1 Which Factors Matter Most Can Startup Valuation be Micro -Targeted1

2025-04-30 0 0 1.04MB 34 页 10玖币

侵权投诉

Which Factors Matter Most?

Can Startup Valuation be Micro-Targeted?

Max Berre

ABSTRACT

While startup valuations are influenced by revenues, risks, age, and macroeconomic conditions, specific

causality is traditionally a black box. Because valuations are not disclosed, roles played by other factors

(industry, geography, and intellectual property) can often only be guessed at. VC valuation research indicates

the importance of establishing a factor-hierarchy to better understand startup valuations and their dynamics,

suggesting the wisdom of hiring data-scientists for this purpose. Bespoke understanding can be established

via construction of hierarchical prediction models based on decision trees and random forests. These have

the advantage of understanding which factors matter most. In combination with OLS, the also tell us the

circumstances of when specific causalities apply. This study explores the deterministic role of categorical

variables on the valuation of start-ups (i.e. the joint-combination geographic, urban, and sectoral

denomination-variables), in order to be able to build a generalized valuation scorecard approach. Using a

dataset of 1,091 venture-capital investments, containing 1,044 unique EU and EEA, this study examines

microeconomic, sectoral, and local-level impacts on startup valuation. In principle, the study relies on Fixed-

effects and Joint-fixed-effects regressions as well as the analysis and exploration of divergent micro-

populations and fault-lines by means of non-parametric approaches combining econometric and machine-

learning techniques.

Keywords Valuation, Startup Valuation, Venture Capital, Entrepreneurial Finance, Machine Learning, Hierarchical

Analysis.

This article has been sponsored by the Partners of the “Finance for innovation” Chair at Audencia Business School, especially

Early Metrics and Sowefund, who provided financial support to Max Berre during his PhD course. The author wishes to express

sincere gratitude for this support.

PhD Candidate, Audencia Business School, Nantes, France and Université de Lyon, iaelyon, Magellan, Lyon, France

mberre@audencia.com

1. Introduction

Why do startups in California attract higher valuations than those in New York? Or ones based in London

attract higher valuations than those in Paris, Berlin, or Milan, even when based in similarly-sized economies,

sharing the same industries and many of the same investors? What drives this? Which factors matter most?

While classical economic theory describes valuations as being based on revenues, growth-rates, and risk-

adjusted discount-rates, valuation of startups proves the exception to the rule.

Given their opacity, short histories, and vast array of intangible assets, startups are notoriously difficult to

value (Damodaran, 2009). This has given rise to a diversity of valuation approaches dependent on drivers

known to have valuation-impacts on early-stage startups in various phases. Valuation-approaches such as

discounted-cashflow (DCF), multiples-valuation, and scorecard-valuation rely on inputs such as assets or

performance-measures, while other strands of the literature describe the impact of market-characteristics

and competitive-environment on firm-value. Adding to the cacophony, widespread press-coverage,

describes dramatic valuation-divergences along geographic and industry lines– divergences not wholly

explained by growth, risk, revenue, or assets–.

Scarcity of data-availability gives rise to the need for development of empirical research with the aim of

valuation approaches to be deployed in the face of this data-scarcity. Responding to this, econometric-

techniques demonstrate limitations, as revenue-and-risk-based OLS-techniques demonstrate substantial

hidden-variable bias. Meanwhile, as many categories, groups, regions, and clusters known to have

explanatory-power have sparsely-available concrete economic-figures which might explain these valuation-

differences, fixed-effects demonstrate decreasing marginal explanatory-power as these groups are included

and accounted-for. This gives rise to limitations in estimation-accuracy.

To address these methodological-limitations, one could combine known firm-performance indicators and

market-conditions such as growth-rates, business cycles, and risk-premiums, with predictive-segmentation

of categorical variables, and examination of key fault-lines in the startup landscape.

Such approaches are already used in markets, where an often-used startup-valuation approach are scorecard

models, input-based valuation models driven by aggregation of discreet-inputs, and specific discreet

contextual-characteristics in which the startups arise. While these approaches traditionally have limited

generalizability and have accordingly not made much of an impact in the peer-review landscape, fields such

as marketing, psychology, and political science have made extensive use of similar approaches.

