[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.] #  @systematicls sysls sysls posts on X about if you, investment, topics, science the most. They currently have XXXXXX followers and XX posts still getting attention that total XXXXX engagements in the last XX hours. ### Engagements: XXXXX [#](/creator/twitter::1779775861589504000/interactions)  - X Week XXXXXXX -XX% - X Month XXXXXXXXX +213% - X Months XXXXXXXXX +816% - X Year XXXXXXXXX +37% ### Mentions: XX [#](/creator/twitter::1779775861589504000/posts_active)  - X Week XX -XX% - X Month XXX +91% - X Months XXX +1,821% - X Year XXX +36% ### Followers: XXXXXX [#](/creator/twitter::1779775861589504000/followers)  - X Week XXXXXX +13% - X Month XXXXXX +34% - X Months XXXXXX +64% - X Year XXXXXX +93% ### CreatorRank: XXXXXXX [#](/creator/twitter::1779775861589504000/influencer_rank)  ### Social Influence **Social category influence** [finance](/list/finance) XXXXX% [cryptocurrencies](/list/cryptocurrencies) XXXX% [technology brands](/list/technology-brands) XXXX% [social networks](/list/social-networks) XXXX% [stocks](/list/stocks) XXXX% **Social topic influence** [if you](/topic/if-you) 7.35%, [investment](/topic/investment) #3663, [topics](/topic/topics) 1.47%, [science](/topic/science) 1.47%, [the good](/topic/the-good) 1.47%, [tiktok](/topic/tiktok) 1.47%, [carry](/topic/carry) 1.47%, [crypto](/topic/crypto) 1.47%, [relationships](/topic/relationships) 1.47%, [derivatives](/topic/derivatives) XXXX% **Top accounts mentioned or mentioned by** [@humphilomath](/creator/undefined) [@maruushae](/creator/undefined) [@scottph77711570](/creator/undefined) [@friendscallmeap](/creator/undefined) [@majinboson](/creator/undefined) [@macrocephalopod](/creator/undefined) [@soma_as_moon7](/creator/undefined) [@zeonlygui](/creator/undefined) [@0xfdf](/creator/undefined) [@gerardsoreaux](/creator/undefined) [@quantonisland](/creator/undefined) [@richardcraib](/creator/undefined) [@capjsparrrow](/creator/undefined) [@lurkamat](/creator/undefined) [@paperswbacktest](/creator/undefined) [@trad62866](/creator/undefined) [@paleologos](/creator/undefined) [@imotw2](/creator/undefined) [@moreproteinbars](/creator/undefined) [@therobotjames](/creator/undefined) **Top assets mentioned** [XBANKING (XB)](/topic/$xb) [Shopify Inc (SHOP)](/topic/$shop) ### Top Social Posts Top posts by engagements in the last XX hours "I have a story for you guys. This is a TRUE story with some minor touch ups to anonymize our characters. I once employed a man who had all kinds of degrees and accolades to his name. Let's call our man. Valerie. --- Valerie was quite literally in possession of multiple masters all from top universities in very technical and challenging topics (e.g. physics computer science etc). AND he had a PhD Of course he also had the CFA AND the FRM. Very impressive right You'd ask Valerie why he would bring himself to spend XX years studying to get all of these certificates and he would tell you" [X Link](https://x.com/systematicls/status/1994785893669113870) 2025-11-29T15:09Z 20.1K followers, 93.9K engagements "A quants take on productivity proxies in managing global teams. Lets say you manage a large team that spans geographies. You can no longer use the good ol come to office and let me see how often you are on your phone browsing tiktok. The time where you evaluate who gets shitcanned has cometh. How do you evaluate who is a useful team member and who should get the boot Lets say you suffer from the same brand of autism and neuroticism as me. --- a) You have Slack messages an entire year's worth of archives. b) All transcripts of all meetings. c) Direct PnL attribution of researchers and Jira" [X Link](https://x.com/systematicls/status/1995507240414163447) 2025-12-01T14:56Z 20.1K followers, 12K engagements "You cannot use classifiers as a replacement for returns prediction if you care about magnitude. If you regress *returns* on features youre estimating: ER X If you run logistic on *sign(R)* youre estimating: P(R X X) Even if you output probabilities and both models spit out a continuous score they are not solving the same problem and are not representing the same information. --- Key differences: 1) Target a) Linear: sees every bp of the move. b) Logistic-on-sign: only sees green vs red. +1 bp and +5% are identical. Same for X bp and 5%. 