INTRODUCTION

The world of startups is a dynamic place — for entrepreneurs, there is seldom any promise of vested interest or reward once they hit the ground running. Private equity and venture capital firms enjoy a hefty profit because they are supposedly good at identifying which startups sizzle and which ones fizzle, but is there a way to automate the process? My goal was to build a model to answer the question of:

Can a computer model how much funding a startup should receive?

Note: the tables in this article were meant to be viewed on a computer!

|| THE DATA ||

The data used for this article was obtained through PitchBook, one of the best sources for research and financial information on startups and the realm of private investment. PitchBook provided me with the following information:

1. Last Valuation (in millions)
2. Last Funding Size (in millions)
3. Last Funding Round (Series A, Series B, etc.)
4. Last Funding Stage (Seed Round, Early Stage VC, etc.)
5. HQ Location
6. Number of Employees
7. Industry Sector (Healthcare, Consumer Products, etc.)
8. Industry Group (Healthcare Services, Pharmaceuticals, Software, etc.)
9. Total Raised
10. Number of Active Investors

The total sample size used in this experiment consisted of 973 startup companies that had received funding over the past two years.

EXPLORATORY DATA ANALYSIS ||

AngelList has a statistical demonstration of a couple of variables that are involved with the startup culture, but what intrigues me more is the actual startup process. The image below (taken from CBInsights) details the “Venture Capital Funnel,” or how most startups follow a particular predictable life cycle.

|| THE METHOD ||

The given startups were organized into buckets according to the rules defined below (where x is the funding amount):

                   x ≤ $5M –> Class 1(SMALL)

             $5M < x ≤ $25M  –> Class 2(MEDIUM)

             $25M < x ≤ $45M –>  Class 3(LARGE)

                    x > $45 –> Class 4(GIANT)

I did this in order to change the problem I was trying to solve from a regression problem (what is the exact value of funding the startup will receive?) into a classification problem (in which range will the funding fall into?). The loss of accuracy here is not an enormous problem since it is not necessary to try and predict exacts!

Using the R package ‘mice,’ missing values in the dataset were approximated with a neural network using variables like “Total Raised” and “Number of Active Investors” (to name a few). These variables were dropped because, for example, the model may not have access to the number of active investors until after a funding round ends.

The following models were trained on a limited set of features, and were appropriately weighted during testing according to their performances during the training period:

• Random Forest
• Linear XGBoost
• Tree XGBoost
• Extreme Learning Machine (ELM)
• Conditional Inference Tree
• Bagged Classification and Regression Tree (CART)
• Neural Net
• K-Nearest Neighbors

|| THE RESULTS ||

I was quite impressed with the results I was able to get from my fairly complex model! After the dynamic weighting I performed, I got the following results. The first table is a summary of the classification scores I calculated, and the second table is a confusion matrix.

                 ╔═════════════╦══════════════╗

                 ║ accuracy ║ kappa     ║

                 ╠═════════════╬══════════════╣

                 ║ 0.70696 ║ 0.51984    ║

                 ╚═════════════╩══════════════╝

                               Predicted

                Observed     1   2   3   4

                        1 | 71  13 0  2

                        2 | 27 110   6 4

                        3 |  1 12  11 15

                        4 |  0 0   0 1

These results demonstrate that my model was able to predict with 70.6% accuracy and had a moderate-strong measure of agreement. The models did have some difficulty estimating the “Large” category of funding sizes but performed fairly well overall.

Now, for the more interesting part. What variables did the model care about most? The ‘caret’ library in R provides an easy way to measure variable relevance.

