A Financial Brain Scan of the LLM

arXiv:2508.21285 · q-fin.GN, cs.AI, cs.CE, econ.GN, q-fin.EC · Submitted 2025-08-29 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.

Ruth: We're looking at the paper: "A Financial Brain Scan of the LLM".

Warren: Here's the gist of the paper. Emerging techniques in computer science make it possible to “brain scan” large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant This paper shows that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance.

Ruth: Pruning redundant reasoning tokens The method allows researchers to open and control an LLM by inserting an interpretable sparse representation within its architecture.

Warren: Through this representation, a researcher can identify the concepts the model uses to process a given input, expressed in plain English.

Ruth: The array of concepts is extremely large and ranges from optimism or financial risk to fondness for specific cuisines.

Title and authors: Ruth: The paper's title, "A Financial Brain Scan of the LLM," immediately tells us that this is about trying to scan or look into what's happening inside these models concerning finance.

Warren: It suggests they are moving beyond just looking at the final answer and are digging into the reasoning process itself to find those underlying financial concepts.

Iris: They are essentially trying to open up these black boxes by inserting an interpretable sparse representation right in the architecture of the model so researchers can see what's going on.

Ruth: That’s right, they introduce this sparse embedding which encodes all the information flowing through the model and then decodes it back into a representation that is much more readable.

Warren: It sounds like they are essentially giving us a way to open and control an LLM by inserting this layer, which is what page two of the paper talks about when they mention applying techniques from Cunningham et al. in this context.

Iris: They're focusing on the fact that the model's inner representation is dense, meaning it’s hard to interpret, so they are trying to make it sparse and meaningful for economic tasks like these ones (<ref:2508.21285#pg2>).

Ruth: And page one sets up a big problem by pointing out that the scale and density of LLMs make them uninterpretable black boxes for economics and finance, which is why they felt this approach was necessary (<ref:2508.21285#pg1>).

Warren: So, the core idea is using this sparse representation to map forecasts onto concepts like sentiment or timing, and then figuring out which concept actually drives the prediction most.

Iris: And the paper claims they can do that while maintaining performance, which is a big claim because usually when you add interpretability, you expect a hit to accuracy (<ref:2508.21285#pg1>).

Ruth: It seems they are showing that this approach isn't just for curiosity; it’s designed to be transparent and replicable for empirical research in the social sciences (<ref:2508.21285#pg1>).

The paper's summary: Ruth: To summarize what they did, this paper introduces a method where you insert an interpretable sparse representation into an already trained LLM's architecture to allow you to identify the plain-English concepts guiding their reasoning.

Warren: They build this by training a sparse autoencoder which is essentially an encoder-decoder structure that takes all the information flowing through the model and encodes it into this representation before decoding it back.

Iris: This autoencoder is trained specifically on the residual streams produced at a fixed layer in the language model, and it’s loss function encourages both accurate reconstruction and sparsity in that hidden representation (<ref:2508.21285#pg3>).

Ruth: So, the key takeaway here is that this sparse autoencoder approach provides both interpretability and performance by mapping any prompt through an LLM to a semantically rich sparse representation (<ref:2508.21285#pg4>).

Warren: And when they look at what's most important, they found that the most relevant feature groups for financial applications are sentiment, closely followed by market and finance concepts and technical analysis ideas (<ref:2508.21285#pg4>).

Iris: They also show how you can steer the model by identifying a relevant feature in this sparse representation, figuring out what perturbation would happen if that feature were more strongly activated, and adding that to the residual stream to get a steered probability distribution (<ref:2508.21285#pg3>).

Ruth: And they show this steering can be used practically to simulate agents with specific preferences without needing to retrain the base model by just repeating prompts, which is really useful for modeling political economy or psychology (<ref:2508.21285#pg4>).

The paper's improvements: Ruth: The paper points out that this sparse autoencoder method substantially outperforms some recent methods presented in other work, specifically Chen et al. in two thousand twenty-two (ref:Chen et al <ref:2508.21285#pg1>., two thousand twenty-four) <ref:2508.21285#pg1>.

Warren: And the results show that when you use this sparse autoencoder approach, any prompt passed through an LLM can be mapped to a semantically rich sparse representation that you can then apply in the framework from Gentzkow et al. in two thousand nineteen (ref:Gentzkow et al <ref:2508.21285#pg1>., two thousand nineteen) <ref:2508.21285#pg1>.

Iris: The paper highlights that they found that off-the-shelf LLMs are often overly optimistic for trading applications, and reducing positivity actually improves portfolio performance when you steer the model slightly negatively (ref:Figure six).

Ruth: That is a really practical improvement because it gives researchers a way to mitigate documented biases, like those seen in Fedyk et al. in two thousand twenty-four or even deliberately amplify bias for simulation purposes (Figure six).

Warren: And they also found that the dominant driver of short-horizon predictability isn't just any feature, but a small set of sentiment-like features that explains most of the signal even when you aggressively reduce dimensionality (ref:Table III).

Iris: They also show how steering can be used to simulate agents with configurable preferences across political economy and psychology where retraining is costly, which expands the applicability way beyond just finance (ref:Figure six).

Conclusion: Ruth: So, to wrap up, this paper on "A Financial Brain Scan of the LLM" presents a simple, scalable way to open and control large language models by inserting this sparse layer between the model's hidden states and its outputs.

Warren: The main points are that concept steering works when applied to firm-level news, showing that off-the-shelf LLMs lean toward optimism for trading tasks.

Iris: It also clarifies the drivers of short-horizon predictability by showing that a small set of sentiment-like features explains most of the signal even with aggressive dimensionality reduction (<ref:2508.21285#pg4>).

Ruth: This framework extends beyond finance to help us simulate agents with specific dispositions, preferences, and contextual sensitivities where retraining is too expensive for those kinds of controlled variations (ref:Figure six).

Warren: It’s a simple tool that allows you to correct biases by selecting an appropriate steering level for a given feature, or even deliberately amplify bias for research simulations.

Iris: While this technique is still in its early stages, it could evolve into a standard tool for building interpretable, field-ready LLMs across all the social sciences (<ref:2508.21285#pg4>).