2) Estimand a) Linear OLS: best L2 estimate of the" [X Link](https://x.com/systematicls/status/1995740105009709237) 2025-12-02T06:21Z 20.1K followers, 45.4K engagements "I really like decision trees and so should you Geometrically: - Linear regression is essentially one global hyperplane; it's useful at small number of features but once you have a large number of features you should think about more complex models to get a better fit. - Trees keeps asking yes/no questions and are essentially carving the space into boxes. See the photo that I "borrowed" from Cornell. Each leaf in a tree is X box in feature space with its own prediction. IF you think about this this means that in X dimensions the best linear regression can do is one slanted line whilst a tree" [X Link](https://x.com/systematicls/status/1995391608066634000) 2025-12-01T07:16Z 20.1K followers, 33.5K engagements "If you are a researcher and are relegated to the alpha mines in a fairly modern team most of your research assumes no transaction costs. You live in this fairytale world where trading is free and daily (vwap-close/(close+epsilon)) is a X sharpe alpha in US equities. Everyday you churn out banger after banger "2 sharpe" "3 sharpe" and you ignore the turnover number on your sim. You are after all paid roughly as a function of live OS sharpe before costs. The turnover number is just a distraction right --- Meanwhile the PM tasked with trading your signals berates you everyday. "Dave you freaking" [X Link](https://x.com/systematicls/status/1996578896842531016) 2025-12-04T13:54Z 20.1K followers, 14K engagements "Everyone knows funding rate carry works in crypto. It's the easy sell to get you to "try systematic strategies". Here's what most people miss about WHY it works: The funding rate is a signal. It tells you there's an imbalance between longs and shorts in the perpetual market. When funding is positive it means there's excess demand for longs - the market is crowded on one side. Traditional futures solve this with expiration. Price converges to spot as the contract matures. No argument. Perpetuals don't expire. So how do you anchor the perp price to spot Answer: You pay someone to take the other" [X Link](https://x.com/systematicls/status/1992072241589457328) 2025-11-22T03:26Z 20.1K followers, 100.1K engagements "Want to understand how the best models in Numerai crush the competition every era 🧵 Cue: Feature Exposure Feature Exposure is a diagnostic metric measuring how much your model depends on specific features. High exposure creates regime-dependent performance and should generally be reduced. Feature Neutralization is the technique for reducing exposure by removing linear feature relationships from predictions. It: 1) Uses pseudoinverse projection to residualize predictions 2) Preserves higher-order interactions while removing direct effects 3) Can be tuned via proportion parameter and feature" [X Link](https://x.com/systematicls/status/1991498263963582674) 2025-11-20T13:26Z 20.1K followers, 40.5K engagements "You can pick a fight with a man. He can be muscled scarred and tattooed. You might go for it even if he has buddies around. But for the love of god if you see THIS stay away. It is the mark of a demon and you are walking into his lair" [X Link](https://x.com/systematicls/status/1995191750214984005) 2025-11-30T18:02Z 20.1K followers, 151K engagements "Everyone's waiting for certainty before they make the jump. Newsflash: Certainty doesn't exist in any system complex enough to be worth your time. Kelly figured this out in 1956. You don't need to know the outcome. You need: 1) Positive expected value 2) Position sizing that survives being wrong 3) Enough iterations to let compounding do its thing In the short term we regret our failures. In the long term we regret what we never tried. Bezos put it simply: "100% chance of regret if I didn't try. X% chance if I tried and failed." My take on life: - You will never have perfect information -" [X Link](https://x.com/systematicls/status/1995834362815001020) 2025-12-02T12:36Z 20.1K followers, 17.1K engagements "Alpha drop: Numerai forums. Numerai obfuscates all features before giving you data. feature_1. feature_1050 zero semantic meaning. Can't lean on fundamentals or papers forced to discover pure statistical relationships. The forum is people trying to crack why things work without being able to anchor to conventional wisdom. Forced creativity at scale. Some threads: why recent 1yr models beat long history why neutralization backfires handling