This paper’s focus is multifold. First, this study brings recent developments in methodology to bear for

valuation of startups using machine-learning approaches. Second, this study aims to shed light on

divergences between classical valuation-approaches and scorecard and segmented-approaches used in

industry. Most importantly, recent developments in machine-learning approaches make possible the

hierarchical ranking of valuation-factors, thereby minimizing information-asymmetries, enabling more

insightful decision-making. Thirdly, this study seeks to explore the viability of the use of microtargeting on

the basis of non-numerical factors such as geographical, business-model-based, and sectoral factors to

predict startup-valuations to a high degree of accuracy, overcoming firm-level numerical-factor-reliance.

While it has long been established in both classical economic theory and in entrepreneurial finance that

prices and valuations are driven by discreet, measurable factors such as revenues and risk-metrics, the

importance and need for hierarchical-ranking as outlined by Quintero (2019), represents an important and

emerging gap in the literature, which can be addressed via machine-learning.

This study proceeds as follows: The following section examines the relevant literature, while Section 3

describes the dataset and key variables. Subsequently, Sections 4 and 5 describe the research questions,

model-approaches, and findings. Lastly, Section 6 discusses findings and concludes.

2. Literature Review

According to Berre and Le Pendeven (2022), the heritage of contextual-based valuation approaches lies with

scorecard valuation approaches. Examples of this can be found both in peer-review literature such as Hand

(2005) and Sievers et al. (2013), as well as in grey-literature such as Payne (2011) and Berkus (2016), and has

made a substantial impact among practitioners in the private equity and venture capital industries, according

to Ernst & Young (2020).

Building on this, Quintero (2019) describes the importance and relevance of data-science for the

establishment of factor-hierarchy in order to accurately predict valuations within the venture capital industry,

specifically suggesting the use of either Bayesian or machine-learning algorithm approaches.

The validity of machine-learning approaches for finance and for entrepreneurial-finance has also been

expressed in a growing body of published research focusing on several key topics. This includes use of

machine-learning for prediction of both pre- and post-money valuation, for sectoral-clustering, and for

successful-exit (Ang et al., 2020). Furthermore, while appropriate- model-selection represents a unique

challenge among machine-learning practitioners, Ang et al., (2020) and Quintero (2019) demonstrate that

various machine-learning approaches, including Least-Absolute-Shrinkage-and-Selection Operator

(LASSO), Latent Dirichlet Allocation (LDA), Categorization-and-Regression-Tree (CART), Extreme-

Gradient-Boosting (XGBoost), and Bayesian-approaches such as No-U-Turn Sampler (NUTS) and

Markov-Chain-Monte-Carlo (MCMC).

Adding to this, policy sources have also recently begun publishing similar sentiments and views for

applications of economy policy ranging from monetary policy to macroprudential policy, to competition

policy. Jarmulska (2020) is an ECB study comparing the explanatory-power of random forests to those of

more traditional econometric techniques for predicting financial-stress, and finds that the policy-level of

decision-tree-based machine-learning approaches are highly-accurate for financial-prediction purposes, even

in the face of high-dimensionality macrofinancial and macroeconomic data. Meanwhile, similar techniques

are applied to cartel-detection in a competition-policy setting (Huber and Imhof, 2019), for the purposes of

competition-law-screening, focusing both competition-enforcement in convoluted settings, and acting as a

source of actionable-intelligence and demonstrable-evidence. Here again, accurate actionable-predictions

are possible, even in the face of high-dimensionality data.

Mechanically, machine-learning algorithm-reliance on optimal arrangement of multiple decision-factors are

central to their functionality. In parallel, multiple-criteria decision analysis (MCDA), which explicitly

evaluates conflicting-criteria in decision-making is described Zopounidis et al. (2015) as being used for

portfolio and investment-evaluation and selection, is usually implemented in terms of fundamental-factors.

While valuation-factors do not necessarily conflict, the valuation-impact of tradeoffs and fault-lines may

form important valuation model-elements.

Additionally, several key studies in the economics, finance and entrepreneurial fields bring into context the

focus on industry-level, city-level, and business-model focus. Globally, a major debate among the literature

is whether the “horse”, rather than the “jockey” is more likely to drive investment and valuation. While

studies agree that “jockey” refers to management-teams, studies describing the “horse” as industry-level

market-conditions, technological-standards, and market-size, (Gompers and Lerner, 2001; Kaplan et al.,

2009) outline the horse as being most-important, while Gompers et al. (2020), who describe horse as

business-model and firm-related factors, find horse-factors secondary to jockey-factors.