MIT Sloan and NBER · University of Melbourne · Yale School of Management and Organization (SOM) · ESSEC Business School

q-fin.GN, cs.AI, cs.CE, econ.GN, q-fin.EC

Submitted: 2025-08-29

Updated: 2026-11-04

Comments: 82 pages

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 88/100

The gist: Emerging techniques in computer science make it possible to “brain scan” large language models (LLMs), identify the plain-English concepts that guide their reasoning…

Key concepts

Sparse Auto-Encoder (SAE)
A Sparse Auto-Encoder is a neural network trained to compress and then reconstruct information in a very sparse way. It learns an interpretable representation of the input data by forcing the hidden layer to be sparse, meaning only a few features are activated at any given time. This creates a structured way to look inside the LLM's internal workings.
Concept Steering
Concept steering is a control mechanism that allows researchers to influence an LLM's output by identifying and adjusting specific concepts within its sparse representation. By perturbing the residual stream based on a targeted feature, users can manually regulate how strongly the model 'thinks' about that concept, enabling fine-grained control over its reasoning.
Sentiment Steering
Sentiment steering uses a steered feature linked to risk aversion or attention to wealth to simulate agents with configurable preferences. By adjusting this steering coefficient, researchers can observe monotonic shifts in behavior, such as making the model less optimistic for trading applications and improving portfolio performance.

Terminology

Summary

Emerging techniques in computer science make it possible to “brain scan” large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant This paper shows that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance. p. 0

Where the paper says it

A Financial Brain Scan of the LLM Hui Chen∗ , Antoine Didisheim† , Mohammad (Mo) Pourmohammadi‡ , Luciano Somoza§ , Hanqing Tian¶ February, 2026 Abstract Emerging techniques in computer science make it possible to “brain scan” large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant. We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance.

Pruning redundant reasoning tokens

The method allows researchers to open and control an LLM by inserting an interpretable sparse representation within its architecture. Through this representation, a researcher can identify the concepts the model uses to process a given input, expressed in plain English. p. 1

Where the paper says it

The technique allows to open and control an LLM by inserting an interpretable sparse representation within its architecture. Through this representation, a researcher can identify the concepts the model uses to process a given input, expressed in plain English.

The array of concepts is extremely large and ranges from optimism or financial risk to fondness for specific cuisines. p. 1
Where the paper says it

Through this representation, a researcher can identify the concepts the model uses to process a given input, expressed in plain English. The array of concepts is extremely large and ranges from optimism or financial risk to fondness for specific cuisines.

Furthermore, by adjusting the code, we can manually regulate the intensity at which the model “thinks” about any targeted concept while keeping everything else equal. p. 1
Where the paper says it

The array of concepts is extremely large and ranges from optimism or financial risk to fondness for specific cuisines. Furthermore, by adjusting the code, we can manually regulate the intensity at which the model “thinks” about any targeted concept while keeping everything else equal.

How the training signal is built

The technique introduces a sparse embedding within an already trained LLM. All information flowing through the model is encoded into this representation and subsequently decoded. p. 2

Where the paper says it

The technique used in this paper solves this problem by introducing a sparse embedding within an already trained LLM. All information flowing through the model is encoded into this representation and subsequently decoded.

This encoder-decoder structure, referred to as a Sparse Auto-Encoder (SAE), is trained to be benign and not impact the LLMs’ quality. p. 2
Where the paper says it

All information flowing through the model is encoded into this representation and subsequently decoded. This encoder-decoder structure, referred to as a Sparse Auto-Encoder (SAE), is trained to be benign and not impact the LLMs’ quality.

SAEs are trained on the residual streams produced at a fixed layer l in the language model. p. 8
Where the paper says it

Second, the residual streams are not sparse, and they occupy a high dimensional, entangled latent space in which individual coordinates lack clear semantic interpretation. In the human brain, different regions often specialize for distinct tasks such as language, vision, or motor control, making it easier to associate structure with function.

The loss function encourages both accurate reconstruction and sparsity in the hidden representation. p. 9
Where the paper says it

The decoder then attempts to reconstruct the original input from z: xˆ = g(Wdz + bd), (4) where Wd ∈ R d×k and bd ∈ R d are decoder parameters, and g(·) is usually chosen to be the identity function or a sigmoid, depending on the domain of the input data (Goodfellow, Bengio, Courville, and Bengio, 2016). The encoder and decoder are trained jointly to minimize the reconstruction error between the input x and its reconstruction xˆ: LAE(x) = ∥x − xˆ∥ 2 2 | {z } Reconstruction loss .

Where the gains come from

The sparse autoencoder approach provides both interpretability and performance. Using sparse autoencoders, any prompt passed through an LLM can be mapped to a semantically rich sparse representation, which can then be applied in step (i) of the Gentzkow et al. p. 3

Where the paper says it

The methodology proposed in this paper provides both interpretability and performance. Using sparse auto-encoders, any prompt passed through an LLM can be mapped to a semantically rich sparse representation, which can then be applied in step (i) of the Gentzkow et al.

(2019) framework. p. 3
Where the paper says it

Using sparse auto-encoders, any prompt passed through an LLM can be mapped to a semantically rich sparse representation, which can then be applied in step (i) of the Gentzkow et al. (2019) framework.

This approach substantially outperforms recent methods presented in Chen et al. p. 3
Where the paper says it

They use the last layer of LLMs as embeddings and show that this representation outperforms all previous textual methods in the finance literature. To compare our reduced sparse embeddings with the last-layer embeddings from the same LLM, we follow Chen et al.

(2022). The resulting embeddings are extremely high-dimensional, with many features irrelevant for financial applications and the most important feature group is Sentiment, closely followed by Market/Finance and Technical Analysis concepts. p. 5
Where the paper says it

Comparing these models with the full feature model, which jointly incorporates all 17 groups, allows us to quantify the contribution of individual financial concepts to LLM performance in portfolio applications. Unsurprisingly, the most important feature group is Sentiment, closely followed by Market/Finance and Technical Analysis concepts.

Concept Steering

Steering can be understood as follows: 1. Identify a relevant feature in the sparse representation to steer the LLM toward. p. 6

Where the paper says it

For this paper’s interpretability goal, representation learning with autoencoders can summarize risk exposures while remaining economically transparent. This approach applies that idea by replacing dense, opaque LLM embeddings with a sparse, interpretable feature set, then using those features to steer the model’s output.