regime shifts when you can't see what actually changed. Good hunting ground if you want to see statistical thinking divorced from "but Fama-French says" [X Link](https://x.com/systematicls/status/1990818291896725987) 2025-11-18T16:24Z 20K followers, 15.1K engagements "Leveraged ETFs Mechanism: Leveraged ETFs use derivatives (total return swaps futures) with institutional counterparties to achieve 2x or 3x the daily return of an underlying index. They don't hold underlying securities directly. Daily rebalancing: At each day's close they must rebalance derivative positions to maintain target leverage. Example: 2x ETF with $100M at $200M exposure gains X% - now $104M must adjust to $208M exposure. Volatility decay/path dependency: Due to compounding of daily returns multi-day performance diverges significantly from the stated multiple. Choppy markets = decay:" [X Link](https://x.com/systematicls/status/1991167316101656700) 2025-11-19T15:30Z 20K followers, 6863 engagements "Reading the very interesting MS report on Moats and sharing some learnings. I'm not a fundamental investor so take this with a pinch of salt I couldn't tell you GOOG's revenue numbers at gunpoint. But I am broadly fascinated by businesses of all nature and so. --- What is a "moat" and how to actually think and measure about one BUFFET ON MOATS Buffet seems to have popularized the idea of a "moat" from his shareholder letters and meetings: "What we're trying to do is we're trying to find a business with a wide and long-lasting moat around it surround -- protecting a terrific economic castle" [X Link](https://x.com/systematicls/status/1994397394625310906) 2025-11-28T13:26Z 20K followers, 22.8K engagements "But my point exactly is this "have very orthogonal alphas" - I would bet a very significant amount that those alphas are in fact not "very orthogonal". And I say this not in the spirit of throwing shade. I say this factually as someone who has heard a great many "orthogonal alphas"" [X Link](https://x.com/systematicls/status/1996819607618007194) 2025-12-05T05:51Z 20K followers, XXX engagements "@MovieTimeDev Someone should backtest this. X. Grab all events across history. X. PIT XX hours from expiry buy NO tokens X. Compute P_outcome (1 if No X if Yes) - P_buy_no X. Horizontal average for day (this is ur daily returns) X. Do cumulative sum X. Plot the time-series This is the cumulative PnL curve" [X Link](https://x.com/systematicls/status/1981907441282609385) 2025-10-25T02:15Z 20K followers, 4697 engagements "People complaining that "Startup X shouldn't be worth $XB because it's just a wrapper around GPT" fundamentally misunderstand how value is created. Everything is a wrapper. Some of these are trillion dollar wrappers: AWS: Wraps data centers into an API. $500B. Stripe: Wraps banks and payment networks into X lines of code. $70B. Shopify: Wraps web hosting and payments into "start a store". $100B. Salesforce: Wraps databases into "track your customers". $250B. Google: Wraps the internet into a search box. $2T. Nobody uses raw compute raw infrastructure or raw APIs. They use wrappers that" [X Link](https://x.com/systematicls/status/1990794794163929242) 2025-11-18T14:50Z 20K followers, 3931 engagements "I was a very early Numerai user back in 2016-ish when I was already a working quant. Was only active for a short stint but was so tickled by the idea that a firm was trying to crowdsource signals. Numerai had a certain appeal that Quantopian never quite had for me. I really appreciated that you needed to stake NMR to get allocation and that payback was proportional to your stake. This was elegant - if you don't enforce users to have skin in the game you get adversely selected. Incentives and punishment drives behaviour. Over the years I've been involved in a few crowdsourcing programs across" [X Link](https://x.com/systematicls/status/1991378966759068158) 2025-11-20T05:31Z 20K followers, 19.4K engagements "@quantonisland Very few roles will let you work on different sides of things. Normally you hone your skills in alpha generation and learn portfolio opt via osmosis or as a sub-pm" [X Link](https://x.com/systematicls/status/1991511761972068407) 2025-11-20T14:19Z 20K followers, 1463 engagements "@ScottPh77711570 @MartinShkreli I am an impartial juror only assessing alpha generation abilities" [X Link](https://x.com/systematicls/status/1993609491934527618) 2025-11-26T09:15Z 20K followers, XXX engagements "I