Week 6: Alternative Finance & Marketplace Lending

Learning Objectives

Define alternative finance and distinguish crowdfunding models (rewards, equity, debt)
Analyse marketplace lending platforms using two‑sided market theory
Explain how platforms address information asymmetry and adverse selection
Evaluate credit risk assessment using alternative data and machine learning
Assess inclusion benefits and regulatory challenges in alternative finance
Implement credit risk scoring and platform economics analysis in Python

Opening frame: “Welcome to Week 6. We are exploring how technology is reshaping credit markets, enabling peer-to-peer lending, crowdfunding, and new forms of capital allocation.”
Big question: “How do platforms overcome the information problems that have always plagued credit markets? And does alternative finance genuinely expand access, or just repackage exclusion?”
Connect to earlier weeks: Bring back our core lens: measurement, selection, and data quality decisions. In credit markets those choices become underwriting, pricing, and inclusion trade-offs.
Connect to Week 5: Week 5 focused on robo-advisers and portfolio construction. Week 6 shifts to consumer and SME credit, but the shared theme is the same: technology changes intermediation, and the statistical assumptions matter.
Learning objectives roadmap: By the end you’ll understand different crowdfunding models, how marketplace lending works as a two-sided platform, how alternative data improves credit assessment, and the regulatory frontier.
Assessment relevance (if applicable): You may be asked questions on platform economics and credit assessment, or to write a reflective analysis on whether alternative finance expands inclusion (and at what cost).
Practical hook: “We’ll implement a simple credit scoring model using alternative data, the type platforms like LendingClub or Funding Circle use, and see how it differs from traditional credit bureaus.”
Engagement: Ask students if they’ve backed a Kickstarter project or used a P2P lending platform. This grounds the discussion in experience.
Timing: Spend 2-3 minutes here; return to objectives at wrap-up.

Agenda

Part	Theme	~Time
I	What is alternative finance? Taxonomy and models	10 min
II	Crowdfunding: rewards, equity, and debt	15 min
III	Marketplace lending as a two-sided platform	15 min
IV	Credit risk, adverse selection, and alternative data	20 min
V	Inclusion, regulation, and Python implementation	15 min

Part I — What Is Alternative Finance?

Traditional Capital Allocation

Banks (debt intermediation)

Take deposits → make loans
Screen borrowers, hold credit risk on balance sheet
Regulated: capital requirements, deposit insurance
Relationship-based: know the customer over time

Capital markets (equity & bond)

Firms issue shares or bonds to investors directly
Underwritten by investment banks (IPOs, bond issuance)
Requires scale: costly for small firms or individuals
Secondary markets provide liquidity (LSE, NYSE)

The gap: Small businesses, thin-file consumers, and early-stage ventures are too small or too risky for both channels.

Defining Alternative Finance

Financial channels outside traditional banking and capital markets

Why “alternative”?

Internet platforms lowered matching costs from the 2000s onward
Not necessarily better — different intermediation model
Targets underserved segments: SMEs, thin-file borrowers

Common features

Platform intermediation — software replaces branches
Direct matching — fewer layers between capital and borrower
Data-driven underwriting — alternative signals replace credit scores

The paradox: The narrative is “cut out the middleman.” The reality: platforms are middlemen — just ones that don’t take balance-sheet risk.

