A good friend shared this in one of our Whatsapp chats.
Classic #chartcrime. I don’t want to be too hard on my friend, after all, Bay Area real estate has certainly co-moved with the stock market. But this chart is intentionally heavy-handed. The axes don’t start at zero which should immediately cause you to wonder “what’s that smell?”
Look a little closer and you see the dual vertical axes are out of proportion. The blue axis goes from $400 to $1,100 while the red makes a much larger percent jump from $7,000 to $29,000.
Getting fooled by a chart is a forgivable offense. The friend who shared that chart has a grad degree in physics and extensive business and tech experience. It’s tiring and impractical to slow down at every chart we see. Fortunately, spotting chart crimes is just a matter of practice.
For financial #chartcrimes I recommend this thread by my buddy Jake who ruthlessly collects them.
Every chart you see was created by someone who was framing a story. Every chart has intent. Design choices are never accidental. This doesn’t mean every chart crime is nefarious. Often they just reveal how people have fooled themselves.
Here are 2 common failure modes:
- Spurious Correlations
These are best explained by simply looking at ridiculous examples. Call it lazy pattern-matching or uncritical data-mining. Correlation/causation errors are in our DNA. I’m convinced there’s no solution to this. And even when we think we isolate causation we are prone to being exactly wrong. A recurring theme in Moontower is we often say “because of” when we mean “in spite of”.
- Invalid Comparisons
What can cause 2 quantities to be non-comparable? One series might be “stationary” and the other “non-stationary”. These are technical words so I feel like an imposter even writing them. The ELI5 gist of a non-stationary series is one that does not have a stable mean. For example a stock price index or your age. These are quantities that trend (we can debate the stock one, but if you believe in inflation at the very least the price is subject to the trend of the denomination. All prices are relative to a denomination. I can compute the price of oil in USD but I could also compute in terms of eggs per barrel, diplomas per barrel, or Pokemon per barrel).
Examples of stationary or stable quantities would be how many hours you sleep, how many times you go to the bathroom, or returns. So it would not make sense to compare the price level of the SP500 which has a mean which changes over time with the level of the VIX which is mean-reverting.
So when making comparisons it’s important to consider what drives the measurables. If the drivers do not come from the same distribution or behavioral class you might be fooling yourself.
Just before I was set to blast out the newsletter I learned this is the “most liked” tweet of all-time. That’s informative but it does make you wonder what the most liked tweets of all time are normalized by number of active Twitter accounts. And even then you’d like to further account for Russian bot accounts. A common way to normalize a non-stationary series to a more stationary one is to normalize it with a ratio. GDP vs GDP per capita. Inventories vs Inventory/use.