2. Use the SAE’s decoder to identify the perturbation induced by activating this feature; specifically, what would have been added to the residual stream if the feature were more strongly activated. p. 8
Where the paper says it

To address this, we seek to discover a sparse and semantically meaningful representation of the residual stream. Specifically, we aim to identify a transformation: r (l) 7→ z (l) ∈ R k where each coordinate of z (l) corresponds to an interpretable feature or concept.

3. Add this perturbation to the residual stream to obtain a steered probability distribution over the next tokens. p. 8
Where the paper says it

With m total blocks, the final residual stream r (m) is passed to an output module that maps it to a probability distribution over the vocabulary. This distribution represents the model’s prediction for the next word: P(Next Word | Input Text) = Block(Output)(r (m) ).

This approach provides a fine-grained, concept-level control mechanism grounded in interpretable internal features. p. 6
Where the paper says it

We show that this method can make LLMs transparent for economic analysis, allowing researchers to identify the concepts driving predictions without reducing performance. Second, we find that these interpretable features enable precise control over model behavior (such as sentiment or risk aversion), providing a practical approach to correct biases, simulate preferences, and design controlled experiments.

Applied Steering and Bias Correction

Steering features linked as risk aversion and attention to wealth can be used to simulate agents with configurable preferences without retraining the base model. Repeated prompting yields monotonic shifts in behavior consistent with the intended steering direction, providing a practical mechanism to simulate agents with configurable preferences without retraining the base model. p. 6

Where the paper says it

Repeated prompting yields monotonic shifts in behavior consistent with the intended steering direction: stronger risk aversion decreases equity allocations and increases the share of safe choices, whereas stronger attention to wealth produces the opposite effect. This provides a practical mechanism to simulate agents with configurable preferences without retraining the base model.

This method helps clarify the drivers of short-horizon predictability by showing that a small set of sentiment-like features explains most of the signal. p. 6
Where the paper says it

and international markets, reporting sizable gains over bag-of-words and other NLP baselines. Building on that work, this paper retains predictive power but replaces dense embeddings with a sparse, interpretable feature set, showing that most of the signal loads on refined sentiment.

The results indicate that off-the-shelf LLMs are overly optimistic for trading applications, and reducing positivity improves portfolio performance with the highest Sharpe ratio at a modest negative steering level. This flexibility supports applications in political economy, psychology, and sociology where controlled variation in dispositions is useful and retraining is costly. p. 6
Where the paper says it

Second, we find that these interpretable features enable precise control over model behavior (such as sentiment or risk aversion), providing a practical approach to correct biases, simulate preferences, and design controlled experiments. Together, these findings offer a lightweight, replicable method for turning LLMs from opaque systems into useful tools for empirical research in economics and the social sciences.

Conclusion

This paper presents a simple, scalable approach to opening and controlling large language models by inserting a sparse, interpretable layer between the model’s hidden states and its outputs. Concept steering performs as intended when applied to firm-level news, showing that off-the-shelf LLMs are overly optimistic for trading applications. p. 5

Where the paper says it

Steering and Bias Correction Having shown how sparse autoencoders increase the transparency of LLMs, we next highlight the second main advantage: they allow us to force the model to incorporate a specific concept, with a chosen intensity, when processing an input. We start by building a long-short trading strategy based on prompt-extracted sentiment from news in the spirit of Lopez-Lira and Tang (2023a) and Chen et al.

This method also helps us clarifying the drivers of short-horizon predictability by showing that a small set of sentiment-like features explains most of the signal, even under aggressive dimensionality reduction. p. 6
Where the paper says it

and international markets, reporting sizable gains over bag-of-words and other NLP baselines. Building on that work, this paper retains predictive power but replaces dense embeddings with a sparse, interpretable feature set, showing that most of the signal loads on refined sentiment.

The framework extends beyond finance to simulate agents with specific dispositions, preferences, and contextual sensitivities. p. 6
Where the paper says it

Repeated prompting yields monotonic shifts in behavior consistent with the intended steering direction: stronger risk aversion decreases equity allocations and increases the share of safe choices, whereas stronger attention to wealth produces the opposite effect. This provides a practical mechanism to simulate agents with configurable preferences without retraining the base model.

While this technique is still in its infancy, concept steering could evolve into a standard tool for building interpretable, field-ready LLMs for research and practice with countless applications across all social sciences. p. 1
Where the paper says it

Furthermore, by adjusting the code, we can manually regulate the intensity at which the model “thinks” about any targeted concept while keeping everything else equal. The capability of observing an LLM’s “mental” process and manually steering it towards any direction are absent in current financial research and can be valuable for applications across all social sciences.

References

Aguilera, Ana M Manuel Escabias, and Mariano J Valderrama, 2006 Using principal components for estimating logistic regression with high-dimensional multicollinear data, Computational Statistics & Data Analysis 50, 1905–1924.

Aher Gati Rosa I Arriaga, and Adam Tauman Kalai, 2023 Using large language models to simulate multiple humans and replicate human subject studies, in Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Research (PMLR). p. 7

Where the paper says it

Horton (2023) argues that LLMs can serve as “homo silicus” in economic simulations. Aher, Arriaga, and Kalai (2023) formalize “Turing Experiments” and show that recent models replicate classic findings in behavioral economics and psychology.

Argyle Lisa P Ethan C Busby Nancy Fulda Joshua R Gubler Christopher Rytting, and David Wingate, 2023 Out of one, many: Using language models to simulate human samples, Political Analysis 31, 337–351. p. 7

Where the paper says it

Aher, Arriaga, and Kalai (2023) formalize “Turing Experiments” and show that recent models replicate classic findings in behavioral economics and psychology. Argyle, Busby, Fulda, Gubler, Rytting, and Wingate (2023) demonstrate that conditioning LLMs on demographics can reproduce group-level survey response distributions.

Bybee Leland Bryan Kelly Asaf Manela, and Dacheng Xiu, 2024 Business news and business cycles, The Journal of Finance 79, 3105–3147. p. 6

Where the paper says it

The design also complements prompt-based methods that classify news directly (Lopez-Lira and Tang, 2023b), and it follows best-practice evaluation to avoid training-cutoff and look-ahead errors (Sarkar and Vafa, 2024). It also connects to Bybee, Kelly, Manela, and Xiu (2020) by 6

Bybee Leland Bryan T Kelly Asaf Manela, and Dacheng Xiu, 2020 The structure of economic news, SSRN Working Paper No. p. 6

Where the paper says it

The design also complements prompt-based methods that classify news directly (Lopez-Lira and Tang, 2023b), and it follows best-practice evaluation to avoid training-cutoff and look-ahead errors (Sarkar and Vafa, 2024). It also connects to Bybee, Kelly, Manela, and Xiu (2020) by 6

3522305.

