Statistical procedures such as regressions and correlations are powerful tools for data analysis that are very effective in many fields of study. However, I believe that analyzing price data with statistical methods is not the optimal way to make investment decisions, for a variety of reasons.
Financial market data has certain unique characteristics, because markets are influenced by factors such as crowd psychology and the behavior of institutional investors:
- Investors sometimes behave with a herd mentality, meaning that prices tend to swing too far in both directions. At times the stock market can become very overbought (as in January 2018) or very oversold (as in December 2018).
- In many cases, investors who bought a stock at (for example) $100, only to see it drop to $70, decide that they will sell if it ever gets back to $100, in order to break even. Similarly, many investors who missed a big move up in a stock decide that they will buy it on the next pullback.
- This type of psychological behavior creates support and resistance levels. Certain price levels can become significant, and breaks through them can be meaningful.
- When a stock breaks out to a new high, it does not face any resistance, meaning it may be able to advance more easily than a stock that is below its old high.
- Institutions often buy stocks at key support levels such as the 50-day moving average.
- Markets tend to move in trends, in which advances are followed by pullbacks, and declines are followed by rallies.
- Trading volume can be meaningful, and price moves on big volume can be more significant than moves on light volume.
- On any trading day (or other time period such as a week, an hour, etc.) there are four significant prices: the Open, Close, High and Low. The relationships among these four prices can be significant, and can provide insights into how the balance between the buyers and sellers is playing out. These relationships are best seen in candlestick charts. In some cases, candlestick patterns can have meaningful investment implications.
- When candlestick reversal patterns occur at support or resistance levels, they can indicate market turning points. For example, in May 2008 the S&P 500 weekly chart formed a bearish engulfing pattern after rising to the resistance level of its 200-day moving average. This was a noteworthy bearish development (please see the article “Long-term Market Analysis” for more detail). An example of a bullish situation occurred in February 2016, when the S&P 500 daily chart showed a bullish candlestick pattern at the support level of its prior low (please see the article “Short-Term Market Analysis” for more detail.)
Due to these and other related phenomena, the market is not a pure math problem, and the best way to analyze it is not with statistical methods.
… Price data moves randomly, but this does not prove that analyzing charts is worthless. There are many ways of analyzing price and volume activity that add value to the investment process, including identifying overbought or oversold conditions, breakouts to new highs, etc.
This information can be seen in charts, but is not apparent in statistical calculations. Of course, due to the market’s inherent volatility and riskiness, studying charts is not perfect, but there is no perfect tool for analyzing the market. Correlations are not perfect either, as will be discussed below.
Analyzing Market Relationships
My research shows that much of the value from using technical analysis comes not from looking at any one chart by itself, but from studying the relationships among multiple charts. Information that has not been arbitraged away often can be found in market relationships, such as how a stock or industry group is moving relative to the S&P 500.
Figure 6 Electronic Gaming industry group index vs. S&P 500 (daily) March – April 2017
With a good software package, several charts can be aligned vertically (as shown in Figure 6), and in each chart we can add trendlines and other indicators.
For the purpose of making investment decisions, viewing market relationships with this type of analysis can be more valuable than running statistical calculations.
In late April 2017, electronic gaming stock Take Two Interactive (TTWO) broke out to a new high, and subsequently outperformed the market by a wide margin, with a gain of 75% over the remainder of 2017, compared to 12% for the S&P 500. (This case study is discussed in more detail in the article entitled “Managing Data Effectively,” as well as in one of my longer white papers. Please contact me by e-mail for a copy of this report.) In addition to its good fundamentals and good chart pattern, further evidence for buying TTWO in April 2017 was seen in the relationship between the electronic gaming group and the S&P 500.
As seen in the chart above, this industry group had good relative strength during the market pullback of March – April 2017, and then broke out to a new high just as the pullback ended. These were very bullish developments for the group, further supporting the case for buying TTWO.
