{"id":"2078828967407026452","url":"https://x.com/0xTatara/status/2078828967407026452","text":"","author":{"name":"Tatara","username":"0xTatara","avatarUrl":"https://pbs.twimg.com/profile_images/2077648911947247616/Lk8tiYjP_200x200.jpg"},"createdAt":"Sun Jul 19 13:07:03 +0000 2026","engagement":{"replies":8,"retweets":14,"likes":123,"views":314381},"article":{"title":"The Quant Research Pipeline Nobody Explains.","previewText":"From raw market data to tradable alpha. The full process behind every real edge.\nMost people think quant research starts with a strategy idea.\nIt usually starts much earlier.\nBefore the model. Before","coverImageUrl":"https://pbs.twimg.com/media/HNl5wmwW4AATt3s.jpg","content":"## From raw market data to tradable alpha. The full process behind every real edge.\n\nMost people think quant research starts with a strategy idea.\n\nIt usually starts much earlier.\n\nBefore the model. Before the backtest. Before the signal.\n\nA real quant research pipeline begins with a question about the market that can survive contact with data.\n\nWhy does this price move exist?\n\nWho is creating it?\n\nHow long does it last?\n\nCan it be measured before everyone else measures it?\n\nCan it still be traded after costs?\n\nThat is the part almost nobody explains.\n\nPeople see the final strategy.\n\nThey see the Sharpe ratio.\n\nThey see the machine learning model.\n\nThey see the execution system.\n\nWhat they do not see is the long pipeline that sits underneath all of it.\n\nRaw data enters one side.\n\nA tiny piece of tradable information comes out the other.\n\nAlmost everything in between gets rejected.\n\nThat rejection process is quant research.\n\n## Alpha does not begin with a model\n\nThe easiest way to waste six months in quantitative trading is to start with the algorithm.\n\nOpen XGBoost.\n\nBuild a neural network.\n\nFeed it ten years of price data.\n\nAsk it to predict tomorrow.\n\nThis feels like research because the code is difficult.\n\nBut difficulty is not the same as edge.\n\nThe strongest quant research usually starts with a hypothesis about how markets actually behave.\n\nMaybe large institutional orders create temporary price pressure.\n\nMaybe investors react slowly to certain information.\n\nMaybe two related assets occasionally diverge because liquidity disappears in one of them.\n\nMaybe options dealers are forced to hedge in predictable ways when volatility moves.\n\nMaybe an index rebalance creates mechanical demand that has nothing to do with fundamentals.\n\nThe first question is not:\n\nWhich model should I use?\n\nIt is:\n\nWhat mechanism could create a repeatable inefficiency?\n\nThat distinction changes the entire pipeline.\n\nIf you begin with a model, you search the data for anything it can predict.\n\nIf you begin with a market mechanism, you search for evidence that the mechanism leaves a measurable footprint.\n\nOne approach produces patterns.\n\nThe other has a chance of producing alpha.\n\nThe model comes later.\n\nMuch later.\n\n## Stage one: turn the market into data\n\nMarkets do not arrive as clean tables.\n\nThey arrive as trades.\n\nQuotes.\n\nOrder book updates.\n\nCorporate actions.\n\nEconomic releases.\n\nNews.\n\nOptions chains.\n\nFunding rates.\n\nLiquidations.\n\nETF flows.\n\nEarnings transcripts.\n\nAlternative datasets.\n\nMillions of events happening on different clocks.\n\nThe first real problem in quant research is deciding what the market actually looked like at a specific moment in time.\n\nThat sounds simple until you try to build it.\n\nPrices need to be adjusted correctly.\n\nDelisted securities need to remain in historical datasets.\n\nTimestamps need to match.\n\nMissing values need to be understood, not blindly filled.\n\nA macroeconomic number needs to reflect what traders knew when it was originally released, not the revised number available today.\n\nAn earnings report needs to appear in the dataset when it became public, not when the quarter ended.\n\nAt high frequency, even milliseconds matter.\n\nA quote received after a trade cannot be used to explain the trade before it.