really enjoyed reading this if you're an agentic nerd like I am. Here's a neat summary Everyone's building AI agents now. The problem most will face is that AI agents tend to still fail on anything longer than a single context window. Here's what Anthropic learned building harnesses for long-running agents: The core insight: State doesn't belong in the prompt. It belongs in files. Your agent harness needs three things: X. Structured task tracking (JSON not Markdown) X. Unstructured progress notes (plain text) X. Git for checkpoints Why JSON over Markdown Structured data survives context" [X Link](https://x.com/systematicls/status/1993896450644664644) 2025-11-27T04:15Z 20K followers, 17.3K engagements "The line between alpha and risk premia is defined by your risk manager (in consultation with Barra)" [X Link](https://x.com/systematicls/status/1993972921258397935) 2025-11-27T09:19Z 20K followers, 10.6K engagements "I agree with kens conclusions. My indicators are telling me the same things. If you cant see it youre NGMI" [X Link](https://x.com/systematicls/status/1994841183181516928) 2025-11-29T18:49Z 20K followers, 8041 engagements "requested by @0xfdf" [X Link](https://x.com/systematicls/status/1996112435972112759) 2025-12-03T07:01Z 20K followers, 4400 engagements "@richardcraib I think the answer to this can be solved by invoking occam's razor. They are institutional-grade full-timers dedicated to Numerai and are a notch above the rest of the hobbyists" [X Link](https://x.com/systematicls/status/1997364807805313403) 2025-12-06T17:57Z 20K followers, XXX engagements "@richardcraib You might find a handful of them per 1000 active users but it's probably not at a concentration that Numerai strives for to scale Signals" [X Link](https://x.com/systematicls/status/1997365151071424625) 2025-12-06T17:58Z 20K followers, XXX engagements "@CapJSparrrow sending you an mri of my brain" [X Link](https://x.com/systematicls/status/1997664061203300775) 2025-12-07T13:46Z 20K followers, XXX engagements "High frequency bozos will tell you HFT is the only way to make money because predictions decay exponentially with time. Mid frequency clowns will tell you that MFT is the ONLY way to make money because trading too fast will erode all transaction costs. Low frequency jesters will tell you that LFT is SUPREMELY better because it allows you to manage BILLIONS rather than millions. --- The point is for every way of making money there is a subset of people that are diametrically opposed to your beliefs. Fuck them and do your thing. Sincerely Guy who has seen strategies across broad spectrums all" [X Link](https://x.com/systematicls/status/1994276332428251395) 2025-11-28T05:25Z 20.1K followers, 113.7K engagements "Common misconception: "My LR model predicts LARGE (e.g. +8%) returns. It must be confident." No. Prediction magnitude tells you almost nothing about confidence. Often the opposite. --- Here's why: X. What determines Your prediction is = X. Large predictions mean: a. Extreme features (far from training mean) b. Large coefficients c. Both None of these imply confidence. --- X. What actually determines confidence Prediction variance: Var() = x'(X'X)x That x'(X'X)x term is called leverage. It measures how far your feature vector sits from the center of your training data. Extreme features - high" [X Link](https://x.com/systematicls/status/1994622645938364521) 2025-11-29T04:21Z 20.1K followers, 88.4K engagements "Mixed-effects models are what you use once you admit your data is lying about the i.i.d. assumption. Quick thread on what they are when to use them and how they differ from linear regression: 1) The usual toy: y = X + Every row is independent same variance no structure. It's less useful once you have repeated measures panel data or anything hierarchical. 