Chen Hui Antoine Didisheim, and Luciano Somoza, 2024 Out of the (black) box: Ai as conditional probability, Available at SSRN. p. 0

Where the paper says it

A Financial Brain Scan of the LLM Hui Chen∗ , Antoine Didisheim† , Mohammad (Mo) Pourmohammadi‡ , Luciano Somoza§ , Hanqing Tian¶ February, 2026 Abstract Emerging techniques in computer science make it possible to “brain scan” large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant. We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance.

Chen Yifei Bryan T Kelly, and Dacheng Xiu, 2022 Expected returns and large language models, SSRN Working Paper No. p. 6

Where the paper says it

The most related study is Chen et al. (2022), who use contextualized news embeddings from large language models to forecast next-day stock returns in U.

4543999.

Cheng Qiang Pengkai Lin, and Yue Zhao, 2024 Does generative ai facilitate investor trading? p. 1

Where the paper says it

Introduction Research on artificial intelligence is growing rapidly. Large language models (LLMs) are already part of financial analysis, research workflows, and trading strategies (Cheng, Lin, and Zhao, 2024).

evidence from chatgpt outages, Available at SSRN 4872189. p. 5
Where the paper says it

First, LLM forecasts are biased toward positive sentiment. While this finding may not generalize to all models, it is consistent with evidence 5We follow the approach proposed by Jobson and Korkie (1981), incorporating the correction noted in Memmel (2003), to test the significance of our results.

Cunningham Hoagy Aidan Ewart Logan Riggs Robert Huben, and Lee Sharkey, 2023 Sparse autoencoders find highly interpretable features in language models, arXiv preprint arXiv:2309. p. 1

Where the paper says it

, Cunningham, Ewart, Riggs, Huben, and Sharkey, 2023; Gao, 2024; Shi, 2025) to economic tasks. The technique allows to open and control an LLM by inserting an interpretable sparse representation within its architecture.

08600.

Didisheim Antoine Martina Fraschini, and Luciano Somoza, 2025a Ai’s predictable memory and its implications for finance, SSRN Working Paper No. p. 0

Where the paper says it

A Financial Brain Scan of the LLM Hui Chen∗ , Antoine Didisheim† , Mohammad (Mo) Pourmohammadi‡ , Luciano Somoza§ , Hanqing Tian¶ February, 2026 Abstract Emerging techniques in computer science make it possible to “brain scan” large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant. We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance.

5331177.

Didisheim Antoine Martina Fraschini Luciano Somoza, and Hanqing Tian, 2025b Is AI reasoning useful in finance? p. 0

Where the paper says it

A Financial Brain Scan of the LLM Hui Chen∗ , Antoine Didisheim† , Mohammad (Mo) Pourmohammadi‡ , Luciano Somoza§ , Hanqing Tian¶ February, 2026 Abstract Emerging techniques in computer science make it possible to “brain scan” large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant. We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance.

, Unpublished Working Paper. p. 1
Where the paper says it

Thus, it is essential to develop and apply methods that increase transparency and align model behavior with the researchers’ objectives. This paper applies an emerging technique in computer science (see, e.

Didisheim Antoine Shikun Barry Ke Bryan T Kelly, and Semyon Malamud, 2024 Apt or “aipt”? p. 4

Where the paper says it

We repeat the prediction exercise—training logistic regressions on a rolling window and building a long-short strategy from the forecasts—using subsets of the most important features. Consistent with the notion of “Virtue of Complexity” Kelly, Malamud, and Zhou (2024), adding features always increases the Sharpe ratio.

the surprising dominance of large factor models, Technical report, National Bureau of Economic Research. p. 0
Where the paper says it

A Financial Brain Scan of the LLM Hui Chen∗ , Antoine Didisheim† , Mohammad (Mo) Pourmohammadi‡ , Luciano Somoza§ , Hanqing Tian¶ February, 2026 Abstract Emerging techniques in computer science make it possible to “brain scan” large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant. We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance.

et al., Vaswani, 2017 Attention is all you need, Advances in neural information processing systems 30. p. 7

Where the paper says it

g Vaswani et al. , 2017; Radford et al.

Fatahi Somayeh Julita Vassileva, and Chanchal K Roy, 2024 Comparing emotions in chatgpt answers and human answers to the coding questions on stack overflow, Frontiers in Artificial Intelligence 7, 1393903. p. 6

Where the paper says it

in the literature (Fatahi, Vassileva, and Roy, 2024). Second, steering can be used to correct this bias.

Fedyk Anastassia Ali Kakhbod Peiyao Li, and Ulrike Malmendier, 2024 Chatgpt and perception biases in investments: An experimental study, Available at SSRN 4787249. p. 1

Where the paper says it

First, LLMs’ scale and density make them uninterpretable black boxes, limiting their usefulness for research (Ludwig, Mullainathan, and Rambachan, 2025). Second, LLMs contain hard-to-identify biases, for instance tilting output toward specific demographic preferences (Fedyk, Kakhbod, Li, and Malmendier, 2024).

Gabaix Xavier Ralph SJ Koijen Robert Richmond, and Motohiro Yogo, 2024 Asset embeddings, Available at SSRN 4507511.

Gao Leo et al. p. 1

Where the paper says it

This paper applies an emerging technique in computer science (see, e. , Cunningham, Ewart, Riggs, Huben, and Sharkey, 2023; Gao, 2024; Shi, 2025) to economic tasks.

, 2024 Scaling and evaluating sparse autoencoders, arXiv preprint arXiv:2406. p. 0
Where the paper says it

tian1@unimelb. au 1 arXiv:2508.

04093.

Gemini, 2025 Gemini: A family of highly capable multimodal models. p. 8

Where the paper says it

(2) This architecture, shared by many state-of-the-art models, has demonstrated remarkable performance across a wide range of NLP benchmarks (e. , OpenAI, 2024; Llama, 2024; Gemini, 2025), yet it remains notoriously difficult to interpret.