\n\nThis is why institutional quant firms spend enormous effort on data infrastructure.\n\nA sophisticated model trained on contaminated data is still a contaminated model.\n\nGarbage does not become alpha because you passed it through a neural network.\n\nThe first edge is often simply knowing what actually happened.\n\n![](https://pbs.twimg.com/media/HNl6zLDW0AAdoP6.jpg)\n\n## Stage two: create the research universe\n\nBefore testing a signal, a quant needs to decide where the signal is allowed to exist.\n\nThis is the universe.\n\nMaybe it is the largest 500 US equities.\n\nMaybe it is every liquid futures contract.\n\nMaybe it is crypto perpetuals above a minimum daily volume.\n\nMaybe it is equity options with enough open interest.\n\nThis decision looks boring.\n\nIt is not.\n\nChange the universe and you can completely change the result.\n\nSuppose you test a stock strategy using only companies that exist today.\n\nThe bankrupt companies disappear.\n\nThe delisted companies disappear.\n\nThe businesses that collapsed disappear.\n\nYou have accidentally created a market where losers are deleted from history.\n\nThe backtest improves.\n\nReality does not.\n\nLiquidity creates another problem.\n\nA signal might look incredible across thousands of tiny stocks.\n\nBut if your strategy needs to buy $5 million of something that trades $200,000 per day, the signal does not exist at your size.\n\nSo the research universe is not just:\n\nWhere does this idea work?\n\nIt is:\n\nWhere could this idea actually be traded?\n\nThis is where quant research begins separating mathematical possibility from financial reality.\n\n## Stage three: transform observations into features\n\nRaw prices are rarely the final input.\n\nResearchers transform market behavior into variables that models can understand.\n\nReturns.\n\nVolatility.\n\nVolume imbalance.\n\nMomentum.\n\nDistance from a moving average.\n\nRelative valuation.\n\nYield curve changes.\n\nOption skew.\n\nFunding rates.\n\nOrder flow.\n\nSentiment.\n\nCross-sectional rankings.\n\nResidual returns.\n\nThese are features.\n\nBut feature engineering in finance is different from ordinary machine learning.\n\nA normal ML problem often assumes that the relationship between input and output is relatively stable.\n\nFinance does not give you that luxury.\n\nA feature can work for five years and disappear.\n\nIt can reverse.\n\nIt can become crowded.\n\nIt can only work during high volatility.\n\nIt can only exist in small caps.\n\nIt can vanish the moment transaction costs increase.\n\nThis means every feature needs context.\n\nTake momentum.\n\n“Price went up” is not enough.\n\nOver what horizon?\n\nRelative to what?\n\nThe market?\n\nThe sector?\n\nIts own historical volatility?\n\nWas the move driven by one large jump or twenty gradual sessions?\n\nDid volume confirm it?\n\nDid the asset move independently or simply follow the entire market?\n\nThe same raw price series can generate dozens of different signals depending on how you frame the question.\n\nThat is why good quant research is often less about finding more data and more about representing the same data correctly.\n\nThe information was already there.\n\nThe feature makes it visible.\n\n## Stage four: remove what you do not want to trade\n\nOne of the most powerful ideas in quantitative research is decomposition.\n\nA stock moves three percent.\n\nRetail sees a stock that moved three percent.\n\nA quant asks where the three percent came from.\n\nMaybe the entire market moved two percent.\n\nMaybe the technology sector moved another half percent.\n\nMaybe momentum factors explain another twenty basis points.\n\nMaybe the stock's sensitivity to interest rates explains more.\n\nAfter removing all of that, perhaps only thirty basis points remain unexplained.\n\nThat residual can be more interesting than the original three percent move.\n\nBecause the quant may not want to trade the market.\n\nOr technology.\n\nOr interest rates.\n\nThe quant wants the specific abnormal movement that remains after those exposures are removed.