Alternative to what?: Traditional banking (relationship lending, branch networks, balance sheets) and traditional capital markets (IPOs, bond issuance, institutional investors).
Why “alternative”?: Term emerged in the 2000s to describe new financing channels enabled by internet platforms. Not necessarily better, just different mechanisms.
Platform intermediation: All models use digital platforms to match funders with projects/borrowers. This replaces traditional intermediaries (bank branches, VC offices, investment banks) with software.
Direct matching: Platforms enable direct connections between capital suppliers (investors, lenders, backers) and capital demanders (entrepreneurs, borrowers, project creators). This reduces layers but doesn’t eliminate intermediation entirely.
Data-driven underwriting: Platforms use algorithmic assessment rather than relationship-based lending. Alternative data (education, employment, cash flow) supplements or replaces traditional credit scores.
Disintermediation narrative: Early rhetoric was “cutting out the middleman” (banks). Reality: platforms are middlemen too, just different ones. They don’t take balance-sheet risk but provide matching, underwriting, servicing.
Connection to Week 1: This is FinTech product innovation (new services) rather than process innovation (making existing services cheaper). It targets underserved segments—small businesses, consumer borrowers with thin credit files.
Student engagement: “Who’s backed a Kickstarter? Who’s borrowed from a P2P platform? What was your experience?”
Transition: “Now let’s see the core models that make up alternative finance.”

Core Models

1. Crowdfunding

Many small contributors fund projects or businesses

Rewards (product pre-order)
Equity (shares)
Debt (loans)

Kickstarter · Seedrs · Funding Circle

2. Marketplace Lending

Platforms match borrowers and lenders directly

Started retail P2P
Now institutional-dominated
Consumer and SME credit

LendingClub · Prosper · Zopa

3. Invoice Finance

Businesses sell receivables for immediate cash

Solves cash-flow timing gap
Tech-enabled factoring
Short-duration, asset-backed

MarketFinance · Fundbox

Why three categories: These capture the major alternative finance models by funding mechanism. Each has distinct economics, risks, and regulatory treatment.
Interactive element: Links are clickable—encourage students to briefly explore during break. Kickstarter shows creative projects with funding goals, Seedrs shows equity deals with investment minimums, Funding Circle shows business loans with risk grades.
What to notice when visiting: Kickstarter—see funding goals, backer numbers, reward tiers. Seedrs—see equity percentages, company valuations, minimum investments (often £10-100). Funding Circle—see loan purposes, risk ratings (A-E), expected returns. Zopa is UK’s oldest P2P lender (2005), now also a bank.
UK accessibility: Seedrs, Funding Circle, and Zopa are UK-based and FCA-regulated, making them directly relevant to students. Kickstarter and US platforms show international scope.
Crowdfunding origins: Kickstarter (2009), Indiegogo (2008) pioneered rewards-based model. Funding Circle (2010), LendingClub (2006) pioneered debt-based models. Equity crowdfunding came later (post-JOBS Act 2012 in US, post-2012 in UK).
P2P vs. marketplace lending evolution: Originally “peer-to-peer” implied retail investors lending to retail borrowers. As platforms matured, institutional investors (hedge funds, banks) became dominant — LendingClub SEC filings (2015) indicated institutional investors provided ~80% of capital at that point. “Marketplace lending” reflects this shift. ⚠️ Exact current proportions vary by platform and are not systematically reported; treat the “institutional-dominated” characterisation as qualitatively accurate rather than a precise figure.
Invoice finance mechanics: A business delivers goods, sends invoice (payment due in 30-90 days), but needs cash now. Platform buys the invoice at discount (e.g., £95 for £100 invoice), business gets immediate cash (£95), platform collects full £100 when due (£5 profit). This is old factoring model, now tech-enabled.
Scale comparison: Based on Cambridge Centre for Alternative Finance (2018) data: marketplace lending (P2P/debt) dominated at ~$250bn globally, crowdfunding ~$30bn, invoice finance ~$20bn. Marketplace lending dominates because it targets larger market (consumer/SME credit) with repeat usage. Note: These are 2018 industry estimates—exact figures vary by methodology and reporting.
Geographic concentration: Cambridge Centre for Alternative Finance (2018) reports China led with $215.37bn (mostly P2P lending), US $61bn, UK $10.4bn. Then China experienced massive bust (2018-2020) when thousands of platforms failed amid fraud scandals. US and UK markets remained more stable due to stronger regulatory frameworks.
Pedagogical point: Before diving into details, establish the landscape—three core models with different risk profiles, return expectations, and regulatory treatments.
Student task: “Which of these have you heard of? Which seems riskiest for investors? Take 2 minutes to explore one platform—what strikes you?”
Transition: “Now let’s break down each model in detail, starting with the full taxonomy.”

