{"id":241054,"date":"2026-04-02T15:44:41","date_gmt":"2026-04-02T19:44:41","guid":{"rendered":"https:\/\/ibkrcampus.com\/campus\/?p=241054"},"modified":"2026-04-02T15:46:47","modified_gmt":"2026-04-02T19:46:47","slug":"brave-new-backtest","status":"publish","type":"post","link":"https:\/\/www.interactivebrokers.com\/campus\/ibkr-quant-news\/brave-new-backtest\/","title":{"rendered":"Brave New Backtest"},"content":{"rendered":"\n

The article “Brave New Backtest” was originally posted on Robot Wealth<\/a> blog.<\/em><\/p>\n\n\n\n

My last two articles on AI and trading research got more engagement than almost anything I\u2019ve written.<\/p>\n\n\n\n

\u201cMore of the Disease, Faster<\/a>\u201d argued that LLMs can\u2019t answer the critical question: who pays you and why?<\/em><\/p>\n\n\n\n

\u201cAI Will Create Millions of Quants<\/a>\u201d went deeper on the why: AI makes beautiful backtests trivially easy to produce, which means more false discoveries, more overfitting dressed up as research, and more people confusing a good-looking equity curve with a real edge.<\/p>\n\n\n\n

Both pieces argued from experience. I\u2019ve been doing this long enough to know what works and what doesn\u2019t, and I was pretty confident in the arguments. But the most telling thing about the comments wasn\u2019t the disagreement. It was how many people clearly don\u2019t understand that backtesting, statistics, and pattern matching aren\u2019t research.<\/p>\n\n\n\n

Their LLMs don\u2019t understand this either. And you can\u2019t outsource the thinking part of trading research to a machine that doesn\u2019t understand what research is<\/em>.<\/p>\n\n\n\n

Two recent papers suggest this isn\u2019t a temporary limitation that better models will fix. It\u2019s structural.<\/p>\n\n\n\n

A feature<\/em> of the system, rather than a bug<\/em>.<\/p>\n\n\n\n

And they provide the scientific backing for what I\u2019ve been saying.<\/p>\n\n\n\n

My earlier articles argued that the training data is the problem. That\u2019s true. But it\u2019s only Problem 1. Here are Problems 2 and 3.<\/p>\n\n\n\n

Problem 1: The Training Data (Quick Recap)<\/strong><\/h2>\n\n\n\n

For anyone who missed the first two articles, the short version:<\/p>\n\n\n\n

LLMs are trained on the internet. The internet\u2019s trading content is overwhelmingly noise. \u201cDon\u2019t fight the trend.\u201d \u201cUse RSI for entries.\u201d \u201cPaper trade for six months.\u201d \u201cValidate with out of sample data.\u201d \u201cCointegration matters for pairs trading\u201d.<\/p>\n\n\n\n

The dominant paradigm online is that backtesting and statistics IS research, that finding a pattern in historical data IS finding an edge.<\/p>\n\n\n\n

The critical question, \u201cwho pays you and why?\u201d<\/strong><\/em>, barely exists in the training data.<\/p>\n\n\n\n

Mechanism-based thinking, structural edges, understanding participant constraints\u2026 this stuff lives in the tail of the distribution. The vast majority of trading content online is conventional wisdom dressed up as insight.<\/p>\n\n\n\n

So the LLM doesn\u2019t know that backtesting isn\u2019t research because the internet doesn\u2019t know it either. It reproduces the dominant paradigm with extraordinary confidence. Which is bad.<\/p>\n\n\n\n

But some people reasonably argued: fix the training data, and you fix the problem. Train the model on better material. Use RAG. Curate your knowledge base.<\/p>\n\n\n\n

Two recent papers suggest that\u2019s only partly true. Even with perfect training data, two more problems remain. And these are architectural, meaning they\u2019re baked into how these models work.<\/p>\n\n\n\n

Problem 2: The Forgetful Machine<\/strong><\/h2>\n\n\n\n

A paper called \u201cUnable to Forget<\/a>\u201d (Wang & Sun, 2025) tested something deceptively simple: can LLMs track a value that changes over time?<\/p>\n\n\n\n

Imagine a patient\u2019s blood pressure being recorded throughout a hospital visit. BP at triage: 120. Ten minutes later: 128. At discharge: 125. Ask the model: what\u2019s the current blood pressure?<\/em><\/p>\n\n\n\n