2) Mixed-effects model: y = X + Zb + = fixed effects (population-level coefficients) b = random effects (group-level deviations) Z = design matrix that ties observations to groups Now observations that share the same b are correlated by" [X Link](https://x.com/systematicls/status/1996112375162777640) 2025-12-03T07:00Z 20.1K followers, 28.5K engagements "According to @macrocephalopod if I dont get people to retweet to follow me I am going to get cooked on my bet that I will get to 30K followers by year end. I made this bet X weeks ago and my plan was to post frequently for X weeks and then beg for my highbie followers to lend a brother a hand to teach my counterparty not to make bets where reflexivity is involved and I have a chance to shape the outcome" [X Link](https://x.com/systematicls/status/1996170300095111496) 2025-12-03T10:51Z 20.1K followers, 158.4K engagements "Lets say you are a quant researcher and you want to make the leap to being a systematic PM. Let me tell you that the name of the game is to reduce the interviewers perception of how many unknowns they have to deal with. --- If you were an experienced PM the expectation is that you understand every stage in the investment process and have a track record. But for first-time PMs the bar is different. And you should absolutely take advantage of that. Chances are for first-time PMs you will join a team as a sub-PM which means you are essentially at an experience level where the main PM thinks you" [X Link](https://x.com/systematicls/status/1996799798956147055) 2025-12-05T04:32Z 20.1K followers, 56.1K engagements "You get some raw SysLS (and not ChatGPT SLOP) thoughts and analysis for free and I get a retweet like and a follow FAIR" [X Link](https://x.com/systematicls/status/1997247280559395070) 2025-12-06T10:10Z 20.1K followers, 1813 engagements "Anybody who retweets and contributes to getting me above the 20K line will get +20% sharpe in 2026. No exceptions" [X Link](https://x.com/systematicls/status/1997746380740456716) 2025-12-07T19:13Z 20.1K followers, 14.2K engagements
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
@systematicls sysls
sysls posts on X about if you, investment, topics, science the most. They currently have XXXXXX followers and XX posts still getting attention that total XXXXX engagements in the last XX hours.
Engagements: XXXXX #
- X Week XXXXXXX -XX%
- X Month XXXXXXXXX +213%
- X Months XXXXXXXXX +816%
- X Year XXXXXXXXX +37%
Mentions: XX #
- X Week XX -XX%
- X Month XXX +91%
- X Months XXX +1,821%
- X Year XXX +36%
Followers: XXXXXX #
- X Week XXXXXX +13%
- X Month XXXXXX +34%
- X Months XXXXXX +64%
- X Year XXXXXX +93%
CreatorRank: XXXXXXX #
Social Influence
Social category influence finance XXXXX% cryptocurrencies XXXX% technology brands XXXX% social networks XXXX% stocks XXXX%
Social topic influence if you 7.35%, investment #3663, topics 1.47%, science 1.47%, the good 1.47%, tiktok 1.47%, carry 1.47%, crypto 1.47%, relationships 1.47%, derivatives XXXX%
Top accounts mentioned or mentioned by @humphilomath @maruushae @scottph77711570 @friendscallmeap @majinboson @macrocephalopod @soma_as_moon7 @zeonlygui @0xfdf @gerardsoreaux @quantonisland @richardcraib @capjsparrrow @lurkamat @paperswbacktest @trad62866 @paleologos @imotw2 @moreproteinbars @therobotjames
Top assets mentioned XBANKING (XB) Shopify Inc (SHOP)
Top Social Posts
Top posts by engagements in the last XX hours
"I have a story for you guys. This is a TRUE story with some minor touch ups to anonymize our characters. I once employed a man who had all kinds of degrees and accolades to his name. Let's call our man. Valerie. --- Valerie was quite literally in possession of multiple masters all from top universities in very technical and challenging topics (e.g. physics computer science etc). AND he had a PhD Of course he also had the CFA AND the FRM. Very impressive right You'd ask Valerie why he would bring himself to spend XX years studying to get all of these certificates and he would tell you"
X Link 2025-11-29T15:09Z 20.1K followers, 93.9K engagements
"A quants take on productivity proxies in managing global teams. Lets say you manage a large team that spans geographies. You can no longer use the good ol come to office and let me see how often you are on your phone browsing tiktok. The time where you evaluate who gets shitcanned has cometh. How do you evaluate who is a useful team member and who should get the boot Lets say you suffer from the same brand of autism and neuroticism as me. --- a) You have Slack messages an entire year's worth of archives. b) All transcripts of all meetings. c) Direct PnL attribution of researchers and Jira"
X Link 2025-12-01T14:56Z 20.1K followers, 12K engagements
"You cannot use classifiers as a replacement for returns prediction if you care about magnitude. If you regress returns on features youre estimating: ER X If you run logistic on sign(R) youre estimating: P(R X X) Even if you output probabilities and both models spit out a continuous score they are not solving the same problem and are not representing the same information. --- Key differences: 1) Target a) Linear: sees every bp of the move. b) Logistic-on-sign: only sees green vs red. +1 bp and +5% are identical. Same for X bp and 5%. 2) Estimand a) Linear OLS: best L2 estimate of the"
