Joshua Foster
Joshua Foster has a total of 24 co-authors. Here are the top co-authors by number of shared publications:
- Shannon L. Rawski
Organizational Behaviour Group, Ivey Business School, Western University, London, Canada
4 shared publications - Fredrik Ødegaard
Ivey Business School Western University London Ontario Canada
2 shared publications M. Ryan Haley
2 shared publicationsAmy Farmer
University of Arkansas, 238 Harmon Ave., Fayetteville, AR 72701, US
1 shared publication- Andrew T. Soderberg
Department of Management & Human Resources, College of Business, University of Wisconsin Oshkosh, Oshkosh, USA
1 shared publication Bryan Engelhardt
University of Wisconsin - Oshkosh - Department of Economics
1 shared publication
Latest Publications
- Open Access
- 13 November 2026
- Published by Elsevier in SSRN Electronic Journal
Setting Optimal Reserve Prices using Fine-Tuned Language Models
Problem definition. Online auction houses must set reserve prices for highly differentiated products whose unstructured descriptions contain valuable demand information. Existing approaches generally require managers to specify product attributes in advance. We ask whether a language model can instead learn from listing descriptions and historical bids to set item-specific reserve prices before bidding begins. Methodology/results. We fine-tune a pretrained text embedding language model using the structural likelihood of auction bids as its training objective. This directs representation learning toward the valuation distribution that determines the reserve decision. We show that consecutive upper bids identify the tail information needed to set an optimal reserve without observing or modeling the number of potential bidders. A split-sample estimator learns the representation on one half of the data and conducts structural estimation and inference conditional on that representation on the other. We apply the method to approximately 65,800 online used-vehicle auctions. With seller values calibrated to match the sale rate among listings, the model implies a counterfactual net increase of $165.8 million, representing a 14.6% increase above the incumbent reserve policy. Managerial implications. The framework translates a listing description and a seller's value into an item-specific reserve recommendation without constructing a product taxonomy or requiring a bid history for the item being priced. It provides an approach to adapting language models for operational decisions in marketplaces where products are heterogeneous and demand-relevant information is recorded in text.
- 13 November 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSetting Optimal Reserve Prices using Fine-Tuned Language Models
Problem definition. Online auction houses must set reserve prices for highly differentiated products whose unstructured descriptions contain valuable demand information. Existing approaches generally require managers to specify product attributes in advance. We ask whether a language model can instead learn from listing descriptions and historical bids to set item-specific reserve prices before bidding begins. Methodology/results. We fine-tune a pretrained text embedding language model using the structural likelihood of auction bids as its training objective. This directs representation learning toward the valuation distribution that determines the reserve decision. We show that consecutive upper bids identify the tail information needed to set an optimal reserve without observing or modeling the number of potential bidders. A split-sample estimator learns the representation on one half of the data and conducts structural estimation and inference conditional on that representation on the other. We apply the method to approximately 65,800 online used-vehicle auctions. With seller values calibrated to match the sale rate among listings, the model implies a counterfactual net increase of $165.8 million, representing a 14.6% increase above the incumbent reserve policy. Managerial implications. The framework translates a listing description and a seller's value into an item-specific reserve recommendation without constructing a product taxonomy or requiring a bid history for the item being priced. It provides an approach to adapting language models for operational decisions in marketplaces where products are heterogeneous and demand-relevant information is recorded in text.
- Open Access
- 7 August 2026
- Published by Elsevier in SSRN Electronic Journal
The Innate Economic Preferences of Language Models
Language models increasingly settle real resource tradeoffs on behalf of principals yet their economic preferences remain unobserved. We demonstrate their generation rule is isomorphic to the random utility model of discrete choice. This allows internal logit scores to structurally identify preferences. Estimating risk attitudes across twelve models in a portfolio task reveals universal but heterogeneous risk aversion. Although models reject strictly dominated options, their elicited preferences fail invariance tests and violate the independence of irrelevant alternatives across varying experimental prompts. Finally, fine tuning establishes that a principal can explicitly engineer a target risk attitude.
- 7 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessThe Innate Economic Preferences of Language Models
Language models increasingly settle real resource tradeoffs on behalf of principals yet their economic preferences remain unobserved. We demonstrate their generation rule is isomorphic to the random utility model of discrete choice. This allows internal logit scores to structurally identify preferences. Estimating risk attitudes across twelve models in a portfolio task reveals universal but heterogeneous risk aversion. Although models reject strictly dominated options, their elicited preferences fail invariance tests and violate the independence of irrelevant alternatives across varying experimental prompts. Finally, fine tuning establishes that a principal can explicitly engineer a target risk attitude.