Taxonomy of Alternative Finance

Five distinct models with different risk-return profiles:

1. Donation-based
GoFundMe, JustGiving — No financial return expected

2. Rewards-based
Kickstarter, Indiegogo — Pre-purchase product or receive perk

3. Equity crowdfunding
Seedrs, Crowdcube — Receive equity stake in company

4. Debt-based / P2P lending
Funding Circle, LendingClub — Lend money, receive interest

5. Invoice trading
MarketInvoice, Fundbox — Buy invoices at discount for cash flow

Donation-based: Pure altruism or social causes. Not really “finance” (no expectation of return), but shares platform infrastructure and dynamics. GoFundMe raised $10bn+ for medical costs, disasters, social causes (platform reported data).
Rewards-based: Backers pre-purchase products or receive perks (early access, merchandise). Kickstarter: $7bn+ raised, ~220k projects funded (Kickstarter 2024). Success rate 44-48% (projects that reach funding goal) (Mollick 2014). Social proof (early backers) predicts success.
Equity crowdfunding: Backers become shareholders. Highly regulated (securities laws). In UK, £1.5bn raised via equity crowdfunding (2011-2020). Success stories: BrewDog, Monzo. But many failures—investors face illiquidity and high risk.
Debt-based (P2P/marketplace): Largest segment globally. LendingClub originated $60bn+ loans (2007-2020) (LendingClub 2007--2020). Funding Circle £15bn+ (2010-2020, platform data). Zopa £6bn+ (2005-2020, platform data). Returns: investors earned 3-7% after defaults; borrowers paid 5-12% (vs. 15-25% for payday loans, platform reported data).
Invoice trading: Businesses sell receivables (unpaid invoices) to investors at discount. Solves cash flow timing problem. Grew to $3bn+ annually in UK.
Key differences across models: Return expectations (none → equity upside), risk (donation < rewards < equity < debt), regulation (lightest for donation, heaviest for equity/debt), liquidity (all illiquid except maybe secondary markets).
Platform role varies: Matching, screening, pricing, servicing, trust/reputation. Debt platforms do most underwriting; rewards platforms do less screening (backers judge project quality).
Student task: “Rank these by riskiness for investors. What information would you need for each?”
Assessment (if applicable): You may be asked to classify platforms or identify key differences between models.
Transition: “Now let’s synthesize what makes alternative finance distinct, and where it’s grown.”

Alternative Finance: Scale and Growth

Global market evolution: (Cambridge Centre for Alternative Finance 2020)

2013: ~$11bn globally (nascent stage)
2020: $300bn+ globally (mature but uneven growth)

Geographic concentration: (Cambridge Centre for Alternative Finance 2018)

China: $215bn (P2P lending boom, then bust 2018-2020)
United States: $61bn (LendingClub, Prosper, marketplace lending)
United Kingdom: $10.4bn (Funding Circle, Zopa, regulatory leader)

COVID-19 impact: Accelerated growth in developed markets as traditional banks pulled back from riskier segments