X Link 2025-12-02T06:21Z 20.1K followers, 45.4K engagements
"I really like decision trees and so should you Geometrically: - Linear regression is essentially one global hyperplane; it's useful at small number of features but once you have a large number of features you should think about more complex models to get a better fit. - Trees keeps asking yes/no questions and are essentially carving the space into boxes. See the photo that I "borrowed" from Cornell. Each leaf in a tree is X box in feature space with its own prediction. IF you think about this this means that in X dimensions the best linear regression can do is one slanted line whilst a tree"
X Link 2025-12-01T07:16Z 20.1K followers, 33.5K engagements
"If you are a researcher and are relegated to the alpha mines in a fairly modern team most of your research assumes no transaction costs. You live in this fairytale world where trading is free and daily (vwap-close/(close+epsilon)) is a X sharpe alpha in US equities. Everyday you churn out banger after banger "2 sharpe" "3 sharpe" and you ignore the turnover number on your sim. You are after all paid roughly as a function of live OS sharpe before costs. The turnover number is just a distraction right --- Meanwhile the PM tasked with trading your signals berates you everyday. "Dave you freaking"
X Link 2025-12-04T13:54Z 20.1K followers, 14K engagements
"Everyone knows funding rate carry works in crypto. It's the easy sell to get you to "try systematic strategies". Here's what most people miss about WHY it works: The funding rate is a signal. It tells you there's an imbalance between longs and shorts in the perpetual market. When funding is positive it means there's excess demand for longs - the market is crowded on one side. Traditional futures solve this with expiration. Price converges to spot as the contract matures. No argument. Perpetuals don't expire. So how do you anchor the perp price to spot Answer: You pay someone to take the other"
X Link 2025-11-22T03:26Z 20.1K followers, 100.1K engagements
"Want to understand how the best models in Numerai crush the competition every era 🧵 Cue: Feature Exposure Feature Exposure is a diagnostic metric measuring how much your model depends on specific features. High exposure creates regime-dependent performance and should generally be reduced. Feature Neutralization is the technique for reducing exposure by removing linear feature relationships from predictions. It: 1) Uses pseudoinverse projection to residualize predictions 2) Preserves higher-order interactions while removing direct effects 3) Can be tuned via proportion parameter and feature"
X Link 2025-11-20T13:26Z 20.1K followers, 40.5K engagements
"You can pick a fight with a man. He can be muscled scarred and tattooed. You might go for it even if he has buddies around. But for the love of god if you see THIS stay away. It is the mark of a demon and you are walking into his lair"
X Link 2025-11-30T18:02Z 20.1K followers, 151K engagements
"Everyone's waiting for certainty before they make the jump. Newsflash: Certainty doesn't exist in any system complex enough to be worth your time. Kelly figured this out in 1956. You don't need to know the outcome. You need: 1) Positive expected value 2) Position sizing that survives being wrong 3) Enough iterations to let compounding do its thing In the short term we regret our failures. In the long term we regret what we never tried. Bezos put it simply: "100% chance of regret if I didn't try. X% chance if I tried and failed." My take on life: - You will never have perfect information -"
X Link 2025-12-02T12:36Z 20.1K followers, 17.1K engagements
"Alpha drop: Numerai forums. Numerai obfuscates all features before giving you data. feature_1. feature_1050 zero semantic meaning. Can't lean on fundamentals or papers forced to discover pure statistical relationships. The forum is people trying to crack why things work without being able to anchor to conventional wisdom. Forced creativity at scale. Some threads: why recent 1yr models beat long history why neutralization backfires handling regime shifts when you can't see what actually changed. Good hunting ground if you want to see statistical thinking divorced from "but Fama-French says"