- Open Access
- 7 August 2025
- Published by Elsevier in Journal of Public Economics
- Vol. 249
Auctions for risk-averse non-profits
- 7 August 2025
- Published by Elsevier in Journal of Public Economics
- Vol. 249
Open AccessAuctions for risk-averse non-profits
- Open Access
Decoding Consumer Preferences Using Attention-Based Language Models
This paper proposes a new demand estimation method using attention-based language models. An encoder-only language model is trained in a two-stage process to analyze the natural language descriptions of used cars from a large US-based online auction marketplace. The approach enables semi-nonparametrically estimation for the demand primitives of a structural model representing the private valuations and market size for each vehicle listing. In the first stage, the language model is fine-tuned to encode the target auction outcomes using the natural language vehicle descriptions. In the second stage, the trained language model's encodings are projected into the parameter space of the structural model. The model's capability to conduct counterfactual analyses within the trained market space is validated using a subsample of withheld auction data, which includes a set of unique "zero shot" instances.
Open AccessDecoding Consumer Preferences Using Attention-Based Language Models
This paper proposes a new demand estimation method using attention-based language models. An encoder-only language model is trained in a two-stage process to analyze the natural language descriptions of used cars from a large US-based online auction marketplace. The approach enables semi-nonparametrically estimation for the demand primitives of a structural model representing the private valuations and market size for each vehicle listing. In the first stage, the language model is fine-tuned to encode the target auction outcomes using the natural language vehicle descriptions. In the second stage, the trained language model's encodings are projected into the parameter space of the structural model. The model's capability to conduct counterfactual analyses within the trained market space is validated using a subsample of withheld auction data, which includes a set of unique "zero shot" instances.
- 15 July 2023
- Published by Springer Nature in Journal of Business Ethics
- Vol. 195 (4) , 879-898
The Devil is in the Details: Sexual Harassment e-Training Design Choices and Perceived Messenger Integrity
- 15 July 2023
- Published by Springer Nature in Journal of Business Ethics
- Vol. 195 (4) , 879-898
The Devil is in the Details: Sexual Harassment e-Training Design Choices and Perceived Messenger Integrity
- 11 September 2022
- Published by Elsevier in Economics Letters
- Vol. 220
Semi-nonparametric estimation of secret reserve prices in auctions
- 11 September 2022
- Published by Elsevier in Economics Letters
- Vol. 220
Semi-nonparametric estimation of secret reserve prices in auctions
- Open Access
- 29 April 2022
- Published by American Psychological Association (APA) in Technology, Mind, and Behavior
- Vol. 3 (2)
Supplemental Material for Sexual harassment bystander training effectiveness: Experimentally comparing 2D video to virtual reality practice.
Sexual harassment (hereafter SH) is a dysfunctional workplace behavior, resulting in negative outcomes for individuals, work groups, and organizations. Since #MeToo, companies have been innovating to increase the effectiveness of SH training by incorporating new content (e.g., bystander intervention skills) and new technology (e.g., virtual reality, hereafter VR). However, research has yet to determine the best practices or the effectiveness of these new innovations. The present study hypothesizes that SH bystander intervention training will be more effective when VR practice scenarios are used rather than two-dimensional (2D) video practice scenarios. We argue that the increased presence (i.e., the perception that people and places in a virtual simulation are real) afforded by VR should better replicate bystander experiences in real SH situations, thereby allowing trainees to develop bystander skills in a more realistic practice experience than 2D video provides. We experimentally test our hypothesis in a laboratory setting (N = 100). Our results show that the VR practice condition differed from the 2D video condition by increasing trainees’ intentions to engage in indirect, nonconfrontational, and widely applicable interventions (e.g., intervene by removing the target from the situation, approach the target to offer support later). However, our manipulation showed a negative effect on practice quantity (i.e., those in the VR condition explored fewer response options) and no effect on other operationalizations of training effectiveness (e.g., motivation to learn, knowledge, attitudes toward the training, intentions to directly confront the harasser, and intentions to formally report the harassment). Implications, limitations, and directions for future research are discussed.
- 29 April 2022
- Published by American Psychological Association (APA) in Technology, Mind, and Behavior
- Vol. 3 (2)
Open AccessSupplemental Material for Sexual harassment bystander training effectiveness: Experimentally comparing 2D video to virtual reality practice.
Sexual harassment (hereafter SH) is a dysfunctional workplace behavior, resulting in negative outcomes for individuals, work groups, and organizations. Since #MeToo, companies have been innovating to increase the effectiveness of SH training by incorporating new content (e.g., bystander intervention skills) and new technology (e.g., virtual reality, hereafter VR). However, research has yet to determine the best practices or the effectiveness of these new innovations. The present study hypothesizes that SH bystander intervention training will be more effective when VR practice scenarios are used rather than two-dimensional (2D) video practice scenarios. We argue that the increased presence (i.e., the perception that people and places in a virtual simulation are real) afforded by VR should better replicate bystander experiences in real SH situations, thereby allowing trainees to develop bystander skills in a more realistic practice experience than 2D video provides. We experimentally test our hypothesis in a laboratory setting (N = 100). Our results show that the VR practice condition differed from the 2D video condition by increasing trainees’ intentions to engage in indirect, nonconfrontational, and widely applicable interventions (e.g., intervene by removing the target from the situation, approach the target to offer support later). However, our manipulation showed a negative effect on practice quantity (i.e., those in the VR condition explored fewer response options) and no effect on other operationalizations of training effectiveness (e.g., motivation to learn, knowledge, attitudes toward the training, intentions to directly confront the harasser, and intentions to formally report the harassment). Implications, limitations, and directions for future research are discussed.