Simple, right? The answer is right there at the end of the context. The model was explicitly told to retrieve the most recent value.<\/p>\n\n\n\n

Across 35+ models<\/strong> (GPT, Claude, Llama, Gemini, DeepSeek, the lot), accuracy declined log-linearly toward zero as the number of prior updates increased. The more historical values the model had seen for the same variable, the worse it got at retrieving the current one.<\/p>\n\n\n\n

And this wasn\u2019t a gentle, noisy, or hard-to-discern decline. It was consistent<\/em>, relentless, and universal<\/em>.<\/p>\n\n\n\n

The researchers call this \u201cproactive interference,\u201d borrowed from cognitive science. Earlier values compete with the current value in the model\u2019s retrieval process. Old information interferes with new information. And the interference gets worse, continuously, as context accumulates.<\/p>\n\n\n\n

Three things make this particularly frightening:<\/p>\n\n\n\n

Prompt engineering doesn\u2019t fix it.<\/strong> They tried telling the model to \u201cforget\u201d old values. They tried \u201cfocus on the most recent update.\u201d They tried meta-prompts asking the model to self-assess what to prioritise. Marginal improvement at best. In some cases, the \u201cforget\u201d instruction actually made things worse, anchoring errors around the point where the instruction was inserted.<\/p>\n\n\n\n

Bigger context windows don\u2019t help.<\/strong> A bigger window just gives the model more room to accumulate more interference.<\/p>\n\n\n\n

It\u2019s universal.<\/strong> Every<\/em> model tested showed the same pattern<\/em>. From tiny open-source models to the biggest proprietary ones. The curve shape was the same. Bigger models declined more slowly, but they all declined.<\/p>\n\n\n\n

Now think about what trading research requires.<\/p>\n\n\n\n

Some trading concepts are evergreen: the theory of edge, the importance of mechanism, portfolio construction principles. An LLM can learn those from training data, and it does a reasonable job of reproducing them, when it isn\u2019t drowning them in conventional wisdom, per Problem 1.<\/p>\n\n\n\n

But even when explicitly told to use a specific database of curated material, it will still throw in the conventional wisdom with all the conviction in the world (more on this below).<\/p>\n\n\n\n

But applying<\/em> those concepts requires tracking an ever-changing environment. Market regimes shift. Carry changes direction. Volatility spikes and mean-reverts. Correlations break down. Borrow costs change. Liquidity dries up and returns. New players enter. Regulations get updated and replaced.<\/p>\n\n\n\n

A model that can\u2019t reliably track \u201cthe current value of blood pressure\u201d in a simple key-value test certainly can\u2019t track evolving states of things that affect markets.<\/p>\n\n\n\n

Knowing the theory of edge is one thing. Applying it in a market that adapts and evolves is where the dynamic tracking matters. And the architecture fails at exactly that task.<\/p>\n\n\n\n

Crucially, this isn\u2019t a training data problem. You could train the model on perfect data and it would still fail at tracking sequential updates. The architecture can\u2019t handle it.<\/p>\n\n\n\n

I\u2019ve long felt in my bones that the conviction with which LLMs speak about trading is truly misplaced. And now there\u2019s some published research that says so too.<\/p>\n\n\n\n

Problem 3: The Artificial Hivemind<\/strong><\/h2>\n\n\n\n

A second paper, \u201cArtificial Hivemind<\/a>\u201d (Jiang et al., NeurIPS 2025<\/a>), measured something equally concerning: when you ask LLMs open-ended questions, how diverse are their answers?<\/p>\n\n\n\n

They tested 25+ models across 100 open-ended queries, generating 50 responses per model per query. The results are shocking<\/strong><\/em>\u2026 but I\u2019d wager they align with your recent experience of online content.<\/p>\n\n\n\n

Intra-model repetition:<\/strong> A single model gives you the same answer over and over. In 79% of queries, the pairwise similarity between responses from the same model<\/em> exceeded 0.8. Ask it the same question fifty times, you get essentially the same answer fifty times.<\/strong> Even with aggressive sampling parameters designed to maximise diversity.<\/p>\n\n\n\n

Inter-model homogeneity:<\/strong> Different models, built by different organisations, with different architectures and different training data, produce strikingly similar outputs. Average pairwise similarity between responses from different<\/strong><\/em> models ranged from 71% to 82%. DeepSeek-V3 and GPT-4o hit 0.81 similarity. These are supposed to be independent systems.<\/p>\n\n\n\n