In this situation, there were several specific events which showed that buying TTWO was a good idea in late April 2017:
1) the S&P 500 broke up through a downward-sloping trendline
2) the industry group index broke out to a new high
3) the stock broke out to a new high
However, such developments are not apparent in statistical calculations, such as a regression or correlation based on data for the past 12 months. The specific price moves that lead to valuable investment conclusions will be buried among all the other data from prior weeks and months that go into the calculation.
Statistical calculations do a good job of summarizing the relationship between two assets over a given time period such as the past 12 months. However, they do not consider current market conditions, such as whether the market is overbought, oversold, in a correction, etc., and they do not consider specific developments that may have investment significance.
In general, by reviewing numerous charts and indicators, in conjunction with the fundamentals, we can see how the whole mosaic of market activity is unfolding. This process can lead to valuable investment conclusions. (Examples are shown in other articles, such as “Investment Strategy Case Study.”)
Correlations Do Not Always Lead to Good Investment Conclusions
Correlations are frequently used to analyze the relationships among various assets. However, just as charts do not always have good predictive ability, correlations do not always work as expected. The following comments are from one of my white papers:
… The following charts provide a historical example of why correlations are not the optimal way to analyze asset allocation. Correlations frequently change over time, in some cases dramatically.
As an example, if we were using correlations to build a diversified portfolio in mid-2008, we might look at the correlation between oil prices and stocks, and see that these two asset classes had a negative correlation. As seen in Figure 35 below, oil prices rose sharply from mid-2007 to mid-2008, while stocks were in a downtrend as of mid-2008. In other words, oil was going up while stocks were going down – the two asset classes had a negative correlation.
Figure 35 – As of mid-2008, Oil prices and Stocks had a Negative Correlation
Therefore, if we were using correlations to build a client portfolio in mid-2008, we might conclude that in order to build a diversified portfolio, with low risk, we should buy both oil and stocks. The negative correlation tells us that they move in opposite directions. If one asset goes up, the other goes down. Therefore we should own both oil and stocks, and this would reduce the risk of the portfolio.
However, over the next six months, both oil and stocks plunged. Oil prices dropped sharply in late 2008, falling from $146 a barrel to $40. At the same time, stocks plunged in a market crash of historic proportions.
The two asset classes both dropped sharply, and the correlation suddenly turned from negative to positive (see charts below). Owning these two asset classes in this time period did not provide any reduction in risk, in fact it created a highly risky portfolio. In this situation, studying correlations to determine asset allocation did not work at all, and led to the wrong conclusion.
In fact, if we were building a client portfolio in this time period, the correct investment decision was not to buy both oil and stocks. It was to sell both oil and stocks, and move into cash.
Figure 36 – In late 2008, both Oil and Stocks plunged together – their Correlation turned Positive
Figure 37 shows how the correlation between oil and stocks suddenly changed dramatically when both asset classes plunged in late 2008.
So how could we know that we should be reducing exposure to these two asset classes before they plunged? Oil had become extremely overbought by mid-2008. It had reached its upper trend channel, and then in July 2008 it formed a bearish candlestick sell signal (see chart below).
This situation is an example of “offensive selling,” or selling into strength after an asset has made a dramatic move up and has become extremely overbought. In such a situation, its gains are unsustainable, and the prudent course of action would be to reduce exposure, or at least place stop loss orders to lock in gains if the asset starts to drop.
Note that in order to reduce exposure to oil when it was near $140 a barrel, we did not need to predict that oil prices would drop to $40. We simply needed to realize that oil prices had made an abnormal, unsustainable move up to an extremely overbought level, so reducing exposure would be the prudent course of action.
As for stocks, market conditions had clearly deteriorated by mid-2008, calling for a reduction in equity exposure. Please see the article “Long-term Market Analysis” for a more detailed discussion.
Figure 38 Oil prices had become very overbought by mid-2008
