elliott wave pattern

As can be seen in the above diagram, the yellow ellipses outline the four main peaks. With each passing day, NIST publishes a list that includes the first five of these major peaks, plus more. The full National Peak List (PDF) lists all known known peaks for the last half-century. We would like to explore the behavior of this Elliot Wave Pattern with respect to time. In short, what does this tell us about our data over time? First let’s look at the relationship between peak height or value and periodicity, for that we will use periods as time units. We will see that a high / low value on the graph should predict a long / short period for the entire dataset. Specifically, it would tell us whether there is any kind of correlation between these two things. A nice linear pattern would suggest there is no trend associated with anything with respect to periodicity, whereas an outlier (an example of what data science looks for: missing / anomalous values) would be highly suspicious of such a finding. To make sure we are clear on this, the next plot shows the values versus periodicity for some of those first peaks in the data set:

Elliott wave pattern

As you can see, there is a lot of data right up there that looks highly suspect. Most of this region, though, should have relatively little effect on anything: this should not be too surprising if a given measurement is only made at random instances! In any case, we are interested primarily in trends for periods less than one year, so the range down to the two-year mark is most instructive to consider here. As we look more closely at the first-peak plots for that period, the “outliers” in both periods can mostly be eliminated. This suggests a possible signal of linearity: there is not yet an obvious strong tendency towards either of the aforementioned states. Perhaps there is some sort of general pattern. But then why is that occurring again just seven months later? (Note that I am very cautious of drawing conclusions based on too few observations in this type of analysis.) To explore if it could be possible to separate and identify signals, as well as possible periods for each signal, I will consider two measures. One method would separate different signal periods that share a common starting point. By looking at this trendline through each group of separated period measurements, it can be seen whether there is a strong separation between them from a statistical standpoint.