X Link 2025-11-18T16:24Z 20K followers, 15.1K engagements
"Leveraged ETFs Mechanism: Leveraged ETFs use derivatives (total return swaps futures) with institutional counterparties to achieve 2x or 3x the daily return of an underlying index. They don't hold underlying securities directly. Daily rebalancing: At each day's close they must rebalance derivative positions to maintain target leverage. Example: 2x ETF with $100M at $200M exposure gains X% - now $104M must adjust to $208M exposure. Volatility decay/path dependency: Due to compounding of daily returns multi-day performance diverges significantly from the stated multiple. Choppy markets = decay:"
X Link 2025-11-19T15:30Z 20K followers, 6863 engagements
"Reading the very interesting MS report on Moats and sharing some learnings. I'm not a fundamental investor so take this with a pinch of salt I couldn't tell you GOOG's revenue numbers at gunpoint. But I am broadly fascinated by businesses of all nature and so. --- What is a "moat" and how to actually think and measure about one BUFFET ON MOATS Buffet seems to have popularized the idea of a "moat" from his shareholder letters and meetings: "What we're trying to do is we're trying to find a business with a wide and long-lasting moat around it surround -- protecting a terrific economic castle"
X Link 2025-11-28T13:26Z 20K followers, 22.8K engagements
"But my point exactly is this "have very orthogonal alphas" - I would bet a very significant amount that those alphas are in fact not "very orthogonal". And I say this not in the spirit of throwing shade. I say this factually as someone who has heard a great many "orthogonal alphas""
X Link 2025-12-05T05:51Z 20K followers, XXX engagements
"@MovieTimeDev Someone should backtest this. X. Grab all events across history. X. PIT XX hours from expiry buy NO tokens X. Compute P_outcome (1 if No X if Yes) - P_buy_no X. Horizontal average for day (this is ur daily returns) X. Do cumulative sum X. Plot the time-series This is the cumulative PnL curve"
X Link 2025-10-25T02:15Z 20K followers, 4697 engagements
"People complaining that "Startup X shouldn't be worth $XB because it's just a wrapper around GPT" fundamentally misunderstand how value is created. Everything is a wrapper. Some of these are trillion dollar wrappers: AWS: Wraps data centers into an API. $500B. Stripe: Wraps banks and payment networks into X lines of code. $70B. Shopify: Wraps web hosting and payments into "start a store". $100B. Salesforce: Wraps databases into "track your customers". $250B. Google: Wraps the internet into a search box. $2T. Nobody uses raw compute raw infrastructure or raw APIs. They use wrappers that"
X Link 2025-11-18T14:50Z 20K followers, 3931 engagements
"I was a very early Numerai user back in 2016-ish when I was already a working quant. Was only active for a short stint but was so tickled by the idea that a firm was trying to crowdsource signals. Numerai had a certain appeal that Quantopian never quite had for me. I really appreciated that you needed to stake NMR to get allocation and that payback was proportional to your stake. This was elegant - if you don't enforce users to have skin in the game you get adversely selected. Incentives and punishment drives behaviour. Over the years I've been involved in a few crowdsourcing programs across"
X Link 2025-11-20T05:31Z 20K followers, 19.4K engagements
"@quantonisland Very few roles will let you work on different sides of things. Normally you hone your skills in alpha generation and learn portfolio opt via osmosis or as a sub-pm"
X Link 2025-11-20T14:19Z 20K followers, 1463 engagements
"@ScottPh77711570 @MartinShkreli I am an impartial juror only assessing alpha generation abilities"
X Link 2025-11-26T09:15Z 20K followers, XXX engagements
"I really enjoyed reading this if you're an agentic nerd like I am. Here's a neat summary Everyone's building AI agents now. The problem most will face is that AI agents tend to still fail on anything longer than a single context window. Here's what Anthropic learned building harnesses for long-running agents: The core insight: State doesn't belong in the prompt. It belongs in files. Your agent harness needs three things: X. Structured task tracking (JSON not Markdown) X. Unstructured progress notes (plain text) X. Git for checkpoints Why JSON over Markdown Structured data survives context"
X Link 2025-11-27T04:15Z 20K followers, 17.3K engagements