                Class 1  Class 2 Class 3  Class 4  Average Rank

Funding Round   | 100.00   73.286 58.538   100.00 82.956 2

Employees       |  83.11   98.774 53.270    98.77 83.481 1

Funding Stage   |  73.62   72.776 60.768    73.62 70.196 3

HQ Location     |  17.79   21.155 17.786    21.15 19.470 4

Industry Group  |  12.54    0.000 13.984    12.54 9.766 5

Industry Sector |  10.48    5.845 5.972    10.48 8.194 6

From this table, it looks like the “Funding Round” (Series A, B, etc.) and the number of employees were the most important variables. This makes sense because as the startup enters more funding rounds or is larger (i.e. has more employees), the cheque size increases. More interestingly, however, is the fact that the “Industry Group” and “Industry Sector” variables contributed very little to the model — this means that the industry of the startup (Software, Pharmaceuticals, etc.) makes no difference! It was interesting to see that the funding landscape is not being partial to just software startups. Lastly, it seems as if the location of the company’s headquarters really does give some sort of indication as to whether or not that startup will receive higher cheque sums!

However, the ideal “startup conditions,” in essence, are masked in that there are a huge amount of other variablesthat play a larger factor in deciding the cheque size.

Some real-world examples of this phenomenon include Dwolla and AgCode, two very successful startups that began their journeys in Des Moines, IA and Glenwood, MN, respectively. The real predictor of their success was not their industry (financial services and agricultural technology) and a naive model trained only on location would have discredited them (seeing as how they are based in the Midwestern region of the United States). My model emphasizes the way the startup was able to grow and expand itself as a better predictor, citing the number of total employees and the corresponding funding round as stronger inputs. In other words, a startup that can grow faster (more employees) with less funding rounds is a healthy one.

Entrepreneurship and the private investment market are incredibly complex spaces. I invite you to conduct your own experiments relating to this topic and would love to hear any findings or progress! Give this article some claps if you liked it and stay tuned for more data-driven idea exploration by following TDS Team and me (Abhinav Raghunathan)!

INVEST IN CANADA DO BUSINESS WITH THE WORLD

Ecompanies Canada offers Information, Knowledge & Expert Advice to Help You Start, Grow, Market and Expand a Strong Successful Business in Canada.

Company Registration services for Canadian residents.

To incorporate a new company in Canada., one or more persons(called “the incorporators”) may form a company by filing articles of incorporation, notice of address and notice of directors with the Corporate Registry Office of its jurisdiction of incorporation.

We offer fast & easy Canada incorporation and business registration services anywhere in Canada. At Ecompanies we help you step-by-step and take care of the entire business registration process from start to finish. Incorporating a business with us is fast, easy and takes just minutes.

Register today your new business:

https://www.ecompaniescanada.com/incorporation-service/

Canada Incorporation Service for Non Canadian Residents.

Throughout Canada, corporations are the most widely used legal vehicle for operating a business. A corporation has the same rights and obligations under Canadian law as a natural person. Among other things, this means it can acquire assets, go into debt, enter into contracts, sue or be sued,.

and even be found guilty of committing a crime.

To incorporate a new company in Canada., one or more persons(called “the incorporators”) may form a company by filing articles of incorporation, notice of address and notice of directors with the Corporate Registry Office of the desired jurisdiction of registration.

Ecompanies Canada offers fast & easy Canada online incorporation and business registration services to non-Canadian residents interested in doing business in Canada. At Ecompanies Canada we help you step-by-step and take care of the entire business registration process from start to finish. Incorporating a business with us is fast, easy and takes just minutes.

Register today your company in Canada as a Non-Canadian resident

https://www.ecompaniescanada.com/incorporation-service-non-canadian-residents/

Incorporation Service for Foreign Companies in Canada

Ecompanies Canada offers fast & easy Canada online incorporation and business registration services to foreign companies interested in doing business in Canada. At Ecompanies Canada we help you step-by-step and take care of the entire business registration process from start to finish. Incorporating a business with us is fast, easy and takes just minutes.

Register today your foreign company in Canada

https://www.ecompaniescanada.com/incorporation-service-foreign-companies-canada/

This Article first appeared in towardsdatascience

The post Learned Entrepreneurship What do the most funded startups have in common? appeared first on Ecompanies Canada.

from Ecompanies Canada http://bit.ly/2IpS8FA