Scale evolution demonstrates FinTech boom: From $11bn (2013) when alternative finance was experimental, to $300bn+ (2020) when it became mainstream (Cambridge Centre for Alternative Finance 2020). This 25x growth in 7 years reflects: (1) Platform maturation, (2) Regulatory acceptance, (3) Institutional capital entry, (4) Banks pulling back post-2008. Note: These figures are industry estimates from Cambridge Centre for Alternative Finance (2020) benchmarking reports—treat as approximate market indicators, not precise academic measurements. Methodology varies by report and includes different platform types and regions.
China’s P2P boom and bust: China had 2,000+ P2P platforms in 2017, lending $215bn annually (Cambridge Centre for Alternative Finance 2018). Then massive fraud, Ponzi schemes discovered. Government cracked down 2018-2020. By 2020, only ~30 licensed platforms remained. Thousands failed, millions of investors lost money. This is cautionary tale—light regulation enables growth but also fraud.
US marketplace lending maturity: LendingClub (founded 2006) went public 2014 (IPO $9bn valuation), then struggled (scandal 2016, acquired 2020). Prosper ($20bn+ originated) remains private. SoFi ($60bn valuation 2021) expanded beyond lending. US market is mature—institutional investors dominate, retail participation declined.
UK as regulatory leader: UK FCA authorized P2P lending 2011, created regulatory framework before other countries. Result: orderly growth, some failures (Lendy, Collateral) but no systemic crisis. Funding Circle (founded 2010) went public 2018 (valuation £1.5bn), now £15bn+ originated.
COVID-19 dual impact: (1) Traditional banks tightened lending (uncertain economy) → opportunity for alternative finance. (2) But alternative finance platforms also restricted lending (default risk spiked) → some failed, survivors focused on prime borrowers. Net effect: growth in developed markets with strong platforms, contraction in weaker platforms and emerging markets.
Market structure differences: US dominated by consumer unsecured lending (personal loans, student loan refinancing). UK balanced between consumer and SME lending. China was mostly consumer P2P.
Pedagogical point: Scale demonstrates alternative finance is real (not just hype), but geographic concentration and China’s bust show risks. Regulation and governance matter enormously.
Student engagement: “Why did China’s P2P market collapse while UK/US markets survived?” (Regulation, fraud, Ponzi schemes, lack of oversight.)
Assessment (if applicable): You may be asked about market scale or geographic patterns, or to analyse a boom–bust episode as a governance failure.
Transition: “We’ve covered the taxonomy and scale. Now let’s dive deep into each crowdfunding model.”

Part II — Crowdfunding Models

Rewards Crowdfunding

How it works:

Creator: Sets funding goal and timeline, defines rewards
Backers: Pledge money (pre-purchase product or receive perk)
All-or-nothing: Goal reached → creator gets funds; Goal missed → refunds
Delivery: Creator must deliver rewards (but ~9% fail to deliver)

Platform economics:

Platform fee: 5-10% of funds raised (Kickstarter 5%, Indiegogo 5%)
Payment processing: ~3%
Creator keeps: 85-90% of raised funds

Source: Platform fee schedules as of 2024

Kickstarter all-or-nothing model: Projects must hit funding goal or get $0 (pledges refunded). This creates urgency and reduces free-rider problem (if everyone waits, no one backs). Indiegogo offers flexible funding (keep whatever you raise), but evidence suggests all-or-nothing performs better.
Why all-or-nothing works better: Forces creators to set realistic goals (they get nothing if too ambitious). Signals credibility to backers (creator confident they’ll hit goal). Creates urgency (back now or project might fail).
Platform fees: Kickstarter charges 5% + payment processing (~3%). On a $100k project, that’s $8k to platform. Expensive, but reflects value of: (1) Audience access (millions of potential backers), (2) Trust infrastructure (payment handling, dispute resolution), (3) Discovery (search, recommendations).
Failure to deliver: ~9% of successfully funded projects fail to deliver rewards on time or at all (Kickstarter/platform data). Reasons: (1) Underestimating production costs, (2) Manufacturing delays, (3) Technical challenges, (4) Fraud (rare but exists—creator takes money and disappears). Backers have little recourse—no investor protections, platforms don’t guarantee delivery.
Information asymmetry: Backers can’t verify product feasibility claims. Creator has more information than backers about technical challenges, costs, timeline. This creates adverse selection (optimistic or fraudulent creators over-promise). Social proof and reputation mitigate but don’t eliminate risk.
Comparison to traditional finance: Venture capital would conduct due diligence (technical assessment, market validation, financial projections), demand equity stake, impose governance (board seats, reporting requirements). Crowdfunding democratizes access by eliminating vetting, but this trades inclusion for quality control.
Student engagement: “If you backed a Kickstarter that failed to deliver, how did you feel? What recourse did you have?” (Likely: frustrated, felt scammed, but accepted risk as part of crowdfunding.)
Transition: “Now let’s see what predicts success—research evidence on Kickstarter campaigns.”