Gentzkow Matthew Bryan Kelly, and Matt Taddy, 2019 Text as data, Journal of Economic Literature 57, 535–574. p. 3

Where the paper says it

steering can detect and correct biases in canonical finance applications and how this technique can be applied in broader settings to tune LLM’s preferences and biases. Interpretable LLMs’ embeddings In their seminal review, Gentzkow, Kelly, and Taddy (2019) observe that a large fraction of empirical applications of textual analysis in economics and finance can be summarized in three broad steps: (i) constructing a numerical representation of the text; (ii) mapping these representations to predicted values of an unknown outcome; and (iii) employing the predicted values in subsequent descriptive or causal analyses.

Gu Shihao Bryan T Kelly, and Dacheng Xiu, 2020 Empirical asset pricing via machine learning, Review of Financial Studies 33, 2223–2273. p. 6

Where the paper says it

Related Literature Modern asset-pricing research shows that flexible, high-dimensional representations can improve prediction and interpretation. (2020) find that machine learning methods extract return-relevant structure from large predictor sets.

Gu Shihao Bryan T Kelly, and Dacheng Xiu, 2021 Autoencoder asset pricing models, Journal of Econometrics 222, 429–450. p. 9

Where the paper says it

, Gu, Kelly, and Xiu, 2021). Formally, an autoencoder consists of two components: an encoder and a decoder.

Horton John J. p. 7

Where the paper says it

Furthermore, our paper speaks to the fast-growing line of work that treats LLMs as stand-ins for human subjects. Horton (2023) argues that LLMs can serve as “homo silicus” in economic simulations.

, 2023 Large language models as simulated economic agents: What can we learn from homo silicus? p. 7
Where the paper says it

Horton (2023) argues that LLMs can serve as “homo silicus” in economic simulations. Aher, Arriaga, and Kalai (2023) formalize “Turing Experiments” and show that recent models replicate classic findings in behavioral economics and psychology.

, NBER Working Paper 31122, National Bureau of Economic Research. p. 0
Where the paper says it

The method is transparent, lightweight, and replicable for empirical research in the social sciences. ∗MIT Sloan and NBER.

Jobson J Dave and Bob M Korkie, 1981 Performance hypothesis testing with the sharpe and treynor measures, Journal of Finance 889–908. p. 5

Where the paper says it

First, LLM forecasts are biased toward positive sentiment. While this finding may not generalize to all models, it is consistent with evidence 5We follow the approach proposed by Jobson and Korkie (1981), incorporating the correction noted in Memmel (2003), to test the significance of our results.

Kelly Bryan Semyon Malamud, and Kangying Zhou, 2024 The virtue of complexity in return prediction, The Journal of Finance 79, 459–503. p. 4

Where the paper says it

We repeat the prediction exercise—training logistic regressions on a rolling window and building a long-short strategy from the forecasts—using subsets of the most important features. Consistent with the notion of “Virtue of Complexity” Kelly, Malamud, and Zhou (2024), adding features always increases the Sharpe ratio.

Lieberum Tom et al. p. 4

Where the paper says it

Each feature in a sparse representation corresponds to a specific semantic meaning. In this paper, we use the pre-trained and labeled sparse model released by Google DeepMind (Lieberum et al.

, 2024 Gemma scope: Open sparse autoencoders everywhere all at once on gemma 2, arXiv preprint arXiv:2408. p. 0
Where the paper says it

tian1@unimelb. au 1 arXiv:2508.

05147.

Liu Xiao Zirui Wu Xueqing Wu Pan Lu Kai-Wei Chang, and Yansong Feng, 2024 Are llms capable of data-based statistical and causal reasoning? benchmarking advanced quantitative reasoning with data, arXiv preprint arXiv:2402. p. 0

Where the paper says it

tian1@unimelb. au 1 arXiv:2508.

17644.

Llama, 2024 The llama 3 herd of models. p. 8

Where the paper says it

(2) This architecture, shared by many state-of-the-art models, has demonstrated remarkable performance across a wide range of NLP benchmarks (e. , OpenAI, 2024; Llama, 2024; Gemini, 2025), yet it remains notoriously difficult to interpret.

Lopez-Lira Alejandro, and Yuehua Tang, 2023a Can chatgpt forecast stock price movements? p. 2

Where the paper says it

We divide our analysis in two equally important parts. First, we focus on interpretability and use the sparse embedding as input to a forecasting model, to understand what generates the LLM-driven excess returns documented in recent papers (Chen, Kelly, and Xiu, 2022; Lopez-Lira and Tang, 2023a; Chen, Didisheim, and Somoza, 2024).

return predictability and large language models, arXiv preprint arXiv:41>. p. 0
Where the paper says it

tian1@unimelb. au 1 arXiv:2508.

p. 0
Where the paper says it

tian1@unimelb. au 1 arXiv:2508.

Lopez-Lira Alejandro, and Yuehua Tang, 2023b Can chatgpt forecast stock price movements? p. 6

Where the paper says it

Building on that work, this paper retains predictive power but replaces dense embeddings with a sparse, interpretable feature set, showing that most of the signal loads on refined sentiment. The design also complements prompt-based methods that classify news directly (Lopez-Lira and Tang, 2023b), and it follows best-practice evaluation to avoid training-cutoff and look-ahead errors (Sarkar and Vafa, 2024).

return predictability and large language models, arXiv preprint arXiv:41>.

Loughran Tim Bill McDonald, 2011 When is a liability not a liability? p. 3

Where the paper says it

The first step (numerical representation) entails a trade-off between interpretability and efficiency. Simpler methods, such as dictionary-based approaches and bag-of-words models, provide highly transparent and interpretable representations (Tetlock, 2007; Loughran and McDonald, 2011).

textual analysis, dictionaries, and 10-ks, The Journal of finance 66, 35–65. p. 3
Where the paper says it

steering can detect and correct biases in canonical finance applications and how this technique can be applied in broader settings to tune LLM’s preferences and biases. Interpretable LLMs’ embeddings In their seminal review, Gentzkow, Kelly, and Taddy (2019) observe that a large fraction of empirical applications of textual analysis in economics and finance can be summarized in three broad steps: (i) constructing a numerical representation of the text; (ii) mapping these representations to predicted values of an unknown outcome; and (iii) employing the predicted values in subsequent descriptive or causal analyses.

