RIsk Ranges
- This topic has 111 replies, 20 voices, and was last updated 2 years ago by
.
-
Incidentally, on the Macro Show this week KM gave further clues on the Risk Ranges. He mentioned the UST risk range high going higher based on the MOVE index going lower. He also mentioned the TREND signals that he uses (3 months) are a function of the VIX trend. So I guess his trend signals / levels could be derived from the 90 day (or 63 working day) trend in the VIX? It seems the Vola component is vital for the risk range AND trend.
1 user thanked author for this post.
Hi-
A lot of good points here I won’t repeat.
I scrapped the last 2 years of Risk Ranges, and every morning I update these in the minutes before the Macro Show starts.
I was surprised that price did not travel KM’s expected range most days. Price spends much of its time at the top end of the risk range, AAPL, MSFT, etc. But, they seem to work well with those assets that have lots of shadows in their candles, like Gold. You can see that you usually only get 1-2 days per month to buy near the lower end of the risk range to catch a good uptrends
The 3 regimes that KM outlines on VIX lead to much wider, normal, narrower expected move ranges as compared to traditional Gaussian-based expectation models.
So, in late Mar 2020, conventional expected $SPX moves were maybe 5% and KM would be like 15%, with VIX> 30 in the F-Bucket.
I have done some work with Hurst but its very sensitive to the time ranges you are looking at .. so, is this trending over a 15d period or 30d period.. gives very different H values, so it has been of limited value to me.
More to follow,
Risk
1 user thanked author for this post.
Thanks guys. Some great additions to the debate!
On the Macro show yesterday Keith mentioned a bit of ‘his secret’ to the risk ranges as he mentioned a NASDAQ stock (can’t recall which one) and the fact the VXN had gone DOWN which allowed him to raise the upper risk band level on X stock. I think we have sussed this from above but the key for me was that he appears to be using the stock specific Vola level (be that VXN or VIX etc) for each stock RATHER than the individual stocks’ implied volatility. After all, the VIX is effectively the IVOL for the SP500 anyway!
So maybe Hurst Component for trend? He has previously mentioned that his SPX TREND is derived from the VIX trend too. So maybe the VOLA level is what dictates the TREND (3+ months) in the Vola specific index?
1 user thanked author for this post.
Hi All,
Great to see fellow HE subs trying to back solve the RR and signal lines. I’ve been trying to do this solo for nearly three years, imagine my joy I stumbled across this thread!
My findings so far are below on both the Risk Range and the signal lines (Trade/Trend/Tail), I hope this helps the debate and understanding for everyone.
Risk Range
The Volatility parameter is the Implied Volatility for that asset scaled down to daily. Currencies and Treasuries might be a bit harder for some to get, but they do exist (mainly OTC derivatives). Remember the VIX is a weighted aggregate options of the S&P 500, not the index itself. So if you can get S&P 500 IVOL, use thatm if not then the VIX will suffice.
I have used realised volatility as a substitute when there is no options market readily available.
On the Hurst Exponent, I’m thinking this is a scaling factor of volatility, a power-law relationship. H is equivalent to 1/alpha in fractal math e.g is H = 0.5, alpha = 2. So if H<0.5 this would provide a wider range for a security which is mean-reverting and a narrow one for those that are trending strongly.
The range itself is around a dynamic moving average. I’m thinking this has to do with fractal time (“Risk happens slowly, then all at once”, the “or” when KM mentions the duration). It is discussed in the Behaviour of Markets. Haven’t worked the math out yet, my skills are not what they used to be when I was at university! I have tried Volatility & Volume Weighted Moving Averages, but it’s not really solving the issue at hand and overcomplicating the Volatility and Volume aspect of it. IMO, VWAP falls into the same boat here.
I’m not sure where the skew in the range comes from, but I know it does not come from using the Max/Min used in the Rescaled Range calculation. Tried that and it produced a mess.
On the Vol of Vol, I don’t think this is VVIX or equivalent but the evolution of volatility over time and is not involved in the calculation.
On volume, I am not sure how this works with FX and Bonds. It could be a variable, but I think it is the same as Vol of Vol, e.g. Daily volume was down X% vs the 1M average, and the range is signalling a lower high etc.
The “trend” range I’d imagine would use front-month options still as they are more frequently traded vs the quarterly options but only time will tell.
Signal Lines
This is where the 1-month, 3-month, 6-month and longer momentum chasing comes in! The goal is to find the threshold prior to when these systems will flip, it’s not too dissimilar in principle from Charlie M’s (from Nomura) CTA model.