Kickstarter: What Works?

Success rate: 44-48% of projects reach funding goal (Mollick 2014)

Key predictors of success:

1. Early momentum

Funds raised in first 48 hours strongest predictor
Creates social proof cascade (others see backing → join)

2. Social proof

Number of backers matters more than dollar amount
100 backers pledging $50 each > 10 backers pledging $500

3. Quality signals

Video presentation (explains concept, builds trust)
Detailed description (reduces uncertainty)
Regular updates during campaign (maintains engagement)

4. Creator reputation

Prior successful projects → roughly 2× success rate (Mollick 2014)
Large social media following → substantially more funds raised (Mollick 2014)

Mollick (2014) foundational study: Analyzed 48,562 Kickstarter projects (2009-2012), Journal of Business Venturing. Dataset success rate 48.1%; Kickstarter-reported rate 44.7% at that time. This is the gold-standard empirical study of rewards crowdfunding success factors. ⚠️ Specific effect-size percentages in slides (video, updates, social media) are directional interpretations of Mollick’s regression findings — not exact point estimates. Consult original paper for precise coefficients.
Early momentum critical: Funds raised in first 48 hours is the single strongest predictor of eventual success — signals quality to later backers, triggers Kickstarter’s discovery algorithm, and creates bandwagon effects. This is Mollick’s core finding on social dynamics.
Social proof vs. dollar amount: A project with 100 backers at $50 each signals broader appeal than 10 backers at $500 each (same total). Many small backers implies genuine demand; few large backers may be friends/family. Mollick finds backer count matters beyond capital amount.
Quality signals reduce uncertainty: Video lets backers assess creator credibility. Projects with video succeed at substantially higher rates (Mollick’s analysis shows statistically significant positive effect — “50% higher” on slides is an approximate interpretation, not a precise point estimate from the paper).
Updates during campaign: Mollick finds frequent updates positively predict success (approximate interpretation on slides: “30% higher” — directional not precise). Updates signal transparency and commitment.
Creator reputation effects: Projects by creators with prior successful campaigns have approximately 2x success rate (this is consistent with Mollick’s findings on reputation effects — treat as order-of-magnitude).
Social media following: Mollick finds social capital (Facebook friends proxy) positively predicts funding. “40% more” on slides is an approximate interpretation — the key insight is the effect is statistically significant and economically meaningful, favouring creators with existing audiences.
Category variation: Success rates differ substantially by category — performing arts (higher) vs. technology (lower, vaporware risk) vs. food (lower, execution complexity). Mollick’s paper reports category fixed effects; the specific percentages on slides (62%, 34%, 22%) are illustrative approximations of the direction and ranking, not precise estimates.
Implications for platforms: Kickstarter’s algorithm promotes projects with early momentum, creating winner-take-most dynamics. Popular projects get homepage placement → more backers → more placement. This is efficient (channels capital to credible projects) but creates inequality (some projects ignored despite quality).
Behavioral economics interpretation: Backers exhibit herding (follow others’ decisions), overweight social proof (100 backers must know something), and anchor on initial momentum (early success implies quality). These biases are rational in settings with high uncertainty and information asymmetry.
Student discussion: “If you were launching a Kickstarter, which of these factors could you control? Which are luck?” (Control: video quality, description, updates, social media outreach. Luck: timing, media coverage, virality.)
Assessment (if applicable): You may be asked to explain Mollick’s findings on crowdfunding success factors, and/or to critically evaluate whether social proof leads to efficient capital allocation or just herding.
Transition: “Rewards crowdfunding is low-stakes pre-purchase. Now let’s examine equity crowdfunding—real investment with real risk.”

Equity Crowdfunding: Access?