\n\nThis is the logic behind factor models and many forms of statistical arbitrage.\n\nStart with the observed return.\n\nRemove common components.\n\nStudy the leftover.\n\nSometimes the residual contains nothing.\n\nSometimes it contains temporary pressure.\n\nSometimes it identifies an asset that moved too far relative to everything that normally explains it.\n\nThe goal is not always to predict the entire price.\n\nIt is often to isolate the smallest part of the price that might be predictable.\n\nThat is a much easier problem.\n\n## Stage five: turn the feature into a signal\n\nA feature describes the market.\n\nA signal tells the system what to do.\n\nThose are not the same thing.\n\nSuppose your research shows that unusually negative residual returns tend to reverse over the next two days.\n\nYou still do not have a strategy.\n\nYou need rules.\n\nHow negative is unusual?\n\nDo you trade every event?\n\nOnly the extreme ones?\n\nDo you rank the entire universe and buy the cheapest residuals?\n\nDo you short the richest?\n\nDo you hold for one day?\n\nTwo?\n\nFive?\n\nDo you scale positions by volatility?\n\nDo you neutralize market exposure?\n\nSector exposure?\n\nBeta?\n\nNow the research becomes portfolio construction.\n\nThis is where many promising ideas disappear.\n\nA signal can be statistically predictive and still produce a terrible portfolio.\n\nMaybe all your long positions are technology stocks.\n\nMaybe your supposedly market-neutral strategy secretly has massive beta.\n\nMaybe ten different signals are actually the same momentum trade wearing different names.\n\nMaybe your strongest positions appear in the least liquid assets.\n\nThe signal does not exist independently.\n\nIt lives inside a portfolio.\n\nAnd the portfolio is what actually makes or loses money.\n\n![](https://pbs.twimg.com/media/HNl6-oCWMAAE9z2.jpg)\n\n## Stage six: ask whether the signal is real\n\nThis is where quant research becomes hostile.\n\nA good researcher does not try to prove the signal works.\n\nThe researcher tries to destroy it.\n\nChange the period.\n\nDoes it survive?\n\nChange the universe.\n\nDoes it survive?\n\nShift the signal by one day.\n\nDoes it survive?\n\nUse slightly different parameters.\n\nDoes it survive?\n\nRemove the best-performing year.\n\nDoes it survive?\n\nIncrease transaction costs.\n\nDoes it survive?\n\nTest another market.\n\nDoes it survive?\n\nA weak research process asks:\n\nHow high is the Sharpe ratio?\n\nA strong research process asks:\n\nHow difficult is it to make the Sharpe ratio disappear?\n\nThat is robustness.\n\nA signal that only works with a 37-day lookback and dies at 35 or 40 is suspicious.\n\nA signal that works at 30, 40, 50, and 60 days with different levels of strength is more interesting.\n\nThe goal is not to find the perfect parameter.\n\nPerfect parameters usually belong to the past.\n\nThe goal is to identify a region where the economic effect appears consistently.\n\nReal alpha should bend.\n\nIt should not instantly break.\n\n![](https://pbs.twimg.com/media/HNl7F-_XEAABVZr.jpg)\n\n## Stage seven: separate discovery from verification\n\nThis is one of the most important pieces of the pipeline.\n\nThe same process that discovers a strategy should not be trusted to validate it.\n\nWhy?\n\nBecause research creates attachment.\n\nYou spent three weeks on the signal.\n\nYou understand the theory.\n\nYou wrote the code.\n\nYou saw the beautiful backtest.\n\nYou want it to work.\n\nThat makes you dangerous.\n\nQuant firms solve this by separating creation from verification.\n\nThe researcher produces the idea.\n\nAnother process attacks it.\n\nCheck the data.\n\nCheck the assumptions.\n\nCheck the statistics.\n\nCheck the exposure.\n\nCheck for leakage.\n\nCheck whether the result survives out-of-sample.\n\nThe maker asks:\n\nCan this be alpha?\n\nThe checker asks:\n\nHow is this lying to us?\n\nBoth are necessary.\n\nThis becomes even more important when automation enters the pipeline.\n\nA system can generate hundreds of research hypotheses.\n\nAnother process can write the code.