City-level impacts on valuation can be traced to local-level competitive-environments, supply-chain and

consumer sophistication, and local factor-conditions ranging from infrastructure to human-capital and

workforce-specialization (Porter, 1990).

Meanwhile, Damodaran approaches the valuation landscape with a specifically industry-level outlook. Both

Damodaran (2002) and (2009) approach various valuation techniques using sectoral industry-level figures

to drive valuations and analysis. While Damodaran (1993) explicitly finds the effect of insiders, Damodaran

also publishes aggregated industry-level figures for insider, CEO, and institutional holdings.

For valuation approaches useful to incorporate all of these drivers, Berre and Le Pendeven (2020), a

systematic literature review which examines entrepreneurial finance literature and elaborates a multi-step

valuation meta-model which describes the mechanical process by which start-up valuation emerges. While

the meta-model can accommodate a wide-range of start-up valuation drivers.

3. Methodology and Data

Locally-Supplied Proprietary Data and Commercially-Available Data

The data consist of proprietary venture-capital deal-data shared by Early Metrics, a Paris-based startup

ratings and research agency. To this, we add EU-and EEA-located startup-deals drawn from EIKON and

Crunchbase. These supplementary sources were chosen due to their content-similarity to Early Metrics data.

To further enrich the dataset, each deal was cross-referenced with firm-performance, industry-level,

municipal, and national-level macroeconomic data, and including both proprietary, and commercially-

available data, while also boasting extensive variety of value-adding categorical-variables. While the

categorical-variables have substantial explanatory-power in their own right, they also add value by interaction

with firm-characteristics, as well as macroeconomic and business-cycle market-conditions.

European data grants numerous advantages, such as institutional and macroeconomic diversity meaningful

for geographic fixed-effects taking into taking into account of distinct contextual and geographic factors.

Because the EEA market is a large developed private equity market where the Common Law, French Civil

Law, German Civil Law, and Scandinavian legal families are represented, a European dataset provides the

depth of institutional diversity needed to carefully examine the valuation-impact of institutional factors tied

to legal origin in a meaningful way. Additionally, as outlined by Berre and Le Pendeven (2020), a research-

gap exists concerning model-inclusion of contextual and geographic factors.

Because each line within our dataset is specific per-investor-per-deal, deals with multiple investors occupy

multiple lines within the dataset, identifying data for startup and investor, as well as relevant industry-level,

institutional, and macroeconomic data for both parties. Since a start-up can have several investors, it can

have multiple observations in the regression analysis, reflecting each unique investor–startup pair. The

dataset style is borrowed from Masulis and Nahata (2009). With 1,089 observations representing 1,042 deals

across 673 startups ranging from Q1-2000 to Q1-2020, our dataset-size is substantial, although only 582

observations contain firm-level revenue figures. Nevertheless, this yields regressions with substantial degrees

of freedom compared to prominent studies in the entrepreneurial finance and startup field, such as Gompers

et al. (2020), Greenberg (2013), and Masulis and Nahata (2009), who examine 444, 317, and 273 observations

respectively. This is outlined in Table 1.

Table 1: Dataset Observations

Source

Observations

EIKON

397

Early Metrics

Crunchbase

614

Total

1091

Dependent Variables

The primary dependent variable in this study is pre-money valuation, with valuations expressed in EUR.

From EIKON and Early Metrics, the data were collected in Euro, while Crowdcube data was converted

from GBP, while Crunchbase data was converted from various currencies for deals in which valuations in

EUR were not available. Table 2 outlines the summary statistics of our pre-money valuations data. While

the data’s time and sectoral distribution is somewhat uneven, it does cover several major events, including

the end of the dotcom bubble, the Eurozone crisis, and the start of the Covid-19 Pandemic.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

1WhichFactorsMatterMost?CanStartupValuationbeMicro-Targeted?1MaxBerre2ABSTRACTWhilestartupvaluationsareinfluencedbyrevenues,risks,age,andmacroeconomicconditions,specificcausalityistraditionallyablackbox.Becausevaluationsarenotdisclosed,rolesplayedbyotherfactors(industry,geography,andintellectualprop...

展开>> 收起<<

1 Which Factors Matter Most Can Startup Valuation be Micro -Targeted1.pdf

共34页,预览5页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

1 Which Factors Matter Most Can Startup Valuation be Micro -Targeted1

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: