Quantum of Vol

My goal is to have my own retail Medallion fund. Sometimes I share my opinion - not advising to use it though as it can be addictive

[Quantum of Vol @AlgoSignalVol]

A few words about the model I’ve mentioned several times recently.

Over the past year I’ve been building a model with one very specific objective:
identify stocks likely to drop −10% within the next 5 trading days.

No fundamentals. No news. No alternative data.
Every feature is engineered purely from OHLCV — structure, momentum, volatility, volume behavior, and regime context.
• 2020–2024: training data (>1000 tickers, XGBoost classification)
• H1 2025: strict walk-forward testing + parameter tuning
• Since September 2025: fully out-of-sample live forward run

The recent forward results are holding up — which is what ultimately matters.
The goal is to systematically detect downside convexity before it becomes obvious.

Frankly, the test results were already very strong, but I was concerned about potential overfitting due to extensive tuning during the validation phase. Since September’s fully out-of-sample run, the performance suggests the signal is real and reasonably stable.
I plan to publish signals for some time. If that’s of interest — stay tuned.
Also open to discussions around machine learning applied to market data.

[Quantum of Vol @AlgoSignalVol]

@OptionsDepth That was real sourcery, took 7555 as a lotto - not sure how SPX closed though, it was quite erratic last seconds

[Quantum of Vol @AlgoSignalVol]

@HedgieMarkets It is a structural problem when artificial "thinking" is multiple orders less energy efficient than human "thinking". Either core LLM architecture has to be optimized, or chips energy consumption

[Quantum of Vol @AlgoSignalVol]

@predict_addict Come on, for high school - it is not basic algebra. Any real life practical application?

[Quantum of Vol @AlgoSignalVol]

@raptor7xx PS. there was likely also leveraged ETF rebalancing flow - the same direction. Andy Constan explained it in details in his posts

[Quantum of Vol @AlgoSignalVol]

Just my hypothesis — we had a massive gamma-driven run, which likely means there was heavy long call positioning across the board. By expiration, a large amount of calls finished ITM.

Take a 7400 SPX call as an example. If SPX is trading at 7430 into expiration, that call settles worth 30 points ($3,000). A market maker who was fully hedged would be long roughly 100 SPX equivalent deltas against it. After settlement, they’re left holding inventory they no longer need.

So what do you do if the inventory is huge? You start selling it into the close.

That creates three advantages:

Push the underlying lower, reducing the payout — potentially to zero if SPX drops back below 7400.
Exit inventory at higher prices before everyone else starts unwinding.
Flatten exposure and avoid carrying unnecessary directional risk overnight.

Probably similar mechanics for single names - however less likely to sell underlying as they need it for exercise

[Quantum of Vol @AlgoSignalVol]

The real black swan event: Buffett finally catches FOMO 😭

[Quantum of Vol @AlgoSignalVol]

@Mr_Derivatives The Korean stock market = semis

[Quantum of Vol @AlgoSignalVol]

@tommbotrades @SystematicPeter So if I got correctly the expectancy is +$15-30 per trade (0.3-0.6% of locked buying power), which is not bad if trades are done frequently. Thank you, will take a look at Tammy's videos

[Quantum of Vol @AlgoSignalVol]

@tommbotrades @SystematicPeter What return on deployed capital it gives, if ok to ask?

[Quantum of Vol @AlgoSignalVol]

I have probably similar approach, based on ML signal, with couple of dozens trades per year - x8 of original equity - OoS backtest. The main reason that prevents going live - equity curve is too bumpy. The other reason- no obvious exit strategy - xN of the entry premium cuts some real bangers, and there is no optimal sweet spot which would give stable result

[Adam Mancini @AdamMancini4]

Runners paying from our Wednesday's 7137 Failed Breakdown long at 420pm. As posted, next set of targets were 7232, 7238, 7248, 7264. 7264 hit exact

Plan today: 0 to do until deep #ES_F sell. 7248=micro support (watch traps). 7265, 7276, 7287 next slate. 48 fails, dip 7233, 7219

[Quantum of Vol @AlgoSignalVol]

@Math_files and he is right!

[Quantum of Vol @AlgoSignalVol]

@Ozard_OfWiz your dataset has Y2K problem unless you are too young to know what I mean :)

[Quantum of Vol @AlgoSignalVol]

@DeepDishEnjoyer Being a kid city boy visiting grandparents for the summer school break, I could collect 0.5 liter jar of those in western Ukraine forests. Eaten smashed by fork with a little sugar and soar creme

[Quantum of Vol @AlgoSignalVol]

@MichaelPBento @VolSignals Wouldn't the bid from closeout come closer to expiry not to leave MMs exposed? This is in their best interest to bit power hour and reduce JPM put payout

[Quantum of Vol @AlgoSignalVol]

@VolSignals @NickSchmidt 2024-2025: 1q 24 - flat day (miss), 2q 24 - down (hit), 3q 24 - up (hit), 4q 24 - down (hit), 1q 25 - up (hit), 2q 25 - flat (miss), 3q 25 - up (miss), 4q 2025 - no signal. Overall 4 out of 7 hits

[Quantum of Vol @AlgoSignalVol]

@Pedalo1979 @VolSignals One was using chatGPT, the other one Claude

[Quantum of Vol @AlgoSignalVol]

@EdMayak @VolSignals part will apparently offset delta for the next round. But when are they buying back the rest is the $1m question