\n\nAnother can optimize parameters.\n\nThe speed of discovery becomes almost unlimited.\n\nBut faster discovery also means faster production of false positives.\n\nIf you can test one thousand times more strategies, your verification system needs to become one thousand times more skeptical.\n\nAutomation does not remove the scientific method.\n\nIt makes the scientific method more important.\n\n## Stage eight: simulate the trade, not the signal\n\nThis is where academic alpha meets the actual market.\n\nYour signal says buy.\n\nAt what price?\n\nThat question can kill the entire strategy.\n\nThe backtest assumes you traded at the close.\n\nCould you actually have traded at the close?\n\nDid the information used by the signal arrive before the closing auction?\n\nHow much volume was available?\n\nWould your order move the market?\n\nWould you cross the spread?\n\nWould the fill happen immediately?\n\nWould the signal decay while the order was executing?\n\nExecution is not a small adjustment added after research.\n\nFor many strategies, execution is part of the research.\n\nSuppose a signal predicts ten basis points of return.\n\nIf trading costs eight basis points, you have a weak strategy.\n\nIf trading costs twelve, you have no strategy.\n\nAt high frequency, the raw alpha can be almost invisible.\n\nA few basis points.\n\nSometimes less.\n\nThe entire business depends on extracting that edge without destroying it.\n\nThis is why two firms can discover the same signal and produce completely different returns.\n\nOne has better execution.\n\nThe signal is public.\n\nThe implementation is the moat.\n\n## Stage nine: size the edge\n\nEven after finding a genuine signal, one question remains.\n\nHow much should you bet?\n\nThis is where prediction becomes risk.\n\nA signal can be profitable and still destroy the portfolio if sized incorrectly.\n\nExpected return matters.\n\nBut so does volatility.\n\nCorrelation.\n\nDrawdown.\n\nTail risk.\n\nLiquidity.\n\nCapacity.\n\nA strategy with high expected return and extreme downside risk may deserve less capital than a weaker but more stable strategy.\n\nMultiple signals also interact.\n\nFive independent edges can create a strong portfolio.\n\nFive highly correlated edges can create one giant hidden bet.\n\nPosition sizing turns research into a capital allocation problem.\n\nThe question is no longer:\n\nIs this trade good?\n\nIt becomes:\n\nHow much capital should this edge receive relative to every other edge we have?\n\nThat is a completely different level of thinking.\n\nThe strategy is not competing against cash.\n\nIt is competing against every other possible use of capital.\n\n## Stage ten: production is another research experiment\n\nThe biggest mistake is thinking the pipeline ends when the strategy goes live.\n\nThat is when the most valuable dataset begins.\n\nLive performance.\n\nNow you can compare the real strategy with the simulated one.\n\nExpected slippage versus actual slippage.\n\nExpected fills versus actual fills.\n\nExpected turnover versus actual turnover.\n\nExpected alpha decay versus actual decay.\n\nThe difference between simulation and production contains information.\n\nMaybe the signal is correct but execution is poor.\n\nMaybe market impact was underestimated.\n\nMaybe the strategy performs differently at certain times of day.\n\nMaybe one venue consistently produces worse fills.\n\nMaybe the alpha disappears faster than expected.\n\nMaybe the backtest assumed liquidity that never actually existed.\n\nProduction generates feedback.\n\nThat feedback returns to research.\n\nThe pipeline becomes a loop.\n\nData.\n\nHypothesis.\n\nFeature.\n\nSignal.\n\nValidation.\n\nPortfolio.\n\nExecution.\n\nMonitoring.\n\nFeedback.\n\nThen back to data.\n\nThe quant system is not a collection of strategies.\n\nIt is a machine for continuously testing reality.\n\n![](https://pbs.twimg.com/media/HNl7So-XoAA7SwJ.jpg)\n\n## Most research ideas should die\n\nThis is the part nobody likes to talk about.\n\nA healthy quant research pipeline destroys most ideas.