Ludwig Jens Sendhil Mullainathan and Ashesh Rambachan, 2025 Large language models: An applied econometric framework, Technical report, National Bureau of Economic Research. p. 1

Where the paper says it

However, there are two main concerns for researchers in economics and finance. First, LLMs’ scale and density make them uninterpretable black boxes, limiting their usefulness for research (Ludwig, Mullainathan, and Rambachan, 2025).

McQueen James B, 1967 Some methods of classification and analysis of multivariate observations, in Proc. of 5th Berkeley Symposium on Math. Stat. and Prob., 281–297.

Memmel Christoph, 2003 Performance hypothesis testing with the sharpe ratio, Available at SSRN 412588. p. 5

Where the paper says it

First, LLM forecasts are biased toward positive sentiment. While this finding may not generalize to all models, it is consistent with evidence 5We follow the approach proposed by Jobson and Korkie (1981), incorporating the correction noted in Memmel (2003), to test the significance of our results.

OpenAI 2024 Gpt-4 technical report. p. 8

Where the paper says it

(2) This architecture, shared by many state-of-the-art models, has demonstrated remarkable performance across a wide range of NLP benchmarks (e. , OpenAI, 2024; Llama, 2024; Gemini, 2025), yet it remains notoriously difficult to interpret.

Park Joon Sung Joseph C O’Brien Carrie J Cai Meredith Ringel Morris Percy Liang, and Michael S Bernstein, 2023 Generative agents: Interactive simulacra of human behavior, in Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology (UIST ’23) (ACM). p. 7

Where the paper says it

Argyle, Busby, Fulda, Gubler, Rytting, and Wingate (2023) demonstrate that conditioning LLMs on demographics can reproduce group-level survey response distributions. In HCI, Park, O’Brien, Cai, Morris, Liang, and Bernstein (2023) build multi-agent, memory-based “generative agents” that exhibit believable individual and social behavior.

Radford Alec et al. p. 7

Where the paper says it

g Vaswani et al. , 2017; Radford et al.

, 2019 Language models are unsupervised multitask learners, Advances in neural information processing systems 30. p. 9
Where the paper says it

Sparse Autoencoders (SAEs) Autoencoders are a class of neural network models designed for unsupervised dimensionality reduction. Like Principal Component Analysis (PCA), they aim to compress input data into a lower-dimensional latent representation which minimizes the reconstruction error.

Rousseeuw Peter J, 1987 Silhouettes: a graphical aid to the interpretation and validation of cluster analysis, Journal of computational and applied mathematics 20, 53–65. p. 4

Where the paper says it

4 The methodology produces 17 clusters with distinct economic interpretation. This separation enables the construction of two model types and their corresponding long–short portfolios: (i) models trained exclusively on features from a single group, and (ii) models trained on all but one group, following the Shapley value framework (Gu, Kelly, and 2For both models, we use logistic regression, trained on a three-year rolling window with an additional one-year validation set and re-estimated annually.

Sarkar Suproteem K, 2025 Economic representations. p. 3

Where the paper says it

(2022) demonstrate that complex, opaque embeddings generated by modern machine learning methods (see, e. , Sarkar, 2025) can significantly outperform simpler representations in important financial prediction tasks.

Sarkar Suproteem K. p. 3

Where the paper says it

(2022) demonstrate that complex, opaque embeddings generated by modern machine learning methods (see, e. , Sarkar, 2025) can significantly outperform simpler representations in important financial prediction tasks.

, and Keyon Vafa, 2024 Lookahead bias in pretrained language models, SSRN Working Paper No. p. 6
Where the paper says it

Building on that work, this paper retains predictive power but replaces dense embeddings with a sparse, interpretable feature set, showing that most of the signal loads on refined sentiment. The design also complements prompt-based methods that classify news directly (Lopez-Lira and Tang, 2023b), and it follows best-practice evaluation to avoid training-cutoff and look-ahead errors (Sarkar and Vafa, 2024).

4754678.

Shi Wei et al. p. 1

Where the paper says it

This paper applies an emerging technique in computer science (see, e. , Cunningham, Ewart, Riggs, Huben, and Sharkey, 2023; Gao, 2024; Shi, 2025) to economic tasks.

, 2025 Route sparse autoencoder to interpret large language models, arXiv preprint arXiv:2503. p. 8
Where the paper says it

(2) This architecture, shared by many state-of-the-art models, has demonstrated remarkable performance across a wide range of NLP benchmarks (e. , OpenAI, 2024; Llama, 2024; Gemini, 2025), yet it remains notoriously difficult to interpret.

08200.

Tetlock Paul C, 2007 Giving content to investor sentiment: The role of media in the stock market, The Journal of finance 62, 1139–1168. p. 5

Where the paper says it

Comparing these models with the full feature model, which jointly incorporates all 17 groups, allows us to quantify the contribution of individual financial concepts to LLM performance in portfolio applications. Unsurprisingly, the most important feature group is Sentiment, closely followed by Market/Finance and Technical Analysis concepts.

Vaswani Ashish et al. p. 7

Where the paper says it

This layered design lets the model gradually learn more complex language patterns as information moves up through the layers. g Vaswani et al.

, 2017 Attention is all you need. p. 7
Where the paper says it

g Vaswani et al. , 2017; Radford et al.

Wang Liang Nan Yang Xiaolong Huang Linjun Yang Rangan Majumder, and Furu Wei, 2023 Improving text embeddings with large language models, arXiv preprint arXiv:41>.

Appendix A transparently reports the choices made to merge the 25 clusters into 17 groups and provides the final naming of each group

The choice of 25 clusters is guided by the maximization of the silhouette coefficient (Rousseeuw, 1987), consistent with standard practice in the literature. The results yielded the following classification: • 0, 3, 11, 22 → Punctuation and Symbols • 1, 18 → Healthcare • 2 → Sentiment • 4 → Finance/Markets • 6 → Technology • 7 → Temporal Concepts • 8 → Scientific Jargon • 9 → Finance/Corporate • 12 → Legal • 5, 13 → Technical Analysis ●. p. 5

Where the paper says it

Comparing these models with the full feature model, which jointly incorporates all 17 groups, allows us to quantify the contribution of individual financial concepts to LLM performance in portfolio applications. Unsurprisingly, the most important feature group is Sentiment, closely followed by Market/Finance and Technical Analysis concepts.

