Distilling Bourbon and Markets

In June 2018, Dan and I traveled with several friends to Louisville, Kentucky to purchase a barrel of Four Roses Bourbon. Throughout the trip, we toured bourbon distilleries and learned about the process of making bourbon. The grains are cooked into a mash, then yeast is added and fermentation occurs. The result is something called “distillers beer,” and it’s honestly pretty gross. Looking at it, it’s hard to imagine how the roiling, sickly yellow sludge will transform into a smooth, auburn whiskey. A key step along the way is distillation: where the alcohol is liberated from its grainy prison. Since the evaporation point of alcohol is lower than the evaporation point of water, heating the distillers beer can allow extraction of the evaporated alcohol into steam. This is then condensed into a clear liquid known as “white dog.” The final color of whiskey comes from the aging process. The white dog is placed in charred oak barrels to age over several years. As the seasons change, the barrels expand and contract, pushing and pulling the whiskey in and out of the wood and giving it a charred caramel color and taste.

Four Roses Bourbon in particular has a special place in my family history. My maternal grandfather liked the brand, and when he left to fight in the Korean War as part of the US Navy, my grandmother continued to drink it as a way to feel connected to him. They remained loyal consumers through the decades, and she was very excited when I brought her a small batch bottle from our trip, along with some new tasting glasses. Perhaps unbeknownst to my grandparents, the “Four Roses Bourbon” they were buying changed dramatically in the late 1950s. Seagram had purchased the brand in 1943, and decided in the late 1950s to sell legitimate Four Roses Bourbon only in the growing markets of Europe and Asia, and to slap the Four Roses label on inferior blended whiskey to sell in the United States. This practice lasted until 1994. High quality had (mercifully) been restored long before I fell in love with the brand in my early 30s.

For all those intervening years, my grandparents had not complained. Failing (or perhaps refusing) to separate their nostalgia from the actual taste, they had preserved a cherished experience. Experience, after all, is a strange and multi-faceted thing. What we think of as “taste” is actually a superposition of senses and time: the smell, the look, the feel of a glass of whiskey can transport us back to a moment long past and trigger a reaction not wholly of the present. Unlike the process of extracting alcohol from distillers beer, separating out the constituent parts of our human experience is messy and imprecise.

Price movements in the US equities market are more like human experience than like bourbon distillation. What happens in a given stock on a given day is a confusing mixed reflection of general market trends, sector trends, factor trends, trading behavior in closely correlated symbols, and of course, meaningful fluctuations in supply or demand for the security itself. When we seek to understand the “impact” of large trades, we would like to isolate the effects of our trading behavior in the security itself from these extraneous forces, but this is a very challenging task. Measuring slippage versus arrival (i.e. looking at how the price changes over the course of our trading relative to where it was when we entered the market) gives us a very noisy view of our impact, since it lumps together the effects of all the external forces with the effects of our own behavior. Analysts often attempt to get around this problem by measuring slippage versus VWAP instead, a metric that compares the volume-weighted average price attained for our trades to the volume-weighted average price attained by all traders in the market for the same security over the same time period. But this metric is subject to a circularity: when our trading impacts the price of the security, it impacts the price for everyone! And hence we are grading ourselves relative to a standard that we also influence. This is less noisy than slippage vs. arrival, because our benchmark has the same external forces that affected our trade prices baked into it, but it introduces a blind spot to our own influence that grows larger as our trading activity get heavier, which is when our impact matters most.

It would be ideal if the impact of our trades had some kind of distinguishing property that would allow us to separate it from the impact of wider market forces, like the lower boiling point of alcohol that makes spirit distillation possible. Though it will never be as clean as a physical process, there is some basis for hope. Wider market forces should behave a little differently than our trading impact: they should affect more symbols more evenly.

This isn’t a perfect basis for separation. Our trading behavior in one symbol may have secondary effects in correlated symbols, but we would expect the impact to be strongest in the symbol itself. In contrast, we expect the effect of broader trends to be equally (or perhaps more) visible in other symbols. This suggests an approach for removing some of the wider market effects from our metric without introducing the full circularity of VWAP: we can compare ourselves to a benchmark computed from the prices achieved in other securities during the time period we were trading.

So which other securities should we look at? One place to start is ETFs that tend to follow market/sector/factor trends. An ETF like SPY can give us a sense of overall market movement, ETFs like XLE, XLF, etc. can give us a sense of sector movement, and ETFs like VB can give us some sense of movement across factors. Naturally, the raw prices of these ETFs are not a good basis for comparison, as they may differ arbitrarily in magnitude from the typical price of the security we are trading in. But instead we can look at how much they change relatively over the price in question. For example, if the price of SPY has increased by 0.1% over the same time period, we might expect a similar related increase in the relative price of other symbols.

For each symbol, we can look at historical data to get of sense of how relative price increases/decreases in that symbol tend to correlate with relative price movements in specified ETFs. Then, if we look at a time period where we were trading in a particular symbol, we can use this correlation information plus the relative price movements in those ETFs over the same time period to “guess” what the relative price movement in the symbol would have been without our trading activity. We can then subtract this from the actual price movement we observed, and what remains is our estimate of our own impact. However crudely, we have thus attempted to distill our own impact from the distracting stew of market activity.

This is not a particularly original idea or endeavor. We suspect that many people have developed much more sophisticated approaches to this, and we look forward to learning from them as well as continuing to innovate on our own approach. As a result, we expect that we can reduce the noise in our distilled impact measurements much more drastically over time. But in true Proof-y fashion, we wanted to share this research with you as it happens from the beginning, and we have chronicled our progress so far in detail in the whitepaper you can find here. It’s kind of a long read. Best to digest it alongside a good glass of bourbon.