\n\nYou might start with one hundred hypotheses.\n\nFifty show no statistical relationship.\n\nTwenty work only during one historical period.\n\nTen disappear after fixing data leakage.\n\nEight die after transaction costs.\n\nFive collapse out-of-sample.\n\nFour are too correlated with strategies already in the portfolio.\n\nTwo have insufficient capacity.\n\nOne survives.\n\nThat one might not even be spectacular.\n\nMaybe the Sharpe ratio is 1.1.\n\nMaybe the edge is only a few basis points.\n\nMaybe it works only in one small corner of the market.\n\nThat is fine.\n\nThe purpose of the pipeline was never to make every idea work.\n\nThe purpose was to prevent bad ideas from reaching capital.\n\nA quant firm with great researchers and a weak rejection process can still lose money.\n\nA firm with a strong research pipeline can survive thousands of failed experiments.\n\nFailure becomes cheap.\n\nThat is the advantage.\n\n## The real moat is the pipeline\n\nPeople obsess over secret algorithms.\n\nThe hidden indicator.\n\nThe magical model.\n\nThe system that predicts markets.\n\nThat is usually the wrong place to look.\n\nIndividual signals decay.\n\nStrategies get crowded.\n\nData becomes available to competitors.\n\nModels become open source.\n\nWhat survives is the system around them.\n\nHow quickly can you test a new hypothesis?\n\nHow clean is your data?\n\nHow aggressively do you reject false discoveries?\n\nHow accurately can you simulate execution?\n\nHow quickly do you detect signal decay?\n\nHow safely can you deploy new strategies?\n\nHow much knowledge from failed research gets preserved?\n\nThis is the real compounding mechanism.\n\nOne strategy can die.\n\nThe pipeline produces another.\n\nOne data source loses value.\n\nThe pipeline evaluates a new one.\n\nOne market becomes too competitive.\n\nThe research moves somewhere else.\n\nThe permanent edge is not always the alpha itself.\n\nThe permanent edge is the ability to keep finding, testing, deploying, and killing alpha faster than everyone else.\n\n## What quant research actually looks like\n\nIt is less glamorous than people imagine.\n\nMost days are not spent discovering billion-dollar strategies.\n\nThey are spent cleaning data.\n\nDebugging timestamps.\n\nQuestioning suspicious results.\n\nFinding hidden exposures.\n\nWatching promising ideas disappear after costs.\n\nRejecting models.\n\nRebuilding tests.\n\nComparing live performance with simulation.\n\nThe final strategy may look simple.\n\nThe process that allowed you to trust it is not.\n\nThat is the part outsiders never see.\n\nThey see one equation.\n\nOne model.\n\nOne trading bot.\n\nUnderneath it is an entire research architecture built around one principle:\n\nDo not let randomness reach capital.\n\n## The final reframe\n\nQuant research is not a search for predictions.\n\nIt is a filtration system.\n\nThe market produces an almost infinite amount of noise.\n\nThe pipeline removes pieces of it.\n\nFirst bad data.\n\nThen meaningless patterns.\n\nThen common risk.\n\nThen unstable signals.\n\nThen statistical accidents.\n\nThen strategies that cannot survive costs.\n\nThen strategies that cannot scale.\n\nThen strategies that fail in production.\n\nWhat remains is usually small.\n\nA few basis points.\n\nA weak statistical tendency.\n\nA temporary inefficiency.\n\nSomething most people would ignore.\n\nBut if it is real.\n\nIf it survives verification.\n\nIf it survives execution.\n\nIf it can be sized safely.\n\nIf it can be repeated.\n\nThen that small piece becomes alpha.\n\nThe model was never the whole game.\n\nThe backtest was never the whole game.\n\nThe signal was never the whole game.\n\nThe pipeline is the game.\n\nBecause the best quant firms are not better at predicting the future once.\n\nThey are better at building systems that continuously separate what is real from what only looks real.\n\nAnd in a market made almost entirely of noise, that may be the most valuable edge of all.\n\n## If you read this far:\n\n- BOOKMARK THIS.\n\n- Follow [@0xTatara](https://x.com/@0xTatara)"}}