The paper\u2019s most vivid example asked 25 different models to \u201cwrite a metaphor about time.\u201d Fifty responses each. 1,250 total responses from independent systems. They cluster into just two metaphors<\/strong><\/em>: \u201ctime is a river\u201d and \u201ctime is a weaver.\u201d That\u2019s it.<\/p>\n\n\n\n

Twenty-five different models, all converging on the same two ideas.<\/strong><\/em><\/p>\n\n\n\n

And if you\u2019ve been feeling a growing disdain for all that similarly soulless LLM-generated content popping up in your feed, now you know the mechanism. What we consume is collapsing to the mode of anti-creativity.<\/p>\n\n\n\n

\u201cTime is a river\u201d dominates because it\u2019s the most common metaphor about time in the training data.<\/p>\n\n\n\n

And that\u2019s the mechanism that matters for trading.<\/p>\n\n\n\n

Mode collapse means convergence toward the most represented patterns in the training data.<\/strong> For trading, the mode of internet content is conventional wisdom: \u201cuse a stop loss,\u201d \u201cpaper trade first,\u201d \u201cbacktest with moving averages,\u201d \u201cvalidate with out of sample data,\u201d \u201ccointegration matters for pairs trading\u201d. All the usual bollocks.<\/p>\n\n\n\n

The actually useful insights (mechanism-based thinking, structural edges, \u201cwho pays you and why?\u201d) live in the tail of the distribution of online content.<\/p>\n\n\n\n

Mode collapse systematically suppresses the tail and amplifies the centre. So it\u2019s not just that the training data is bad (Problem 1). The architecture preferentially surfaces the bad stuff<\/em> because that\u2019s what convergence toward the mode means.<\/p>\n\n\n\n

\u201cUse a stop loss\u201d is the \u201ctime is a river\u201d of trading content. It\u2019s the modal output. The rare, useful insights get pushed out by convergence.<\/p>\n\n\n\n

I experienced this first-hand.<\/p>\n\n\n\n

I built a RAG system on our TLQ Bootcamp<\/a> material, which is high-quality, mechanism-first content. Explicitly told the LLM to only<\/em> respond using data from the RAG database. It still mixed in conventional wisdom that was clearly not in the database: generic stop loss advice, paper trading suggestions. Both problems at once: couldn\u2019t suppress prior training (proactive interference from Problem 2), and defaulted to the most common trading advice (mode collapse from Problem 3).<\/p>\n\n\n\n

The Stepford Quants<\/strong><\/h3>\n\n\n\n

Remember the backtest cycle of doom from Article 2<\/a>? AI makes it trivially easy to produce beautiful backtests that prove nothing. The unsuspecting trader ends up with a complicated algorithm for fitting to past noise and learns nothing useful in the process.<\/p>\n\n\n\n

Mode collapse implies an added layer of insidiousness: everyone using LLMs for trading research converges on roughly the same conventional-wisdom-flavoured strategies.<\/p>\n\n\n\n

It\u2019s the perfect tool to tell you what already fits into your preconceived frameworks.<\/p>\n\n\n\n

The implication is that millions of AI quants aren\u2019t just running the cycle of doom independently. They\u2019re running the same<\/em> doom cycle. And the \u201cstrategies\u201d being discovered aren\u2019t real edges in the first place. They\u2019re the modal output of bad training data, amplified by architectural convergence.<\/p>\n\n\n\n

The Stepford Quants: pleasant, productive, and identical.<\/p>\n\n\n\n

LLMs Are Brilliant Coders but Terrible Traders<\/strong><\/h2>\n\n\n\n

This is the part that makes the whole thing click.<\/p>\n\n\n\n

I use AI for coding every day. It\u2019s extraordinary. It writes boilerplate, handles data wrangling, produces charts, writes tests. I\u2019ve said this in both previous articles, and I\u2019ll keep saying it: AI is the best research assistant<\/em> I\u2019ve ever had.<\/p>\n\n\n\n

Why does it work so well for code?<\/p>\n\n\n\n

Because mode collapse in coding is convergence toward best practices<\/em>.<\/p>\n\n\n\n