How it works:

Company: Posts pitch (business plan, financials, video) on platform
Platform: Vets companies (only ~5-10% approved), ensures disclosure
Investors: Buy shares (typical investment £500-£2,000)
Outcome: Investor owns equity stake, awaits exit (acquisition, IPO, or failure)

Key difference from rewards crowdfunding:

Real ownership stake (not just pre-purchase)
Regulated as securities offering
Illiquid (can’t easily sell shares)
High risk, high potential return

Regulatory context: Equity crowdfunding was illegal until recently. Selling securities to non-accredited investors violated securities laws. US JOBS Act (2012) created Regulation Crowdfunding (Reg CF) exemption allowing companies to raise up to $5M/year. UK FCA authorized in 2011. EU followed with various national frameworks.
Why it was illegal: Securities laws (Securities Act 1933 in US, Financial Services and Markets Act 2000 in UK) prohibit selling shares to general public without registration. Rationale: protect unsophisticated investors from fraud and losses. Equity crowdfunding required legislative change (JOBS Act, FCA authorization) to create exemption.
Disclosure requirements: Companies must provide: business plan, use of funds, financial statements (2 years), risk factors, management bios. More than Kickstarter (which requires almost nothing) but less than IPO (audited financials, 100+ page prospectus, lawyer review). This is compromise—enable access whilst requiring minimum transparency.
Platform screening process: Crowdcube and Seedrs (UK’s largest) screen applications: (1) Check company is legitimate (not fraud), (2) Assess pitch quality (business plan coherence), (3) Evaluate market potential (is this investable?). Only ~5-10% approved (platform reported data). Why screen? Platform reputation—if too many companies fail, investors lose trust. But screening is light compared to VC (which does months of due diligence).
Typical investment size: £500-£2,000 per investor. This is retail scale—small enough for middle-class investors but large enough to be meaningful. Some sophisticated investors put in £10k-£50k. Regulations often cap investments (e.g., UK: 10% of net worth unless self-certified sophisticated).
Investor profile: UK FCA research (circa 2020): approximately 50% of equity crowdfunding investors are “sophisticated” (high income, financial literacy, prior investment experience), while 50% are first-time investors attracted by democratization narrative. Concern: retail investors often underestimate risk, expect Monzo-like returns, don’t understand illiquidity. Note: This is based on FCA consultations and market studies—exact figure varies by survey methodology.
Student engagement: “Would you invest £1,000 in an equity crowdfunding deal? What information would you need to assess risk?”
Transition: “Now let’s look at the economics and scale of equity crowdfunding.”

Equity: Scale in Context

Channel	Annual UK volume	Typical deal
Equity crowdfunding	~£300–500M/yr (Cambridge Centre for Alternative Finance 2020)	£200k–£500k
UK VC	~£10–15bn/yr (BVCA)	£1M–£50M
LSE new listings	£5–20bn/yr (LSE data)	£50M–£1bn+

Equity crowdfunding is ~2–5% of VC by volume — a genuine niche for seed and community rounds, not a mainstream market.

Notable exits (survivor bias warning):

Monzo — £20m crowdfunded → £4bn+ valuation (Crowdcube / press 2021)
BrewDog — £70m+ across “Equity for Punks” rounds (company data 2010–2020)
Revolut — £12m crowdfunded → £25bn+ valuation (Crowdcube 2016 / press)

Equity: Platform Fees

Who pays what (Seedrs / Crowdcube fee schedules, 2024):

Company

5–7% of funds raised
Possible success / completion fee
Legal and due-diligence costs on top

Investor

0–5% transaction fee (varies by platform)
Secondary market: exists on Seedrs, but thin — most shares never trade

High fees relative to deal size are why equity crowdfunding suits £200k–£500k raises, not £5M rounds.

