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.
Figures & tables
Figure 1: Sparse Autoencoder Representation of the Residual Stream. The left panel shows a stylized four-layer language model. A prompt passes through successive transformer layers to produce a probability distribution over the next token. The red arrows trace the residual stream , the running hidden representation updated and passed from one layer to the next. The right panel shows the same model with an SAE inserted between Layers 2 and 3. The encoder maps the dense residual-stream vector into a substantially wider feature layer. Only a small number of these features are active for a given input, as shown in dark blue. The decoder then reconstructs the vector in its original dimensions and passes it to Layer 3.
Theme
Topics
Corporate Finance
Leverage; Financial Distress; CAPEX; Investments; Revenues growth; Mergers & Acquisitions (M&A); Dividend Policy; Cost of Capital; Working Capital; Inventory; Governance; Capital Budgeting; Human resources
Table I: Taxonomy of Economic and Financial Topics
Prompt 1: Word-Cloud Probe This neutral prompt elicits a short completion from the LLM while its residual stream is steered along the features assigned to a given topic.
Figure 2: Word Clouds for Example Economic Topics
Variable
Corr. (Level)
Corr. (Revision)
Employment
0.277**
0.256**
Federal Gov. Consumption Growth
0.197*
0.064
Housing Starts
0.246**
0.396***
Industrial Production
-0.155
0.356***
Real Consumption Growth
0.418***
0.326***
Real Nonresidential Inv. Growth
0.590***
0.465***
Table II: Expectation Correlation of Macro Variables This table reports the correlation between LLM (Gemma-2-9B-IT) expectations and Survey of Professional Forecasters (SPF) expectations. Following Bybee (2023) , we report the correlation separately for level and revision forecasts. The frequency is quarterly and the sample period is 2000–2021.
Figure 3: Brainscan: Example of a Single News Article.
Prompt 2: Macro-Direction Forecast The prompt asks the LLM to forecast the directional impact of a given news item on a target macroeconomic variable, with the response constrained to a single token from the set {P,U,N} . The brackets {⋯} indicate dynamic inserts where we place the news (text of the news article) and the variable (e.g., the S&P 500, CPI, real GDP).
Figure 4: Brainscan: Aggregation at Economic Concepts Panels A, B, and C report the decompositions of AAA bond rate, real consumption, and real nonresidential investment expectations, respectively. For each variable, the pie chart on the left shows each of the five economic themes’ share of total contribution magnitude, averaged over the sample. The bar chart on the right decomposes the dominant theme into its constituent topics, expressed as shares of that theme’s contribution. Contributions combine SAE feature activations with the gradients of the LLM’s directional assessment, as described in the text.
Prompt 3: Topic-Weighted Macro-Direction Forecast The prompt asks the LLM to forecast the directional impact of a given news item on a target macroeconomic variable while also identifying the topics that are genuinely relevant to the prediction.
Figure 5: Mechanistic versus self-reported topic decomposition. For each macroeconomic variable, we compute the Pearson correlation across the 46 economic topics between the sample-average absolute brain-scan topic contribution, T1∑t∣Ckt∣ , and the sample-average self-reported importance score, T1∑tCkt , as defined in Equation ( 6 ). Each bar reports this correlation for the corresponding forecast target. A higher value indicates greater agreement between the brain scan and the LLM’s self-reports about which topics matter for that target. ***, **, and * denote significance at the 1%, 5%, and 10% level, respectively, based on the Pearson correlation test across the 46 topics.
Prompt 4: Risk Aversion and Investment Choice: S&P500 vs Treasuries The prompt asks the LLM to choose between a higher-risk option, an ETF tracking the S&P500, and a lower-risk option, US Treasuries. We use it to assess whether the model’s choice aligns with expected risk preferences under a forced-choice investment setting.
Figure 6: Prompt Steering versus SAE Steering. The figure illustrates the interventions using the investment choice in Prompt 4 . Panel A shows the unmodified model. Panel B leaves the prompt unchanged and adds a perturbation along the selected risk-aversion feature’s decoder direction to the residual stream. Panel C adds a risk-aversion instruction to the prompt, changing internal representations from the first layer onward. Model weights remain fixed across panels. The four-layer network, activation shadings, and probabilities are illustrative rather than empirical results.
Figure 7: Risk Aversion Steering: Behavioral Responses and Concept Leakage. Both panels use Prompt 4 , which asks the LLM to choose between an S&P500 ETF and US Treasuries. Panel A compares SAE and prompt steering. The vertical axis reports the inner probability of selecting the safer asset, US Treasuries. Panel B reports the ten topics with the largest changes in SAE feature activations when the prompt asks the model to be 100% more risk-averse. Changes are calculated by summing steered minus unsteered feature activations within each topic.
Prompt 5: NPV Dominance, Identical Risk This prompt presents a financial project choice where Project Alpha has a higher Net Present Value than Project Beta, while both projects carry exactly the same risk. Since risk is identical across the two options, risk aversion is irrelevant and the dominant choice is the project with the higher NPV (option 1).
Prompt 6: Factual Bond Price and Interest Rate Relationship This prompt tests whether the model correctly identifies the inverse relationship between bond prices and interest rates. When central bank interest rates increase, existing bond prices generally fall, so the expected answer is option 2.
Figure 8: This figure compares the behavioral and representational effects of risk-aversion steering on deterministic financial prompts. Panels A and B report the change in probability assigned to the expected, correct choice for the NPV dominance prompt and the bond-price factual prompt, respectively. Panels C and D report the top 10 SAE topics with the largest feature changes under prompt steering for the same two prompts. The topic change is computed as the sum of difference in SAE feature activations z between steered and unsteered prompts (steered 100% more risk-averse for prompt 5 and 100% less risk-averse for prompt 6 ). A positive or negative difference indicates that prompt steering caused the model to allocate more or less internal activation to the corresponding topic.
Figure 9: Effect of Quantitative Easing Steering on Treasury Bill Rate Expectations This figure compares LLM-generated expectations for revisions in the 3-month Treasury bill rate with revisions from the SPF. Both panels use the same prompt and news inputs. Panel (a) reports the unsteered forecast, which produces an erroneous rate spike during the 2008 financial crisis. Panel (b) reports the forecast after we increase the activation of the SAE feature most strongly associated with Quantitative Easing . Steering removes the spike and produces a decline that closely tracks the SPF revisions.
Default (Unsteered)
Steered (Quantitative Easing)
Prompt: Here is a piece of news: The US economy is experiencing a crisis because of the housing bubble. Do you think this news will increase or decrease the 3-month treasury bill rate? Please include a very short explanation.
This news would likely increase the 3-month Treasury bill rate. Here’s why: Risk Aversion: A housing crisis signals economic instability, making investors more risk-averse. They flock to safe-haven assets like Treasury bills, increasing demand and pushing up their rates. Let me know if you have any other questions!
This news would likely decrease the 3-month Treasury bill rate. Here’s why: Crisis signals low growth: A housing bubble crisis suggests weak economic conditions and potential recession. Central bank response: To stimulate the economy, central banks (like the Federal Reserve) typically lower interest rates, including short-term rates like Treasury bills. Lower rates encourage borrowing and investment, hoping to boost economic activity. Let me know if you have any other questions!
Table III: Model Response Comparison: Correcting Spurious Logic via SAE Steering.
Prompt 7: News Classification This prompt asks the LLM to classify each news alert as positive or negative for the firm’s stock return. The model must respond with a single letter, P for positive or N for negative. The brackets {⋯} indicate dynamic inserts for the alert’s headline and body .
SAE Steering
Prompt Steering
Avg. Return
Avg. Return
Steering
Pos. Class.
Positive
Negative
Steering
Pos. Class.