- Open Access
- 1 January 2022
- Published by American Psychological Association (APA) in Technology, Mind, and Behavior
- Vol. 3 (2) , 302-321
Sexual harassment bystander training effectiveness: Experimentally comparing 2D video to virtual reality practice.
Sexual harassment (hereafter SH) is a dysfunctional workplace behavior, resulting in negative outcomes for individuals, work groups, and organizations. Since #MeToo, companies have been innovating to increase the effectiveness of SH training by incorporating new content (e.g., bystander intervention skills) and new technology (e.g., virtual reality, hereafter VR). However, research has yet to determine the best practices or the effectiveness of these new innovations. The present study hypothesizes that SH bystander intervention training will be more effective when VR practice scenarios are used rather than two-dimensional (2D) video practice scenarios. We argue that the increased presence (i.e., the perception that people and places in a virtual simulation are real) afforded by VR should better replicate bystander experiences in real SH situations, thereby allowing trainees to develop bystander skills in a more realistic practice experience than 2D video provides. We experimentally test our hypothesis in a laboratory setting (N = 100). Our results show that the VR practice condition differed from the 2D video condition by increasing trainees’ intentions to engage in indirect, nonconfrontational, and widely applicable interventions (e.g., intervene by removing the target from the situation, approach the target to offer support later). However, our manipulation showed a negative effect on practice quantity (i.e., those in the VR condition explored fewer response options) and no effect on other operationalizations of training effectiveness (e.g., motivation to learn, knowledge, attitudes toward the training, intentions to directly confront the harasser, and intentions to formally report the harassment). Implications, limitations, and directions for future research are discussed.
- 1 January 2022
- Published by American Psychological Association (APA) in Technology, Mind, and Behavior
- Vol. 3 (2) , 302-321
Open AccessSexual harassment bystander training effectiveness: Experimentally comparing 2D video to virtual reality practice.
Sexual harassment (hereafter SH) is a dysfunctional workplace behavior, resulting in negative outcomes for individuals, work groups, and organizations. Since #MeToo, companies have been innovating to increase the effectiveness of SH training by incorporating new content (e.g., bystander intervention skills) and new technology (e.g., virtual reality, hereafter VR). However, research has yet to determine the best practices or the effectiveness of these new innovations. The present study hypothesizes that SH bystander intervention training will be more effective when VR practice scenarios are used rather than two-dimensional (2D) video practice scenarios. We argue that the increased presence (i.e., the perception that people and places in a virtual simulation are real) afforded by VR should better replicate bystander experiences in real SH situations, thereby allowing trainees to develop bystander skills in a more realistic practice experience than 2D video provides. We experimentally test our hypothesis in a laboratory setting (N = 100). Our results show that the VR practice condition differed from the 2D video condition by increasing trainees’ intentions to engage in indirect, nonconfrontational, and widely applicable interventions (e.g., intervene by removing the target from the situation, approach the target to offer support later). However, our manipulation showed a negative effect on practice quantity (i.e., those in the VR condition explored fewer response options) and no effect on other operationalizations of training effectiveness (e.g., motivation to learn, knowledge, attitudes toward the training, intentions to directly confront the harasser, and intentions to formally report the harassment). Implications, limitations, and directions for future research are discussed.
- Open Access
- 14 January 2021
- Published by Wiley in Contemporary Economic Policy
- Vol. 39 (4) , 760-777
The relationship between in‐person voting and COVID‐19: Evidence from the Wisconsin primary
On April 7, 2020, Wisconsin held its presidential primary election, and news reports showed long lines of voters due to fewer polling locations. We use county‐level variation in voting patterns and weekly county‐level COVID test data to examine whether in‐person voting increased COVID‐19 cases. We find a statistically significant association between in‐person voting density and the spread of COVID‐19 2–3 weeks after the election. In our main results, a 10% increase in in‐person voters per polling location is associated with an 18.4% increase in the COVID‐19 positive test rate 2–3 weeks later.
- 14 January 2021
- Published by Wiley in Contemporary Economic Policy
- Vol. 39 (4) , 760-777
Open AccessThe relationship between in‐person voting and COVID‐19: Evidence from the Wisconsin primary
On April 7, 2020, Wisconsin held its presidential primary election, and news reports showed long lines of voters due to fewer polling locations. We use county‐level variation in voting patterns and weekly county‐level COVID test data to examine whether in‐person voting increased COVID‐19 cases. We find a statistically significant association between in‐person voting density and the spread of COVID‐19 2–3 weeks after the election. In our main results, a 10% increase in in‐person voters per polling location is associated with an 18.4% increase in the COVID‐19 positive test rate 2–3 weeks later.