The training data (Stack Overflow, GitHub, documentation) is self-correcting: bad code gets downvoted, good patterns get reinforced. Software engineering has well-established paradigms and right answers (or at least well-established good answers). When the model converges on \u201cuse a dictionary for O(1) lookups\u201d or \u201chandle this edge case with a try-except block,\u201d that convergence is helpful<\/em>.<\/p>\n\n\n\n

Mode collapse in trading is convergence toward conventional wisdom<\/em>, which is mostly wrong. The LLM surfaces \u201cvalidate with out of sample data\u201d and \u201cuse cointegration tests to find pairs trading opportunities\u201d because that\u2019s what most trading content says. The actually useful insights (mechanism-based thinking, structural edges) live in the tail and get suppressed by convergence.<\/p>\n\n\n\n

It\u2019s the same architectural property. But it\u2019s beneficial in one domain, lethal in another.<\/p>\n\n\n\n

Similarly, proactive interference doesn\u2019t matter much in coding. You\u2019re working on a specific, well-defined task. The context is relatively static: here\u2019s the codebase, here\u2019s what I want to change, here are the tests. But for tracking an evolving, complex state, where the whole point is that yesterday\u2019s values are different from today\u2019s and the model needs to know which is current, interference is devastating.<\/p>\n\n\n\n

This explains exactly why the appropriate use of LLMs in trading is to implement<\/em> a good idea (coding task, well-suited) rather than to find<\/em> the idea in the first place (trading research, structurally ill-suited).<\/p>\n\n\n\n

The Bug Is Not a Bug<\/strong><\/h2>\n\n\n\n

There are at least three layers of limitation, and only one is even theoretically fixable:<\/p>\n\n\n\n

    \n
  • Problem 1 (training data):<\/strong>\u00a0Fixable in principle. Better data, better RAG, curated knowledge bases. But fixing it doesn\u2019t solve Problems 2 and 3.<\/li>\n\n\n\n
  • Problem 2 (proactive interference):<\/strong>\u00a0Architectural. Tested across 35+ models from every major organisation. Bigger models decline more slowly, but they all decline. Prompt engineering doesn\u2019t fix it. Bigger context windows don\u2019t fix it.<\/li>\n\n\n\n
  • Problem 3 (mode collapse):<\/strong>\u00a0Architectural. Tested across 70+ models. Different organisations, different architectures, different training data, same convergence. Model ensembles don\u2019t help because different models produce the same outputs anyway.<\/li>\n<\/ul>\n\n\n\n

    The \u201cjust wait for GPT-6\u201d argument fails because Problems 2 and 3 are universal across every model tested, spanning all architectures and all organisations. These aren\u2019t bugs that get fixed with the next release. They\u2019re properties of how these systems work.<\/p>\n\n\n\n

    And, look, I could be wrong. Maybe some architectural breakthrough changes this. But these papers tested the full spectrum of current models, and the patterns were universal. I wouldn\u2019t bet on it, short of a completely new and different architectural paradigm.<\/p>\n\n\n\n

    This reinforces the Edge Alchemy framework. The human-driven theory of edge (\u201cwho pays you and why?\u201d) is the step that LLMs cannot do, for multiple independent reasons. The training data doesn\u2019t teach it (Problem 1). The architecture can\u2019t track evolving state (Problem 2). And it converges on the most common answers, which are wrong (Problem 3).<\/p>\n\n\n\n

    Here\u2019s a workflow that stands a chance of working: human generates insight (theory of edge), then AI implements it (coding, data wrangling), then human evaluates results.<\/p>\n\n\n\n

    AI for the implementation, humans for the thinking.<\/p>\n\n\n\n

    Go Easy on the Soma<\/strong><\/h2>\n\n\n\n

    Brave New World<\/em> is my favourite novel. Huxley\u2019s genius was showing that the most dangerous dystopia isn\u2019t the one where people are oppressed. It\u2019s the one where everyone\u2019s perfectly content. Nobody questions anything because the system feels good from the inside.<\/p>\n\n\n\n

    That\u2019s AI-assisted trading research right now.<\/p>\n\n\n\n

    It feels productive. You\u2019re generating strategies, running backtests, getting clean code in minutes<\/em>.<\/p>\n\n\n\n

    The dystopia is that you\u2019re converging on conventional wisdom while feeling like you\u2019re doing cutting-edge work. Work that moves your life forward. The machine tells you what you want to hear, confirms what you already believe, and does it with such confidence and speed that you never stop to ask whether any of it is real.<\/p>\n\n\n\n