Positive
Negative
−100
58.4%
0.0105
−0.0067
−100
0.0%
0.0066
0.0033
−50
61.6%
0.0101
−0.0076
−50
0.2%
0.0282
0.0033
−30
64.2%
0.0098
−0.0083
−30
0.1%
0.0336
0.0033
0
67.6%
0.0095
−0.0094
0
67.6%
0.0095
−0.0094
Table IV: News Classification under SAE and Prompt Steering This table reports news classifications and contemporaneous stock returns under SAE steering and prompt steering. Steering denotes the method-specific intervention strength and is not comparable across panels. Pos. Class. is the percentage of alerts classified as positive, and Average Return is the mean daily stock return conditional on the classification. Both methods use the same LLM and news sample.
Figure 10: SAE Steering Portfolio Performance This figure reports the performance of the news-classification long-short portfolio across SAE steering coefficients. Panel (a) reports the annualized Sharpe ratio. Panel (b) reports alpha relative to the unsteered baseline, estimated as the intercept from a regression of each steered portfolio’s returns on the baseline portfolio’s returns. All portfolios are scaled to 10% annual volatility. Statistical significance is based on HAC standard errors with five lags. ***, **, and * indicate that alpha differs from zero at the 1%, 5%, and 10% levels, respectively, based on two-sided tests.
Figure 11: Collapse in Portfolio Performance under Prompt Steering This figure reports the performance of the news-classification long-short portfolio across prompt-steering strengths. Panel (a) reports the annualized Sharpe ratio. Panel (b) reports alpha relative to the unsteered baseline, estimated as the intercept from a regression of each prompt-steered portfolio’s returns on the baseline portfolio’s returns. All portfolios are scaled to 10% annual volatility. Statistical significance is based on heteroskedasticity- and autocorrelation-consistent (HAC) standard errors with five lags. ***, **, and * indicate that alpha differs from zero at the 1%, 5%, and 10% levels, respectively, based on two-sided tests.
Sparse features
Benchmark
10
50
100
500
1000
3000
5000
ALL
EW Sharpe
3.67
3.44
3.84
4.15
4.50
4.45
4.95
4.78
5.22
VW Sharpe
1.55
1.73
1.71
1.76
1.90
1.92
1.92
1.91
1.99
Accuracy
50.90%
50.92%
51.14%
51.33%
51.38%
51.33%
51.42%
51.38%
51.45%
Table V: Return Predictions and Number of Features This table reports the out-of-sample performance of return-prediction models based on dense final-layer embeddings and sparse SAE representations. The models are re-estimated annually using the preceding four years of data and predict the sign of the subsequent open-to-close return. Each day, firms are ranked by their predicted probabilities of a positive return. The long-short portfolios buy firms in the top 20% and short firms in the bottom 20%. Benchmark denotes the model using standard final-layer embeddings following Chen et al. (2022) . The remaining columns report models using the top N economically labeled SAE features, with All denoting the full set of 7,278 features. EW Sharpe and VW Sharpe report the annualized Sharpe ratios of the equal- and value-weighted portfolios, respectively. Accuracy is the average daily percentage of firm-level predictions that match the sign of the realized return.
Figure 12: Economic Composition of the Most Predictive SAE Features This figure decomposes the top N economically labeled SAE features across five economic themes. Features are ranked by the average absolute magnitude of their fitted coefficients across the annual estimation windows. The horizontal axis reports N , ranging from the 10 highest-ranked features to the full set of 7,278 labeled features. The vertical axis reports the percentage of the top N features assigned to each economic theme.
Figure A1: Illustration of SAE’s integration with a Transformer model. The residual stream r(l) is extracted from an intermediate layer of the language model and projected onto a sparse, interpretable representation z(l) . This sparse code is then decoded to reconstruct r^(l) , which is fed back into the model for subsequent processing.
Prompt 8: Financial Journalism Phrase Generation Prompt This prompt asks the model to act as a financial journalist and generate a fixed number of short examples related to a specified small topic within a broader financial category. The examples are constrained to be brief phrases, clauses, or sentence fragments.