Appendix B Neural Network Predictions

Table VI reports the return predictions and number of features. The Shapley Sharpe captures the incremental predictive power unique to a given cluster, in the sense that the logistic predictor cannot replicate the same information from correlated features in other clusters. p. 4

Where the paper says it

4 The methodology produces 17 clusters with distinct economic interpretation. This separation enables the construction of two model types and their corresponding long–short portfolios: (i) models trained exclusively on features from a single group, and (ii) models trained on all but one group, following the Shapley value framework (Gu, Kelly, and 2For both models, we use logistic regression, trained on a three-year rolling window with an additional one-year validation set and re-estimated annually.

The top three clusters—Sentiment, Finance/Markets, and Technical Analysis—capture fundamental concepts in finance and portfolio construction as reflected in news. p. 5
Where the paper says it

Comparing these models with the full feature model, which jointly incorporates all 17 groups, allows us to quantify the contribution of individual financial concepts to LLM performance in portfolio applications. Unsurprisingly, the most important feature group is Sentiment, closely followed by Market/Finance and Technical Analysis concepts.

The next most important cluster, however, is less expected: temporal concepts. p. 0
Where the paper says it

We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance. We also show that models can be steered to be more or less risk-averse, optimistic, or pessimistic, which allows researchers to correct or simulate biases.

p. 0
Where the paper says it

We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance. We also show that models can be steered to be more or less risk-averse, optimistic, or pessimistic, which allows researchers to correct or simulate biases.

This cluster exhibits a statistically significant Shapley Sharpe of 0. p. 5
Where the paper says it

In the spirit of Lopez-Lira and Tang (2023a), we use these forecasts to construct intra-day long–short portfolios, going long (equally weighted) on stocks with positive sentiment and short on those with negative sentiment. Interestingly, negatively steered predictions achieve a statistically significant higher annualized Sharpe ratio (4.

p. 5
Where the paper says it

In the spirit of Lopez-Lira and Tang (2023a), we use these forecasts to construct intra-day long–short portfolios, going long (equally weighted) on stocks with positive sentiment and short on those with negative sentiment. Interestingly, negatively steered predictions achieve a statistically significant higher annualized Sharpe ratio (4.

41, yet one of the lowest individual Sharpe ratios (1. p. 5
Where the paper says it

Interestingly, time-related features (e. , dates, years, timelines) exhibit the fourth-highest Shapley value but the lowest individual Sharpe ratios, i.

p. 5
Where the paper says it

Interestingly, time-related features (e. , dates, years, timelines) exhibit the fourth-highest Shapley value but the lowest individual Sharpe ratios, i.

27). The dominant driver of such predictions is the directional content of news—whether it conveys positive or negative information about a company. p. 5
Where the paper says it

In the spirit of Lopez-Lira and Tang (2023a), we use these forecasts to construct intra-day long–short portfolios, going long (equally weighted) on stocks with positive sentiment and short on those with negative sentiment. Interestingly, negatively steered predictions achieve a statistically significant higher annualized Sharpe ratio (4.

p. 5
Where the paper says it

In the spirit of Lopez-Lira and Tang (2023a), we use these forecasts to construct intra-day long–short portfolios, going long (equally weighted) on stocks with positive sentiment and short on those with negative sentiment. Interestingly, negatively steered predictions achieve a statistically significant higher annualized Sharpe ratio (4.

Clusters such as Sentiment, Finance/Markets, and Technical Analysis directly capture this dimension. p. 5
Where the paper says it

Comparing these models with the full feature model, which jointly incorporates all 17 groups, allows us to quantify the contribution of individual financial concepts to LLM performance in portfolio applications. Unsurprisingly, the most important feature group is Sentiment, closely followed by Market/Finance and Technical Analysis concepts.

p. 5
Where the paper says it

Comparing these models with the full feature model, which jointly incorporates all 17 groups, allows us to quantify the contribution of individual financial concepts to LLM performance in portfolio applications. Unsurprisingly, the most important feature group is Sentiment, closely followed by Market/Finance and Technical Analysis concepts.

The secondary determinant is the timeframe—whether the news is likely to affect markets in the short or long run. p. 4
Where the paper says it

Features with large absolute loadings strongly affect predictions, while those near zero have little impact. We repeat the prediction exercise—training logistic regressions on a rolling window and building a long-short strategy from the forecasts—using subsets of the most important features.

p. 4
Where the paper says it

Features with large absolute loadings strongly affect predictions, while those near zero have little impact. We repeat the prediction exercise—training logistic regressions on a rolling window and building a long-short strategy from the forecasts—using subsets of the most important features.

Timing features thus add value only in conjunction with sentiment-related features, which is precisely what Table III reflects. p. 5
Where the paper says it

Interestingly, time-related features (e. , dates, years, timelines) exhibit the fourth-highest Shapley value but the lowest individual Sharpe ratios, i.

p. 5
Where the paper says it

Interestingly, time-related features (e. , dates, years, timelines) exhibit the fourth-highest Shapley value but the lowest individual Sharpe ratios, i.

The cluster Punctuation and Symbols exhibits no significant Shapley Sharpe but a very high individual Sharpe of 4. p. 5
Where the paper says it

Interestingly, time-related features (e. , dates, years, timelines) exhibit the fourth-highest Shapley value but the lowest individual Sharpe ratios, i.

65. This result is consistent with the way transformers encode relationships between words and symbols. p. 2
Where the paper says it

In lay terms, just like neuroscientists record which brain regions “light up” when a person sees a face or hears music, here we look at which features in the model activate when it processes certain words or contexts. In both cases, the system is not labeled in advance but the researcher discovers meaning by observing consistent patterns of activation.

Table III Return Predictions and Topics

The top three clusters—Sentiment, Finance/Markets, and Technical Analysis—capture fundamental concepts in finance and portfolio construction as reflected in news. The next most important cluster, however, is less expected: temporal concepts. This cluster exhibits a statistically significant Shapley Sharpe of 0.41, yet one of the lowest individual Sharpe ratios (1.27). The dominant driver of such predictions is the directional content of news—whether it conveys positive or negative information about a company. Clusters such as Sentiment, Finance/Markets, and Technical Analysis directly capture this dimension. The secondary determinant is the timeframe—whether the news is likely to affect markets in the short or long run. Timing features thus add value only in conjunction with sentiment-related features, which is precisely what Table III reflects.

Table IV Validating the Steering Technique

The proportion of positive classifications increases monotonically with the steering coefficient. When the model’s positivity node is activated, it is more likely to predict a positive market reaction. p. 5

Where the paper says it

On the sparse representation, we select a feature linked to “positivity’. For each news item, we obtain a baseline LLM forecast and additional forecasts corresponding to varying levels of positive steering.