    AI is the best research assistant ever built. I use it every day and it makes me faster at everything except the one thing that actually matters: understanding why an edge exists and whether it will persist.<\/p>\n\n\n\n

    That\u2019s three independent reasons to keep doing the hard work of understanding why<\/em> edges exist. You only learn this stuff by reading, talking to other people, and, primarily, by doing. There\u2019s no shortcut, no matter how good the AI gets.<\/p>\n\n\n\n

    I didn\u2019t need research papers to tell me this, but it\u2019s nice to have the receipts.<\/p>\n\n\n\n

    References:<\/strong><\/p>\n\n\n\n

    Wang, C. & Sun, J.V. (2025). \u201cUnable to Forget: Proactive Interference Reveals Working Memory Limits in LLMs Beyond Context Length.\u201d arXiv:2506.08184v3<\/em>.<\/p>\n\n\n\n

    Jiang, L. et al. (2025). \u201cArtificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond).\u201d 39th Conference on Neural Information Processing Systems (NeurIPS 2025)<\/em>.<\/p>\n\n\n\n

    <\/p>\n","protected":false},"excerpt":{"rendered":"

    The dominant paradigm online is that backtesting and statistics IS research, that finding a pattern in historical data IS finding an edge.<\/p>\n","protected":false},"author":271,"featured_media":233527,"comment_status":"open","ping_status":"closed","sticky":true,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":"","jetpack_post_was_ever_published":false},"categories":[339,338,341],"tags":[14939,4873,21341,17952,21004],"contributors-categories":[13676],"class_list":["post-241054","post","type-post","status-publish","format-standard","has-post-thumbnail","category-data-science","category-ibkr-quant-news","category-quant-development","tag-artificial-intelligence-ai","tag-backtesting","tag-coding-best-practices","tag-large-language-models-llms","tag-statistical-analysis","contributors-categories-robot-wealth"],"pp_statuses_selecting_workflow":false,"pp_workflow_action":"current","pp_status_selection":"publish","acf":[],"yoast_head":"\nBrave New Backtest | IBKR Quant<\/title>\n<meta name=\"description\" content=\"The dominant paradigm online is that backtesting and statistics IS research, that finding a pattern in historical data IS finding an edge.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.interactivebrokers.com\/campus\/wp-json\/wp\/v2\/posts\/241054\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Brave New Backtest\" \/>\n<meta property=\"og:description\" content=\"The dominant paradigm online is that backtesting and statistics IS research, that finding a pattern in historical data IS finding an edge.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.interactivebrokers.com\/campus\/ibkr-quant-news\/brave-new-backtest\/\" \/>\n<meta property=\"og:site_name\" content=\"IBKR Campus US\" \/>\n<meta property=\"article:published_time\" content=\"2026-04-02T19:44:41+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-04-02T19:46:47+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.interactivebrokers.com\/campus\/wp-content\/uploads\/sites\/2\/2025\/10\/blue-stock-graph-quant.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1000\" \/>\n\t<meta property=\"og:image:height\" content=\"563\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"Kris Longmore\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Kris Longmore\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"11 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\n\t \"@context\": \"https:\\\/\\\/schema.org\",\n\t \"@graph\": [\n\t {\n\t \"@type\": \"NewsArticle\",\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/ibkr-quant-news\\\/brave-new-backtest\\\/#article\",\n\t \"isPartOf\": {\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/ibkr-quant-news\\\/brave-new-backtest\\\/\"\n\t },\n\t \"author\": {\n\t \"name\": \"Kris Longmore\",\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/#\\\/schema\\\/person\\\/79c2a2775a70a4da1accf0068d731933\"\n\t },\n\t \"headline\": \"Brave New Backtest\",\n\t \"datePublished\": \"2026-04-02T19:44:41+00:00\",\n\t \"dateModified\": \"2026-04-02T19:46:47+00:00\",\n\t \"mainEntityOfPage\": {\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/ibkr-quant-news\\\/brave-new-backtest\\\/\"\n\t },\n\t \"wordCount\": 2511,\n\t \"commentCount\": 0,\n\t \"publisher\": {\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/#organization\"\n\t },\n\t \"image\": {\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/ibkr-quant-news\\\/brave-new-backtest\\\/#primaryimage\"\n\t },\n\t \"thumbnailUrl\": \"https:\\\/\\\/www.interactivebrokers.com\\\/campus\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2025\\\/10\\\/blue-stock-graph-quant.jpg\",\n\t \"keywords\": [\n\t \"Artificial Intelligence (AI)\",\n\t \"backtesting\",\n\t \"Coding Best Practices\",\n\t \"Large Language Models (LLMs)\",\n\t \"Statistical Analysis\"\n\t ],\n\t \"articleSection\": [\n\t \"Data Science\",\n\t \"Quant\",\n\t \"Quant Development\"\n\t ],\n\t \"inLanguage\": \"en-US\",\n\t \"potentialAction\": [\n\t {\n\t \"@type\": \"CommentAction\",\n\t \"name\": \"Comment\",\n\t \"target\": [\n\t \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/ibkr-quant-news\\\/brave-new-backtest\\\/#respond\"\n\t ]\n\t }\n\t ]\n\t },\n\t {\n\t \"@type\": \"WebPage\",\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/ibkr-quant-news\\\/brave-new-backtest\\\/\",\n\t \"url\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/ibkr-quant-news\\\/brave-new-backtest\\\/\",\n\t \"name\": \"Brave New Backtest | IBKR Campus US\",\n\t \"isPartOf\": {\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/#website\"\n\t },\n\t \"primaryImageOfPage\": {\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/ibkr-quant-news\\\/brave-new-backtest\\\/#primaryimage\"\n\t },\n\t \"image\": {\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/ibkr-quant-news\\\/brave-new-backtest\\\/#primaryimage\"\n\t },\n\t \"thumbnailUrl\": \"https:\\\/\\\/www.interactivebrokers.com\\\/campus\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2025\\\/10\\\/blue-stock-graph-quant.jpg\",\n\t \"datePublished\": \"2026-04-02T19:44:41+00:00\",\n\t \"dateModified\": \"2026-04-02T19:46:47+00:00\",\n\t \"description\": \"The dominant paradigm online is that backtesting and statistics IS research, that finding a pattern in historical data IS finding an edge.\",\n\t \"inLanguage\": \"en-US\",\n\t \"potentialAction\": [\n\t {\n\t \"@type\": \"ReadAction\",\n\t \"target\": [\n\t \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/ibkr-quant-news\\\/brave-new-backtest\\\/\"\n\t ]\n\t }\n\t ]\n\t },\n\t {\n\t \"@type\": \"ImageObject\",\n\t \"inLanguage\": \"en-US\",\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/ibkr-quant-news\\\/brave-new-backtest\\\/#primaryimage\",\n\t \"url\": \"https:\\\/\\\/www.interactivebrokers.com\\\/campus\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2025\\\/10\\\/blue-stock-graph-quant.jpg\",\n\t \"contentUrl\": \"https:\\\/\\\/www.interactivebrokers.com\\\/campus\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2025\\\/10\\\/blue-stock-graph-quant.jpg\",\n\t \"width\": 1000,\n\t \"height\": 563,\n\t \"caption\": \"Quant\"\n\t },\n\t {\n\t \"@type\": \"WebSite\",\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/#website\",\n\t \"url\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/\",\n\t \"name\": \"IBKR Campus US\",\n\t \"description\": \"Financial Education from Interactive Brokers\",\n\t \"publisher\": {\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/#organization\"\n\t },\n\t \"potentialAction\": [\n\t {\n\t \"@type\": \"SearchAction\",\n\t \"target\": {\n\t \"@type\": \"EntryPoint\",\n\t \"urlTemplate\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/?s={search_term_string}\"\n\t },\n\t \"query-input\": {\n\t \"@type\": \"PropertyValueSpecification\",\n\t \"valueRequired\": true,\n\t \"valueName\": \"search_term_string\"\n\t }\n\t }\n\t ],\n\t \"inLanguage\": \"en-US\"\n\t },\n\t {\n\t \"@type\": \"Organization\",\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/#organization\",\n\t \"name\": \"Interactive Brokers\",\n\t \"alternateName\": \"IBKR\",\n\t \"url\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/\",\n\t \"logo\": {\n\t \"@type\": \"ImageObject\",\n\t \"inLanguage\": \"en-US\",\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/#\\\/schema\\\/logo\\\/image\\\/\",\n\t \"url\": \"https:\\\/\\\/www.interactivebrokers.com\\\/campus\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2024\\\/05\\\/ibkr-campus-logo.jpg\",\n\t \"contentUrl\": \"https:\\\/\\\/www.interactivebrokers.com\\\/campus\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2024\\\/05\\\/ibkr-campus-logo.jpg\",\n\t \"width\": 669,\n\t \"height\": 669,\n\t \"caption\": \"Interactive Brokers\"\n\t },\n\t \"image\": {\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/#\\\/schema\\\/logo\\\/image\\\/\"\n\t },\n\t \"publishingPrinciples\": \"https:\\\/\\\/www.interactivebrokers.com\\\/campus\\\/about-ibkr-campus\\\/\",\n\t \"ethicsPolicy\": \"https:\\\/\\\/www.interactivebrokers.com\\\/campus\\\/cyber-security-notice\\\/\"\n\t },\n\t {\n\t \"@type\": \"Person\",\n\t \"@id\": \"https:\\\/\\\/ibkrcampus.com\\\/campus\\\/#\\\/schema\\\/person\\\/79c2a2775a70a4da1accf0068d731933\",\n\t \"name\": \"Kris Longmore\",\n\t \"url\": \"https:\\\/\\\/www.interactivebrokers.com\\\/campus\\\/author\\\/krislongmore\\\/\"\n\t }\n\t ]\n\t}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"Brave New Backtest | IBKR Quant","description":"The dominant paradigm online is that backtesting and statistics IS research, that finding a pattern in historical data IS finding an edge.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.interactivebrokers.com\/campus\/wp-json\/wp\/v2\/posts\/241054\/","og_locale":"en_US","og_type":"article","og_title":"Brave New Backtest","og_description":"The dominant paradigm online is that backtesting and statistics IS research, that finding a pattern in historical data IS finding an edge.","og_url":"https:\/\/www.interactivebrokers.com\/campus\/ibkr-quant-news\/brave-new-backtest\/","og_site_name":"IBKR Campus US","article_published_time":"2026-04-02T19:44:41+00:00","article_modified_time":"2026-04-02T19:46:47+00:00","og_image":[{"width":1000,"height":563,"url":"https:\/\/www.interactivebrokers.com\/campus\/wp-content\/uploads\/sites\/2\/2025\/10\/blue-stock-graph-quant.jpg","type":"image\/jpeg"}],"author":"Kris Longmore","twitter_card":"summary_large_image","twitter_misc":{"Written by":"Kris Longmore","Est. reading time":"11 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/ibkrcampus.com\/campus\/ibkr-quant-news\/brave-new-backtest\/#article","isPartOf":{"@id":"https:\/\/ibkrcampus.com\/campus\/ibkr-quant-news\/brave-new-backtest\/"},"author":{"name":"Kris Longmore","@id":"https:\/\/ibkrcampus.com\/campus\/#\/schema\/person\/79c2a2775a70a4da1accf0068d731933"},"headline":"Brave New Backtest","datePublished":"2026-04-02T19:44:41+00:00","dateModified":"2026-04-02T19:46:47+00:00","mainEntityOfPage":{"@id":"https:\/\/ibkrcampus.com\/campus\/ibkr-quant-news\/brave-new-backtest\/"},"wordCount":2511,"commentCount":0,"publisher":{"@id":"https:\/\/ibkrcampus.com\/campus\/#organization"},"image":{"@id":"https:\/\/ibkrcampus.com\/campus\/ibkr-quant-news\/brave-new-backtest\/#primaryimage"},"thumbnailUrl":"https:\/\/www.interactivebrokers.com\/campus\/wp-content\/uploads\/sites\/2\/2025\/10\/blue-stock-graph-quant.jpg","keywords":["Artificial Intelligence (AI)","backtesting","Coding Best Practices","Large Language Models (LLMs)","Statistical Analysis"],"articleSection":["Data Science","Quant","Quant Development"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/ibkrcampus.com\/campus\/ibkr-quant-news\/brave-new-backtest\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/ibkrcampus.com\/campus\/ibkr-quant-news\/brave-new-backtest\/","url":"https:\/\/ibkrcampus.com\/campus\/ibkr-quant-news\/brave-new-backtest\/","name":"Brave