Year
Number of Firms
Number of News
% Return > 0
News length
2015
3,610
209,447
49.2%
92
2016
3,481
203,412
50.1%
90
2017
3,468
203,972
48.4%
91
2018
3,455
199,420
47.9%
87
2019
3,460
215,949
51.1%
88
2020
3,550
225,337
46.5%
88
Table A1: Reuters News Sample This table reports annual summary statistics for the matched Reuters and CRSP sample from 2015 through 2024. Overnight alerts are Reuters alerts published after 4:00 p.m. or before 9:00 a.m. For each year, the table reports the number of unique firms, the number of overnight alerts, the percentage of observations with positive intraday returns, and the average alert length in characters.
Figure A2: Word clouds for steered SAE topics.
Figure A3: Word clouds for steered SAE topics, continued.
Figure A4: Word clouds for steered SAE topics, continued.
Figure A5: Workforce Restructuring and Equity-Market Expectations The panels use the same GE article under S&P 500 and employment prompts. Equity Markets accounts for 12% of excess activation under the equity prompt. Under the employment prompt, Human Resources and Unemployment account for 10% and 8%, respectively. The article discusses both workforce reductions and business disposals, so a decline in GE’s headcount should not be read as an equal decline in aggregate employment. The box reproduces the opening excerpt, while the scans use the full article. Brightness is normalized separately in each panel and does not measure differences in total activation.
Figure A6: Precious Metals, Equity Markets, and Interest Rates The panels use the same silver-market article under S&P 500 and AAA bond rate prompts. Equity Markets and Beta account for 16% and 10% of excess activation under the equity prompt. Under the bond-rate prompt, Yield Curve and Monetary Policy each account for 7%, and Quantitative Easing accounts for 3%. These concepts relate to the article’s discussion of asset returns, lower yields, and central-bank bond purchases. The equity scan is more concentrated, while the bond-rate scan is more diffuse. Brightness is normalized separately in each panel, so this contrast concerns the spatial distribution of excess activation, not its total magnitude. The box reproduces the opening excerpt, while the scans use the full article.
Figure A7: Correlation between LLM and Survey Expectations on the S&P 500 This figure reports the LLM-generated S&P 500 expectation series and the AAII survey series at monthly frequency over 2000–2021. The resulting correlation is comparable in magnitude to the 0.35 reported by Bybee (2023) for the same variable using GPT-3.5.
Figure A8: Decomposition of topic contributions For each variable in the panel, the pie chart on the left reports the share of total feature-contribution mass accounted for by each of the five main themes ( Macro-Finance & Economics , Financial Markets & Instruments , Corporate Finance , Investment & Portfolio Management , Risk Management & Derivatives ), and the bar chart on the right decomposes the dominant theme into its constituent topics.
Figure A9: Decomposition of topic contributions, continued
Figure A10: Decomposition of topic contributions, continued
Figure A11: Decomposition of topic contributions, continued
Figure A12: Decomposition of topic contributions, continued
Figure A13: Decomposition of topic contributions, continued
Figure A14: Decomposition of topic contributions, continued
Prompt 9: Risk Aversion and Investment Choice: Startup vs Bond Fund The prompt asks the LLM to choose between a high-risk option, a high-growth tech startup, and a low-risk option, a municipal bond fund. We use it to assess whether the model’s choice aligns with expected risk preferences under a forced-choice investment setting.
Figure A15: Risk Aversion Steering with an Alternative Investment Prompt. This figure repeats the comparison of SAE and prompt steering using Prompt 9 . The vertical axis reports the inner probability of choosing the low-risk municipal bond fund over the startup.
Figure A16: Topic Activation Changes with the Alternative Investment Prompt. This figure reports the ten topics with the largest changes in SAE feature activations when the model is prompted to be 100% more risk-averse in the startup versus municipal bond choice. Changes sum steered minus unsteered feature activations within each topic.