The corresponding return patterns are consistent with this behavior: under negative steering, positive classifications are associated higher next day average returns, whereas under positive steering, negative classifications are associated with lower next day average returns. p. 5
Where the paper says it

As expected, the proportion of positive classifications increases monotonically with stronger positive steering and the conditional returns are consistent with the model’s induced interpretation of the same headline. In the spirit of Lopez-Lira and Tang (2023a), we use these forecasts to construct intra-day long–short portfolios, going long (equally weighted) on stocks with positive sentiment and short on those with negative sentiment.

These findings support the conclusion that the steering mechanism operates as intended. p. 6
Where the paper says it

Repeated prompting yields monotonic shifts in behavior consistent with the intended steering direction: stronger risk aversion decreases equity allocations and increases the share of safe choices, whereas stronger attention to wealth produces the opposite effect. This provides a practical mechanism to simulate agents with configurable preferences without retraining the base model.

Figure 6 Comparison of model performance metrics

The baseline Sharpe ratio equals 3.87, while negatively steered portfolios systematically achieve higher ratios (4. p. 5

Where the paper says it

In the spirit of Lopez-Lira and Tang (2023a), we use these forecasts to construct intra-day long–short portfolios, going long (equally weighted) on stocks with positive sentiment and short on those with negative sentiment. Interestingly, negatively steered predictions achieve a statistically significant higher annualized Sharpe ratio (4.

07, 4.11, and 4.28). These findings validate steering as a tool to both identify and mitigate LLM bias. p. 6
Where the paper says it

Second, steering can be used to correct this bias. In principle, this bias-correction framework can be applied to any LLM bias or preference that can be mapped onto the set of features in the sparse representation, either to attenuate or to amplify such tendencies.

By selecting an appropriate steering level for a given feature, one can correct biases such as those documented in Fedyk et al. p. 3
Where the paper says it

steering can detect and correct biases in canonical finance applications and how this technique can be applied in broader settings to tune LLM’s preferences and biases. Interpretable LLMs’ embeddings In their seminal review, Gentzkow, Kelly, and Taddy (2019) observe that a large fraction of empirical applications of textual analysis in economics and finance can be summarized in three broad steps: (i) constructing a numerical representation of the text; (ii) mapping these representations to predicted values of an unknown outcome; and (iii) employing the predicted values in subsequent descriptive or causal analyses.

(2024). Conversely, steering can also be employed to deliberately amplify bias for research purposes—for instance, to simulate varying levels of risk aversion or home bias in human decision-making. p. 6
Where the paper says it

In principle, this bias-correction framework can be applied to any LLM bias or preference that can be mapped onto the set of features in the sparse representation, either to attenuate or to amplify such tendencies. We illustrate this by steering features labeled as risk aversion and attention to wealth, and subsequently prompting the model to choose between safe and risky investments or to allocate a budget between Treasuries and an equity index.

Figure 7 Top 10 features by proximity - Topics 0–5

The choice of 25 clusters is guided by the maximization of the silhouette coefficient (Rousseeuw, 1987), consistent with standard practice in the literature. The results yielded the following classification: • 0, 3, 11, 22 → Punctuation and Symbols • 1, 18 → Healthcare • 2 → Sentiment • 4 → Finance/Markets • 6 → Technology. p. 7

Where the paper says it

, positivity or risk aversion-while holding other latent factors fixed, enabling clean comparative statics in finance tasks and portable designs for the social sciences. Methodology The technology underlying generative models like chat-GPT and similar LLMs, is based on generative architectures that model the probability distribution of text sequences.

Figure 8 Top 10 features by proximity - Topics 6–11

The choice of 25 clusters is guided by the maximization of the silhouette coefficient (Rousseeuw, 1987), consistent with standard practice in the literature. The results yielded the following classification: • 6 → Technology • 7 → Temporal Concepts • 8 → Scientific Jargon • 9 → Finance/Corporate • 12 → Legal. p. 7

Where the paper says it

, positivity or risk aversion-while holding other latent factors fixed, enabling clean comparative statics in finance tasks and portable designs for the social sciences. Methodology The technology underlying generative models like chat-GPT and similar LLMs, is based on generative architectures that model the probability distribution of text sequences.

Figure 9 Top 10 features by proximity - Topics 12–17

The choice of 25 clusters is guided by the maximization of the silhouette coefficient (Rousseeuw, 1987), consistent with standard practice in the literature. The results yielded the following classification: • 12 → Legal • 13 → Technical Analysis • 14 → Governance and Politics • 15 → Energy and Manufacturing. p. 0

Where the paper says it

A Financial Brain Scan of the LLM Hui Chen∗ , Antoine Didisheim† , Mohammad (Mo) Pourmohammadi‡ , Luciano Somoza§ , Hanqing Tian¶ February, 2026 Abstract Emerging techniques in computer science make it possible to “brain scan” large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant. We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance.

Figure 10 Top 10 features by proximity - Topics 18–23

The choice of 25 clusters is guided by the maximization of the silhouette coefficient (Rous…

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Sparse Autoencoder based feature extraction can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing without reducing performance. This allows for identifying the plain-English concepts that guide their reasoning in LLMs.

  2. Steering mechanisms allow researchers to manually regulate the intensity at which the model thinks about any targeted concept while keeping everything else equal. This enables models to be steered to be more or less risk-averse, optimistic, or pessimistic, which allows researchers to correct or simulate biases.

  3. The method provides a fine-grained, concept-level control mechanism grounded in interpretable internal features. This control can be used to simulate agents with configurable preferences across diverse contexts like political economy and psychology.

Abstract

We use sparse autoencoders (SAEs) to "brain scan" large language models, mapping internal activations to interpretable concepts and enabling targeted "steering" of specific features. We show that prompting models to adopt specific preferences activates many unintended concepts, complicating causal inference. In contrast, SAEs permit smooth, more isolated steering of target concepts. Furthermore, our scans reveal a stark disconnect between the model's internal feature contributions and its self-generated explanations, showing that these explanations are unreliable guides to its reasoning. Finally, SAEs provide a lightweight intervention for existing models, producing interpretable embeddings and allowing researchers to attenuate specific economic biases.

Sources

Related papers