Tell me something new: startup valuations, information asymmetry, and the mitigating effect of informational updates

ABSTRACT

A high level of information asymmetry is characterizing for venture capital investments making new information about entrepreneurial companies especially valuable for a venture capitalist’s valuation process. This paper uses text classification and text mining methodology to extract structured data about capital allocation plans in a unique sample of 1,550 European funding rounds that serves as proxy for the private informational updates shared with investors by entrepreneurs. We show that venture capitalists incorporate the content and specificity of information into their valuation process. Further, results confirm that the value of new information is dependent on the prevailing level of information asymmetry.

KEYWORDS:

• Venture capital
• valuation
• information asymmetry
• text analysis

Acknowledgments

We gratefully acknowledge access to Dow Jones Venture Source and Thomson Reuters Datastream provided by DALAHO, University of Hohenheim

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. Grossman (1981) argues that general statements about e.g., the quality of a product would always lead a rational buyer of that product to assume the quality to be the minimum that still fits the general statement.

2. While the information that a company will ’grow sales’ is good news, it is very general information. Disclosing more specific details such as ’doubling sales of existing product X’ is easily verifiable ex-post.

3. The sample includes companies from the following countries: Belgium, Austria, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, United Kingdom, Denmark, Finland, Sweden

4. Around 35%, 20%, and 12% of the sample companies are from the United Kingdom, from France, and from Germany, respectively, while the other countries in the sample have roughly equal shares.

5. In text mining applications a corpus is a collection of individual text documents containing natural language. In this paper the corpus consists of the individual textual descriptions of the planned capital allocation of startups.

6. We use the standard English stop word lexicon provided by the nltk python package (Bird, Klein, and Loper 2009) and add further custom context specific stop words.

7. For the regressions in section 4 the variable is log-transformed.

8. This also includes investments with only one single investor, which, strictly speaking, is not a syndicate.

9. An alternative approach would be to look at the relative increase of valuations of the same company between two consecutive funding rounds to further control for company characteristics. However, this would have significantly reduced our already small sample size.

10. We use the granular industries provided by the database and cluster them into seven broad industry groups: Consumer Goods, Consumer Services, Energy, Financial and Professional Services, Health, Industrial Goods, and Information Technology.

11. We include a dummy equal to one for syndicated deals, a dummy equal to one for companies who conducted an IPO, and country, year, industry, and stage fixed effects in the first stage selection model of the two-stage Heckman procedure.

12. As the dependent variable is log transformed the percentage increase is calculated as follows: e^0.088 − 1 = 0.089 ≈ 9%.