Portfolio-Driven Entrepreneurship

How to cite

Kim, Dahyeon; Kim, Hyunjin. 2026. "Portfolio-Driven Entrepreneurship." AEA RCT Registry. AEARCTR-0019159. https://doi.org/10.1257/rct.19159-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Dahyeon Kim* , Hyunjin Kim

* Primary investigator

Registry ID
AEARCTR-0019159
DOI
https://doi.org/10.1257/rct.19159-1.0
First registered
Last updated
Status
In development
Trial dates
2026-07-14 – 2027-09-03
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/19159

Abstract

The experiment randomly varies whether founders focus on a single venture or a portfolio of ventures in parallel to examine the effects on venture-building behavior and performance.

Experimental design

This study tests whether a portfolio-based venture-building strategy – in which founders launch and evaluate multiple ventures in parallel – changes venture performance relative to a traditional ‘focus’ strategy centered on building a single venture. It also examines the mechanisms through which a portfolio strategy may affect performance, including whether the portfolio approach enables founders to select higher-potential ventures based on early market signals, changes the allocation of founder effort and attention, or affects entrepreneurial learning over time.

Design details

Randomization method
We implemented stratified randomization to assign participants to treatment and control. Stratification was based on two elements: previous entrepreneurial experience x domain experience. 1. Previous entrepreneurial experience is a binary measure based on participants' baseline reports of whether they had previously founded a venture. Participants with no prior entrepreneurial experience are coded as “None,” while participants with any prior entrepreneurial experience are coded as “Any.” 2. Domain experience is a binary measure based on participants' reported years of experience in the primary industry or domain in which they have spent their professional careers. Participants with less than 5 years of domain experience are coded as “Low,” while participants with 5 or more years of domain experience are coded as “High.” The 5-year cutoff corresponds to the median level of domain experience in the sample. Randomization was implemented by computer, assigning participants to one of two arms, Focus and Portfolio, within each stratum. We will report descriptive comparisons of baseline characteristics between the Focus and Portfolio groups, including whether participants entered the program with an existing business idea, total years of work experience, years of domain experience, ability to build a product, baseline AI usage and number of AI tools used, age, education, income, and gender. Our primary specifications will account for the stratified randomization design by including randomization-strata fixed effects. We will additionally report specifications that control for prespecified pre-treatment founder characteristics to improve precision.
Randomization unit
The unit of randomization is the founder. Outcomes are measured at the founder level or, where outcomes are observed at the venture level, aggregated to the founder level for analysis.
Clusters
N/A
Observations
Approximately 200 ventures. We have approximately 295 entrepreneurs who have signed up. Before the intervention begins, we allow for a period of attrition, and expect the final sample size to be smaller given the virtual nature of the program.
Arms
Assuming a final sample of approximately 200 ventures, we anticipate an approximately 1:1 allocation between treatment and control arms.
Primary outcomes
Venture-level performance outcomes are constructed from platform data and aggregated to the founder level for analysis. For founders in the Focus arm, the founder-level measure is the outcome of the founder’s single venture. For founders in the Portfolio arm, venture-level outcomes are summarized in three ways: (i) the mean outcome across ventures in the founder’s portfolio, (ii) the median outcome across ventures, and (iii) the outcome of the founder’s highest-performing venture. Our primary outcomes of interest are: 1. Total visitors: The cumulative number of landing page visitors per venture from launch to endline. 2. Total leads: The cumulative number of leads captured per venture from launch to endline. 3. Total purchases: The cumulative number of purchases generated per venture from launch to endline. 4. Total revenue: Cumulative revenue generated per venture from launch to endline. 5. Any traction: Separate binary indicators for whether the venture attracted at least one visitor, captured at least one lead, generated at least one purchase, and generated any revenue by endline 6. Traction index: A composite index (0–3) based on three binary indicators measured at endline: (i) whether the venture attracted any visitors, (ii) whether the venture captured any leads, and (iii) whether the venture had any paying customers. 7. Continuation intent Stated continuation: The founder’s self-reported continued engagement in venture building after the program. Revealed continuation: Total number of hours spent on the platform during the first month following the summer camp 8. Survival of venture Stated survival: A binary indicator equal to one if the founder reports that the venture remains active and has not been shut down Revealed survival: A venture-level indicator equal to one if the platform records show meaningful activity during the final week of the summer camp. Meaningful activity may include edits to the venture, product or landing page updates, marketing actions, lead or customer interactions, or transactions. For founders in the Focus arm, the founder-level survival outcome corresponds to the survival status of the single venture. For founders in the Portfolio arm, venture survival is aggregated to the founder level using two measures: (i) the proportion of ventures in the founder’s portfolio that remain active; and (ii) an indicator equal to one if at least one venture in the portfolio remains active. The first captures survival across the portfolio, while the second captures whether the founder produces at least one continuing venture.
Secondary outcomes
We will examine three key mechanisms through which the Portfolio treatment may affect outcomes: selection, diversification, and learning. The selection mechanism concerns whether founders use early market signals to allocate greater time and financial resources to more promising ventures and progressively concentrate their portfolios. This may improve performance by facilitating the identification of stronger opportunities, but may reduce performance if founders place excessive weight on noisy early signals or discontinue promising ventures prematurely. The diversification mechanism concerns whether maintaining alternative ventures allows founders to remain entrepreneurially active and reallocate resources after discontinuing an individual venture. This may increase founder-level persistence, but may also dilute attention, delay concentration, and weaken commitment to any single opportunity. The learning mechanism concerns whether exposure to multiple ventures generates broader, transferable knowledge through comparison across opportunities, or whether sustained focus instead produces deeper venture-specific knowledge and better judgment. We will therefore measure early venture-level market signals; subsequent time and financial investment; venture continuation, modification, and discontinuation; changes in the number of active ventures and the concentration of resources across them; founder activity following venture discontinuation; founders’ stated reasons for venture decisions; self-reported learning and its perceived applicability across ventures; probabilistic forecasts of venture outcomes; and performance on an endline venture-evaluation task. We will additionally measure total founder effort and the allocation of time across information acquisition, venture comparison, and implementation. These measures will allow us to assess the potential benefits and costs of the Portfolio treatment. In addition to the outcomes above, we will conduct the following exploratory analyses: 1. Treatment effect heterogeneity: We will examine whether the effect of the Portfolio treatment on the primary outcomes varies across founder characteristics, such as prior domain expertise, prior entrepreneurial experience, education level, technical background, and other pre-treatment characteristics measured at baseline. 2. Idea generation: We will examine whether the Portfolio treatment changes what venture opportunities founders pursue. Using survey responses, platform data, and venture descriptions generated throughout the program, we will descriptively compare the characteristics of venture ideas developed by founders in the two conditions. These characteristics may include measures of idea novelty, feasibility, diversity, or other dimensions identified through manual coding, LLM-based coding, or other text-analysis methods. 3. Exploratory text analyses: We will conduct exploratory analyses of text generated throughout the program. These data may include platform-based descriptions of the ventures founders build; open-ended survey responses concerning topics such as founder learnings, customer acquisition approaches, or challenges faced; conversations between founders and the AI agent; and transcripts of program meetings and founder conversations, where available. We may use manual coding, LLM-based coding, natural language processing, or other text-analysis methods to construct measures. Because these analyses are exploratory, we do not commit to specific hypotheses or measures in advance, and we will treat any results as hypothesis-generating rather than confirmatory.

Cited by

No citation data available for this registration yet.

Keywords