Figure A17: Prompt Steering with a Frontier Model (GPT-5.5). This figure replicates the prompt-steering exercise of Figure 7 using GPT-5.5 in place of Gemma-9B, testing whether the discrete, step-wise behavior depends on model size or architecture. Panel (a) uses Prompt 4 (S&P500 vs. Treasuries) and Panel (b) uses Prompt 9 (Startup vs. Municipal Bond Fund). The y-axis measures the model’s internal probability of allocating to the riskier asset (Option 1). Even for this much larger frontier model, prompt steering produces abrupt, discrete jumps rather than a smooth, continuous adjustment in economic preference.
Figure A18: Fine-grained prompt steering results from steering strengths ranging from −10% to 10% . Panels A and B report results for the Gemma-9B model, while Panels C and D report results for the GPT-5.5 model. Panels A and C show the S&P500/Treasuries comparison, and Panels B and D show the Startup/Bond comparison.
Prompt 10: Risk Aversion and Investment Choice: Alternative S&P500 Framing I This prompt is a slightly different wording of the earlier S&P500 versus Treasury choice, while keeping the same underlying assets and risk structure. The prompt frames the equity option as a broad US equity market exposure and the bond option as government-backed securities to test whether wording variations influence the model’s expressed risk preference.
Prompt 11: Risk Aversion and Investment Choice: Alternative S&P500 Framing II This prompt presents the same underlying investment trade-off as the original S&P500 versus Treasury prompt, but with slightly different wording and framing. The equity option is described as a broad US stock market fund, while the Treasury option is framed as near risk-free short-term government debt. We use this variation to examine whether wording differences affect the model’s risk-sensitive decision making.
Figure A19: This Figure reports the similar result of SAE steering with Prompt 4 but with different wording. The steering strength is calibrated using Prompt 4 and applied to these 2 new prompts. The point of this exercise is to show we can use representative prompt to calibrate the steering strength parameter for similar prompts. You can find these prompts at Prompt 10 and Prompt 11 .
Figure A20: Topic Composition within Corporate Finance This figure decomposes the Corporate Finance features contained in the overall top N SAE features across their constituent topics. Features are ranked by the average absolute magnitude of their fitted coefficients across the annual estimation windows. The horizontal axis reports the overall cutoff N , and the vertical axis reports the percentage of retained Corporate Finance features assigned to each topic.
Figure A21: Topic Composition within Financial Markets & Instruments This figure decomposes the Financial Markets & Instruments features contained in the overall top N SAE features across their constituent topics. Features are ranked by the average absolute magnitude of their fitted coefficients across the annual estimation windows. The horizontal axis reports the overall cutoff N , and the vertical axis reports the percentage of retained Financial Markets & Instruments features assigned to each topic.
Figure A22: Topic Composition within Investment & Portfolio Management This figure decomposes the Investment & Portfolio Management features contained in the overall top N SAE features across their constituent topics. Features are ranked by the average absolute magnitude of their fitted coefficients across the annual estimation windows. The horizontal axis reports the overall cutoff N , and the vertical axis reports the percentage of retained Investment & Portfolio Management features assigned to each topic.
Figure A23: Topic Composition within Macro-Finance & Economics This figure decomposes the Macro-Finance & Economics features contained in the overall top N SAE features across their constituent topics. Features are ranked by the average absolute magnitude of their fitted coefficients across the annual estimation windows. The horizontal axis reports the overall cutoff N , and the vertical axis reports the percentage of retained Macro-Finance & Economics features assigned to each topic.
Figure A24: Topic Composition within Risk Management & Derivatives This figure decomposes the Risk Management & Derivatives features contained in the overall top N SAE features across their constituent topics. Features are ranked by the average absolute magnitude of their fitted coefficients across the annual estimation windows. The horizontal axis reports the overall cutoff N , and the vertical axis reports the percentage of retained Risk Management & Derivatives features assigned to each topic.
Large language model (LLM) interpretability studies how internal representations and computations produce outputs, and how they can be explained or controlled.