New Backtest | IBKR Campus US","isPartOf":{"@id":"https:\/\/ibkrcampus.com\/campus\/#website"},"primaryImageOfPage":{"@id":"https:\/\/ibkrcampus.com\/campus\/ibkr-quant-news\/brave-new-backtest\/#primaryimage"},"image":{"@id":"https:\/\/ibkrcampus.com\/campus\/ibkr-quant-news\/brave-new-backtest\/#primaryimage"},"thumbnailUrl":"https:\/\/www.interactivebrokers.com\/campus\/wp-content\/uploads\/sites\/2\/2025\/10\/blue-stock-graph-quant.jpg","datePublished":"2026-04-02T19:44:41+00:00","dateModified":"2026-04-02T19:46:47+00:00","description":"The dominant paradigm online is that backtesting and statistics IS research, that finding a pattern in historical data IS finding an edge.","inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/ibkrcampus.com\/campus\/ibkr-quant-news\/brave-new-backtest\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/ibkrcampus.com\/campus\/ibkr-quant-news\/brave-new-backtest\/#primaryimage","url":"https:\/\/www.interactivebrokers.com\/campus\/wp-content\/uploads\/sites\/2\/2025\/10\/blue-stock-graph-quant.jpg","contentUrl":"https:\/\/www.interactivebrokers.com\/campus\/wp-content\/uploads\/sites\/2\/2025\/10\/blue-stock-graph-quant.jpg","width":1000,"height":563,"caption":"Quant"},{"@type":"WebSite","@id":"https:\/\/ibkrcampus.com\/campus\/#website","url":"https:\/\/ibkrcampus.com\/campus\/","name":"IBKR Campus US","description":"Financial Education from Interactive Brokers","publisher":{"@id":"https:\/\/ibkrcampus.com\/campus\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/ibkrcampus.com\/campus\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/ibkrcampus.com\/campus\/#organization","name":"Interactive Brokers","alternateName":"IBKR","url":"https:\/\/ibkrcampus.com\/campus\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/ibkrcampus.com\/campus\/#\/schema\/logo\/image\/","url":"https:\/\/www.interactivebrokers.com\/campus\/wp-content\/uploads\/sites\/2\/2024\/05\/ibkr-campus-logo.jpg","contentUrl":"https:\/\/www.interactivebrokers.com\/campus\/wp-content\/uploads\/sites\/2\/2024\/05\/ibkr-campus-logo.jpg","width":669,"height":669,"caption":"Interactive Brokers"},"image":{"@id":"https:\/\/ibkrcampus.com\/campus\/#\/schema\/logo\/image\/"},"publishingPrinciples":"https:\/\/www.interactivebrokers.com\/campus\/about-ibkr-campus\/","ethicsPolicy":"https:\/\/www.interactivebrokers.com\/campus\/cyber-security-notice\/"},{"@type":"Person","@id":"https:\/\/ibkrcampus.com\/campus\/#\/schema\/person\/79c2a2775a70a4da1accf0068d731933","name":"Kris Longmore","url":"https:\/\/www.interactivebrokers.com\/campus\/author\/krislongmore\/"}]}},"jetpack_featured_media_url":"https:\/\/www.interactivebrokers.com\/campus\/wp-content\/uploads\/sites\/2\/2025\/10\/blue-stock-graph-quant.jpg","_links":{"self":[{"href":"https:\/\/ibkrcampus.com\/campus\/wp-json\/wp\/v2\/posts\/241054","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ibkrcampus.com\/campus\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ibkrcampus.com\/campus\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ibkrcampus.com\/campus\/wp-json\/wp\/v2\/users\/271"}],"replies":[{"embeddable":true,"href":"https:\/\/ibkrcampus.com\/campus\/wp-json\/wp\/v2\/comments?post=241054"}],"version-history":[{"count":0,"href":"https:\/\/ibkrcampus.com\/campus\/wp-json\/wp\/v2\/posts\/241054\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ibkrcampus.com\/campus\/wp-json\/wp\/v2\/media\/233527"}],"wp:attachment":[{"href":"https:\/\/ibkrcampus.com\/campus\/wp-json\/wp\/v2\/media?parent=241054"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ibkrcampus.com\/campus\/wp-json\/wp\/v2\/categories?post=241054"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ibkrcampus.com\/campus\/wp-json\/wp\/v2\/tags?post=241054"},{"taxonomy":"contributors-categories","embeddable":true,"href":"https:\/\/ibkrcampus.com\/campus\/wp-json\/wp\/v2\/contributors-categories?post=241054"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}