"The line between alpha and risk premia is defined by your risk manager (in consultation with Barra)"
X Link 2025-11-27T09:19Z 20K followers, 10.6K engagements
"I agree with kens conclusions. My indicators are telling me the same things. If you cant see it youre NGMI"
X Link 2025-11-29T18:49Z 20K followers, 8041 engagements
"requested by @0xfdf"
X Link 2025-12-03T07:01Z 20K followers, 4400 engagements
"@richardcraib I think the answer to this can be solved by invoking occam's razor. They are institutional-grade full-timers dedicated to Numerai and are a notch above the rest of the hobbyists"
X Link 2025-12-06T17:57Z 20K followers, XXX engagements
"@richardcraib You might find a handful of them per 1000 active users but it's probably not at a concentration that Numerai strives for to scale Signals"
X Link 2025-12-06T17:58Z 20K followers, XXX engagements
"@CapJSparrrow sending you an mri of my brain"
X Link 2025-12-07T13:46Z 20K followers, XXX engagements
"High frequency bozos will tell you HFT is the only way to make money because predictions decay exponentially with time. Mid frequency clowns will tell you that MFT is the ONLY way to make money because trading too fast will erode all transaction costs. Low frequency jesters will tell you that LFT is SUPREMELY better because it allows you to manage BILLIONS rather than millions. --- The point is for every way of making money there is a subset of people that are diametrically opposed to your beliefs. Fuck them and do your thing. Sincerely Guy who has seen strategies across broad spectrums all"
X Link 2025-11-28T05:25Z 20.1K followers, 113.7K engagements
"Common misconception: "My LR model predicts LARGE (e.g. +8%) returns. It must be confident." No. Prediction magnitude tells you almost nothing about confidence. Often the opposite. --- Here's why: X. What determines Your prediction is = X. Large predictions mean: a. Extreme features (far from training mean) b. Large coefficients c. Both None of these imply confidence. --- X. What actually determines confidence Prediction variance: Var() = x'(X'X)x That x'(X'X)x term is called leverage. It measures how far your feature vector sits from the center of your training data. Extreme features - high"
X Link 2025-11-29T04:21Z 20.1K followers, 88.4K engagements
"Mixed-effects models are what you use once you admit your data is lying about the i.i.d. assumption. Quick thread on what they are when to use them and how they differ from linear regression: 1) The usual toy: y = X + Every row is independent same variance no structure. It's less useful once you have repeated measures panel data or anything hierarchical. 2) Mixed-effects model: y = X + Zb + = fixed effects (population-level coefficients) b = random effects (group-level deviations) Z = design matrix that ties observations to groups Now observations that share the same b are correlated by"
X Link 2025-12-03T07:00Z 20.1K followers, 28.5K engagements
"According to @macrocephalopod if I dont get people to retweet to follow me I am going to get cooked on my bet that I will get to 30K followers by year end. I made this bet X weeks ago and my plan was to post frequently for X weeks and then beg for my highbie followers to lend a brother a hand to teach my counterparty not to make bets where reflexivity is involved and I have a chance to shape the outcome"
X Link 2025-12-03T10:51Z 20.1K followers, 158.4K engagements
"Lets say you are a quant researcher and you want to make the leap to being a systematic PM. Let me tell you that the name of the game is to reduce the interviewers perception of how many unknowns they have to deal with. --- If you were an experienced PM the expectation is that you understand every stage in the investment process and have a track record. But for first-time PMs the bar is different. And you should absolutely take advantage of that. Chances are for first-time PMs you will join a team as a sub-PM which means you are essentially at an experience level where the main PM thinks you"
X Link 2025-12-05T04:32Z 20.1K followers, 56.1K engagements
"You get some raw SysLS (and not ChatGPT SLOP) thoughts and analysis for free and I get a retweet like and a follow FAIR"
X Link 2025-12-06T10:10Z 20.1K followers, 1813 engagements
"Anybody who retweets and contributes to getting me above the 20K line will get +20% sharpe in 2026. No exceptions"
X Link 2025-12-07T19:13Z 20.1K followers, 14.2K engagements