By Jock Mackinlay 05 Mar, 2009

With the recent drop in the Dow Jones, I wondered about the historical trend. I created a workbook with monthly average data I found on the web. Note that my expertise is in visual analysis, not finance, and this analysis comes from that perspective.

With a semi-log plot you can clearly see a generally linear trend to the data:

In Tableau, I can fit a linear trend line to this data by writing a calculation to convert the month to log(month, 10). As you can clearly see, the partial average for this March is currently touching the trend:

You can also see that the Great Depression dipped significantly below the historical trend before the Dow returned to the trend. That is also true for other smaller movements around the historical trend line. If you took these dips to be a precedent for our current market, my visual guess for the bottom of our current correction would be next September around 3,500 for the Dow Jones:

It might be lower since we were above the trend for a long time. Converting back to the a standard trendline, we get the following view:

Of course, all of this assumes that the political and economic situation today is similar to that of the Great Depression, and that our financial system is similarly architected. All of which could be questioned.


However I would note that curve fitting is not a trend line in the sense of stock charting trend lines.
A trend line, on a stock chart, connects the lowest lows prior to the current point. As such, your trend line should be adjusted to connect the low from the Great Depression.
In which case you will note that both the semi-log and the standard chart will show there is plenty of room to fall before we reach that trend line...
I am not saying that we have to come down to the trend line at this juncture in time. Just that we are still very far above the trend line.

Isn't a linear fit of the DJIA data a rather coarse measure? It shows great swaths, decades long, in which the index deviates substantially from the fit.

If you apply a LOESS to the DJIA data, you get a closer model of the data, but you are still able to see what a great positive deviation there was in the late 1920s and what a sharp transition there was to the huge negative deviation in the early 1930s.

In the attached chart (also found at, I show the DJIA (blue data), a linear fit of the log data (green line), and a LOESS fit using alpha=0.25 (red curve).

You analysis is interesting. I wonder how well this procedure would hold up "out-of-time". One might experiment by fitting the line to data points only up to, say, 1980, then seeing how well data points from 1981 onward were predicted. Repeating this test multiple times, using different dates would give at least some idea as to how well this model works.

I realize that the point of your analysis was not to precisely predict individual values in the future, but one would expect that over an extended period of time, future data points should center on the line, if this analysis is valid.

-Will Dwinnell
Data Mining Web log

Nicely done, I found your post while searching for the raw data as I had the same idea. It's interesting to see how severe the Great Depression really was relative to our current drop.

Now I wonder what it might look like if you divide out each data point by the gold value at the time so as to convert the plot from dollars into ounces of gold. I contend that gold is the only real value since the government interferes too much in the value of the dollar. A gold plot may actually follow a linear trend line closer... not sure.

I have been interested in the very same thing you are doing for about a year now. From what I can see of your graph, it tends to be skewed by the last 15 years. I think that part is more of a bubble and should be factored down a little because it is over valued. If you do that, your slop decreases a little. I am guessing your target for great depression level would be more like 2000.

I am looking at this with an idea of changes in rules causing trends to happen. In this case, the changes in laws at the beginning of the Clinton administration would have forced a variation away from the exponential growth you have based your projection on. This would have caused the over evaluation I think you have in the later 15 years of your graph.

As an improvement and I will do this myself when I find the data, do projections based on the different exponentials growths that seem to be shown. Each one is based on changes in banking policy. I think you can see that each one gives a projection as to what value should be if allowed to continue without hiccups in the market.


These plots are just what I have been looking for. Have they been updated to the present? I'd like to see that.

Don't know why plots of things like the Dow aren't always show in logarithmic form!