to evaluate that news and see if I really want to buy. For example, this happened a couple months ago with the stock of Chiron, which fell dramatically on high volume because of the controversy concerning its tainted flu vaccines. The GPS system said "buy" -- but I would certainly have intervened in that case, judging the stock to be fundamentally "broken" by such catastrophic news. Such things will be rare, but part of the art of trading is being ready for them and responding appropriately. Remember, because I say it over and over -- no technical system is a "Holy Grail," nor an end in itself. There's no substitute for good human judgment.

On the web I'll update a running scorecard for both portfolios, and all their positions. New positions will enter the portfolios in approximately equal weights. I'll report each position's gain or loss versus its buy price, and the running overall gain or loss in the portfolios. We will assume that the portfolios are always 100% invested in stocks (that is, they will carry no cash balances at any time).

BACKTESTING METHODOLOGY  The GPS technique has been refined using optimization of test parameters on past stock price data. Such techniques have had considerable criticism heaped on them -- and rightly so -- for being used to "data mine" or "curve fit" past results. When one tightly fits an equation to past data it almost always ceases to fit new data quite quickly. To make my testing as robust as possible to markets as they evolve in real time, I have used techniques documented in the excellent book Design, Testing and Optimization of Treading Systems by Robert Pardo.

The ideal technique is to optimize parameters based on one set of data, and then test it on an entirely independent set of data.  To do this, I collected price and volume data for the six years from July 1998 through June 2004 for both the NASDAQ 100 and the S&P 100. I studied the data in seven overlapping, three-year segments, the first being July 1, 1998 to June 29, 2001 and the last being July 1, 2001 to June 30, 2004. I determined optimized parameters for each three-year segment, and tested each of them on the following six-month segment -- which, of course, was data that had not been included in the optimization. I will adhere to this technique going forward -- that is, I will re-optimize the parameters every six months based on the past three years.

Note that my backtests are based on trading at the closing price on the day trading signals are given, assume no commissions or other transaction costs, and ignore the receipt of dividends. Past performance and backtests are not guarantees of future returns.

THE CASE OF APPLE   Apple Computer in 2004 was a relatively easy stock to chart using Benjamin Crocker's original Price/Volume Charting techniques. That's because it continued to advance all year without much of a retracement, so the chart remained relatively uncluttered. But even at that, it is still nearly impossible to find the most important buy and sell signals that occurred on low volume. One important feature does stand out -- the many high volume thrusts, all of which were followed by higher prices.

Apple Computer (AAPL)
January 2004 through November 24th

A trader looking at this hopelessly cluttered chart would be attracted by the obvious high volume advances. But to study the finer features of price and volume action it is necessary to redraw the chart with fewer days. But on doing so, the prior reference points are lost. Nevertheless, there is value in doing so -- let's look at the first five months in detail so we can see how Price/Volume Charting forms the underpinnings for the new GPS system.

It is only when one takes a closer look that the real value of the Crocker charts becomes clear. In February 2004, as shown in the chart below, Apple made a breakout move a few days after the first trading day of the month (green marker). It advanced almost 10% before giving it all back. However, on the way down it completed a clockwise basic buy loop which I have colored in pink. Following the completed and confirmed buy loop (confirmation requires that the line crossed in completing the loop must not be re-crossed the next day), Apple went on to make the first of several high volume thrusts completed during the year.

I have identified the day the thrust was made in red on the chart below, since it occurred on the last day of the month. In the chart above, it appears as the lowest pink marker. Also in the chart above I marked a much bigger volume thrust with a second pink marker above it. That same thrust on around 27 million shares can be seen in detail in the chart below.

Apple Computer (AAPL)
February 2004

The green marker repeats the last day from the chart directly above, the one with the red marker. Immediately following the high volume thrust, a second thrust was made on even greater volume. As always, it is the low volume moves that are significant in that they alert one to the possibility of an upcoming high volume advance. In March a counterclockwise continuation buy loop followed a thrust and was itself followed by another upmove.

Apple Computer (AAPL)
March 2004

That upturn was followed by yet another one, even bigger than the last. However, Apple turned around, produced a sell loop and made a 50% retracement of its gain from 22 to 28 over the previous two months.

Apple Computer (AAPL)
April 2004

Another bullish low volume pattern followed, this time what Crocker referred to as a star.

Apple Computer (AAPL)
May 2004

In the final month that we'll cover, we can see that Apple made another major high volume thrust and broke out above 33, only half the way to the new highs it was destined to make later in the year.

Apple Computer (AAPL)
June 2004

This illustrates both the value and the difficulty of using Crocker charts for trading individual stocks. Now let's take another look at the Apple chart, but this time using the Goodman Price/Volume Indicator, below. A buy signal is represented by an upside breakout by the indicator (the blue line), above a previous high. The high reached on April 30, 2004 was exceeded at the end of May, and that would have been a very profitable trade if one had held on to it, since Apple advanced from 26 to 68.44 since then.

Note also that there was a change in the chart in late November. After the peak at 68.44 on November 29, the stock started consolidating. The stock action looks worse than the GPVI, which continued to make higher highs and higher lows. I'll discuss this divergence in more detail momentarily.

Apple Computer (AAPL) Goodman Price/Volume Indicator
January 2004 through December 31, 2004

As you can see above, a GPVI chart is far more readable than a Crocker chart. But that still leaves the problem that one must plot hundreds of charts and study them daily. And it leaves the additional problem of subjective interpretation.

So my next step was to transform the GPVI into an oscillator -- a chart that swings between bullish and bearish extremes. I've called that the Goodman Price/Volume Oscillator, or GPVO. Buy signals occur when the GPVO exceeds a computer-generated upper trigger level. Sell signals -- which tell you when to close out the previous buy trade -- occur when the oscillator drops below the lower trigger level. In the chart below I have labeled April 30, 2004 for comparison to the chart above. I selected that date because it marked the first buy signal after a close-out signal. I have also highlighted (pink marker) the February buy loop from the Crocker chart near the beginning of this report.

Now let's return to that divergence between the stock and the GPVI after the stock's top at 69.44 on November 29. Note in the chart below that the Oscillator produced a clear sell signal for Apple on the very day of the top -- November 29. You couldn't have gotten that signal by eye from the stock, nor from the GPVI. But the pattern recognition technology behind the GPVO nailed it to the very day. Of course I'm not saying we'll hit every turning point in every stock perfectly -- and when we do get it right, it usually won't be on the very day. But it's nice to know that this kind of accuracy is possible.

Apple Computer (AAPL) Goodman Price/Volume Oscillator
January 2004 through December 31, 2004


Fred Goodman, CFP, is a fee-only Certified Financial Planner based in Los Angeles. You can send him your questions and comments via e-mail at E-mail sent to Fred may be edited for clarity and brevity and published on this web site, and may include your name unless you request anonymity or specify not for publication. The charts and commentary represent what Fred  thinks about the market and what he is thinking of doing for his own account and for accounts he manages at the time of writing. Fred, his clients, or his family may have positions or may make trades in securities mentioned in these commentaries.  There is no guarantee that you will profit from trading as discussed herein. You may lose money and Fred assumes no responsibility for what you do or do not do with this information. Copyright 2009 Fred Goodman. All rights reserved.

For information purposes only, offered as a periodical of general circulation; not to be deemed to be recommendations for buying or selling specific securities or to constitute personalized investment advice. Derived from sources believed to be reliable, but no warranty is made as to accuracy.