CTA’s and Hedge Funds will often use a weighting system for their momentum systems. Watch MAN AHL’s Quant series on YouTube for more info. In theory, an implied price, P*, could be determined with averaging the momentum durations. This is my line of thought behind what KM is using, he is trying to find P*. This does work somewhat on its own (currently this what I use in the interim) and is similar to a Moving Average but it neglects the role of factors they have on an asset’s momentum 😉
In 2011 KM did provide a hint of what was in his model in the book The StockTwist Edge. KM’s chapter is called Trade, Trend, Tail (https://www.oreilly.com/library/view/the-stocktwits-edge/9781118029053/09_c02.html\)
In that book, he mentioned that there were 27 factors overlayed with the volume and volatility. I think that these 27 factors are Macro in nature e.g. US Equities, CRB Index, Global Equities, 10Y Treasury, DXY, Momentum, Volatility Factor (not volatility itself but High/Low Volatility) etc. Also IRC he did mention that variables on The Macro Show 12 to 18 months ago that are in Trend do not always appear in Trade or Tail lines, eg. Inflation & GDP could make an appearance in Tail for example.
My hypothesis is that he finds P* from individually regressing each factor against the security and does a weighted average based on Correlation^2 or abs(Correlation) for the asset in question to filter what matters to the asset.
Price, Volume and Volatility each have their own individual regressions. The four sub-models are then combined in some form of a weighted average.
It looks complicated, because well it is! When he developed this, not many people knew about factors (mid-2000’s) outside of Hedgies and academics. I inadvertently had done something similar with an FX model I developed in 2017, so this is where my line of thought comes from.
I know its a long post, thanks for reading.
Kauri
5 users thanked author for this post.
Following up on the discussion around volume as it relates to the risk range. It is likely a dummy variable in his formula that follows some type of pattern as below:
- Volume accelerating = 1 (confirmatory signal)
- Volume decelerating = -1 (dis-confirmatory signal)
- Volume unchanged = 0 (no impact)
He always refers to the range identifying the likely probability of the current days price movement.
Will post an update as I progress on the statistical side.
If anyone has additional thoughts on the above, please post.
This could work but it would need to be combined with another dummy e.g. TrendingDummy*VolumeDummy. If we get back to bearish trend, volume accelerating on down days is a good thing.
2 users thanked author for this post.
Fellow Hedgie here! I just came across this thread and have been analyzing this topic for a while. I’ve looked into the Donchian Channels and Buddy Carter models. While I think they are close, I don’t think that’s what the purpose of this thread was for. I think it’s important to state and make clear that the risk range calculation can’t directly take into account volume. ie: UST, USD/EUR, NYXBT, or any other ranges on the risk range product that are listed and have no volume. I’d like to also note that there are missing pieces here in regard to IVOL. UST or NYXBT do not have options and I am sure there are a few others on the list I am missing. The only explanation I could think is KM is taking into account Futures and or Futures Options IVOL but there are price discrepancies from the premium futures trade at to current rates. Also, a reminder that the UST Futures don’t trade in basis points of yields but value for the entire bond. I’m sure this is complicated, but I am trying to consolidate the thread to continue our backtesting. What we do know is volatility affects the risk range. We also know that he can pull risk ranges for securities without any volume or options. Which leaves us with volatility indexes as the most probable component. (VIX, OVX, GVZ, MOVE, VXN). There are a handful of risk ranges that do not have volatility indexes. I saw previously that someone mentioned KM being able to pullup a risk range on the spot which makes me believe he runs an excel sheet pulling data from the terminal to calculate upper and lower risk range bounds. Maybe it’s more weighted on how the security itself trades than all these underlying factors like the volatility of the price of the security itself from day to day. Maybe someone can clarify on anything I am missing? I have access to a terminal and will be willing to supply any additional needed data.
1 user thanked author for this post.
Guys, it has to be some variant of the Donchian Channel. Check this out. This is the closest I’ve gotten to Keith’s sauce, backtested 133 days. The blue lines are KM’s RR, the red lines are my own range, and the black is the closing price. The Donchian Channel is usually measured by the month, but we don’t have to use that scaling, especially in bear markets. You can shrink it down to a 7 or 8 day Donchian Channel. And then you can do some basic calculation of implied vol, (or a vol index if you got it), and then you can add that implied vol to the lower Donchian range, and subtract the implied vol from the higher Donchian range. This effectively creates a buffer where you set your buy and sell price 1 implied standard deviation from the donchian low/high. We won’t ever get to Keith’s exact numbers because his scaling will be arbitrary, but we can get pretty close playing with this formula.
And if you think about it, this makes a lot of sense. Although the Donchian Channel is often an indicator to signal a breakout, no one really knows if it’s going to be a breakout, and according to basic probability, it’s probably not. Most of the time it actually does stay pretty close to that range. This is pretty much the only way to replicate Keith’s upside and downside calculations.
Hopefully that all makes sense, but if it doesn’t, I’m uploading the data I used to make this graph. I also added a half-assed VIX formula that I found on Investopedia, it’ll get you numbers vaguely close to the VIX, and I imagine you can use that equation on other assets.
Same as last time, the spreadsheet is locked. Password is Keith’s low risk range number for Bitcoin (NYXBT) on August 3rd (08/03/20). Repeat: August 3rd low RR for Bitcoin.
Reminder to remember what happened to LTCM guys, don’t rely too heavily on the math.You guys stay awesome!
4 users thanked author for this post.
Find exclusive trading pro-tools on 
