March 18, 2026
I heard former Goldman Sachs CEO, Lloyd Blankfein, say the following on Preet Bharara’s “Stay Tuned with Preet” podcast this last weekend:
““We’ve just gotten through a market with equities on record high levels with a very, very favorable financing environment and a lot still hasn’t been sold, really? So is it going to be easier to do that if the market craters or if financial conditions get so bad that no one is lending anymore into the market? So I think the longer you wait, the longer the amount of time between such reckonings, the more difficult and painful it becomes because more, to just switch metaphors, you accumulate a lot of kindling on the floor of the forest. And at some point, there will be a spark that if the kindling wasn’t there, it wouldn’t have made a difference. But with the kindling there, it might set the forest on fire.”
The idea that stability breeds instability is not new. It dates back at least to economist Hyman Minsky (1992). However, Blankfein’s focuses on it in his market commentary reminded us that it remains relevant and important.
Where is “the amount of time between such reckonings” or “kindling” count reflected in quant finance?
I get ~10 articles on the application of quant to investing in my inbox each week. I can’t remember the last one that pertained directly to the timing of Tops and Bottoms.
True, there is much research that indirectly relate to them – hidden Markov models, regime based investing frameworks, etc.. Marcos López de Prado may have had it in mind when he proposed fractional differencing. Further, we ourselves have shown that Tops and Bottoms (as we identify them) are typically high trading volume events, and there is much academic and industry research supporting the importance of trading volume as a model input.
But there is scant, if any quant research that focuses directly on measurement of the time passed since last major Bottom (or Top). We seek to at least partially address that deficit with this post.
“Days Since Last Top” (DSLT) and “Days Since Last Bottom” (DSLB) are core VecViz inputs.
Both the VecViz Vector Model of price probability and the VecViz V-Score ranking of expected forward price performance use Days Since Last Bottom (DSLB) and Days Since Last Top (DSLT) as inputs.
Linkages VecViz has found of Top-Bottom timing to Expected and Tail Behavior
The machine learning ensemble behind the VecViz V-Score ascribes higher values to tickers with low DSLT and high DSLB, as illustrated in the heatmap below..
This may seem contrary to what Blankfein described. It isn’t. Blankfein was referring to the risk of a severe sell off, one that requires a necessary “spark”. The V-Score, in contrast, represents expected forward relative returns. Recent tops often get re-tested, as do recent bottoms.
More consistently with Blankfein’s comments, we do find that downside risk, as measured by our Vector Model of price probability 99% VaR metric (99D), is most adverse for tickers that have the high DSLB (and DSLT).
Actual forward returns have behaved fairly in line with V-Score and 99D expectations
As anticipated by our V-Scores, forward returns have been greatest for the tickers with low DSLT and high DSLB. The expectation for lower returns is aligned with regard to Top age – high DSLT tends to have lower returns, but most especially when paired with high DSLB (as opposed to low DSLB, as expected).
If our 99D_Ret estimates were perfect, they would be exceeded to the downside only 1% (0.01) of the time. In the heatmap below, we see that the expectation that high DSLB and DSLT tickers would have the worse downside returns was calibrated well – the breakage rate was quite close to 1% (0.0127). The biggest miss to the 1% (0.01) target was for tickers with “mid” aged tops and bottoms, coming in at 1.74%.
Denoted in decimal terms, 1%=0.01.
Lack of a “textbook” definition likely discourages utilization of Tops and Bottoms in quant.
Everyone knows a major top and bottom on a chart when they see it. For example, everyone can agree that March 9, 2009 was the bottom of the GFC and March 23, 2020 was the bottom of the Covid sell off.
However, beyond those dates and a couple others (ex: October 19, 1987, maybe April 9, 2025), everything else is up for argument as to whether or not it is “major”. For example, what about a sharp, significant acceleration upward (downward) from a flat, choppy trend – is that a major bottom (top)? VecViz often flags them as bottoms (tops), but sometimes does not.
Yet somehow the 52 week hi/ lo became a common industry statistic
Almost every investor dashboard on the internet has 52 week hi/lo info. Unfortunately, our data shows those stats provided investors with woefully inadequate representations of a ticker’s relevant range.
In the chart below we plot the DSLT and DSLB for every ticker in our coverage daily, over the last 4+ years. While the average term is in fact just under 1 year, the chart makes clear that much, much longer spans are quite common
In practice, a consistent, reasonably effective Top & Bottom definition is likely the key.
VecViz has a multi-pronged algorithmic approach to identifying Tops and Bottoms. We can’t say the algorithm is perfect – it sometimes takes longer than we would like for a top or bottom to algorithmically register, and in other instances it registers a top or bottom too early. But it is pretty good, as evidenced by the charts below. Appliying it consistently across training and production has been sufficient to generate fairly decent analytic results thus far.
It is difficult to overstate the importance of Tops and Bottoms to the VecViz framework
As seen in the V-Score spider chart earlier, in addition to DSLB and DSLT VecViz also considers Top and Bottom price proximity, channel angles, the distribution of channel lines above and below the current price, etc. What might not be clear from the spider chart is that the channels whose angles and line count locations are VecViz inputs are themselves anchored and defined by the Tops and Bottoms.
“Counting” is sometimes all the quant you need.
Lloyd Blankfein is not a quant, but he knows markets well and under his leadership Goldman navigated most crises, including the GFC, better than most if not all peer institutions. DSLB and DSLT are just day counts, but they quantify the concept of “kindling” Blankfein describes (or something like it), and turn it into functional inputs for forecasting both expected returns and tail risk.