UK market leadership: UK is world’s largest equity crowdfunding market (by volume and maturity). Why? (1) FCA authorized early (2011), (2) Active retail investor culture, (3) Strong platforms (Seedrs, Crowdcube), (4) Success stories (Monzo, BrewDog) generated publicity.
US market slower growth: Despite larger economy, US equity crowdfunding smaller than UK. Why? (1) JOBS Act came later (2016 implementation), (2) Reg CF caps ($5M max raise) limit appeal, (3) SEC more restrictive than FCA, (4) Accredited investor rules exclude most retail investors from best deals.
Average campaign size: £200k-£500k is typical successful raise. This is seed-stage capital—too small for VC (who prefer £1M+ rounds) but meaningful for early-stage companies. Larger companies use equity crowdfunding for marketing/brand (not just capital)—e.g., Brewdog raised to create customer-owners.
Platform fees: 5-7% to company is comparable to traditional IPO underwriting fees (which are 7% for small IPOs). But IPO raises £10M-£1bn; equity crowdfunding raises £200k-£500k. So platform is taking proportionally more for less capital. Reflects platform’s value: audience access, vetting, regulatory compliance, payment infrastructure.
Investor fees: Some platforms (e.g., Crowdcube) charge investors 0-1% transaction fee. Others (e.g., Seedrs) historically charged 7.5% (now reduced). These fees erode returns—£1,000 investment loses £75 upfront at 7.5% fee, needs to appreciate >8% just to break even.
Secondary market illiquidity: Seedrs offers secondary market (investors can sell shares to other investors), but trading is thin. ~90% of shares never trade (platform data). Why? (1) Early-stage companies hard to value (no market price), (2) Information asymmetry (buyer doesn’t know what seller knows), (3) Small investor base (not many buyers). Result: investors should assume 5-10 year lockup.
Success stories are outliers: Monzo, BrewDog, Revolut are exceptional—top 0.1% of equity crowdfunding companies. These are survivor bias examples. For every Monzo, hundreds of companies failed and investors lost 100%. Media focuses on winners; losers fail quietly.
Monzo case study: Raised £20m via Crowdcube (2016-2018), now valued £4bn (press reports, 2021 valuation). Early investors earned 200x+ return (£1,000 → £200,000, illustrative calculation). But Monzo also raised £500m+ from VCs (Accel, General Catalyst, Tencent). VC capital was far more important than crowdfunding capital. Crowdfunding was PR/brand-building as much as fundraising.
BrewDog case study: Raised £70m+ across multiple “Equity for Punks” rounds (2010-2020, company data). Created 200,000+ “punk shareholders” (customers who own equity). Brilliant marketing—customers feel ownership, evangelize brand. But BrewDog also controversial: accused of mistreating employees, overvaluation in secondary market (platform set prices optimistically, media reports).
Student discussion: “Why do you think Monzo and BrewDog succeeded while most equity crowdfunding companies failed?” (Execution, market fit, VC backing, brand/marketing, luck.)
Transition: “Let’s examine BrewDog in detail—it’s both a success story and a cautionary tale.”

BrewDog’s Equity for Punks

A Case Study (2009-2023): (BrewDog 2009--2023)

200,000+ investors raised £100m+ across multiple rounds
Created “punk shareholders” — customers who own equity and evangelise brand
Perks: discounts, exclusive beers, AGM invites, shareholder bars

The genius:

Brand loyalty — investors defend the brand because they own it
Free marketing — each shareholder tells their network
Differentiation — “anti-corporate” punk ethos vs. Big Beer (AB InBev, Heineken)

Equity for Punks as marketing innovation: BrewDog didn’t just raise capital—they created a movement. By selling equity directly to customers, they transformed consumers into brand evangelists. Shareholders promoted BrewDog because they owned it. This is marketing genius disguised as finance.
Scale of community: 200,000+ investors is extraordinary for equity crowdfunding. Most campaigns attract 100-500 investors. BrewDog built an army of advocates. Each shareholder was incentivized to promote (share price rises if brand succeeds).
Perks structure: Discounts (10-20% off in bars), exclusive beer releases (shareholders-only brews), AGM invites (rockstar-style events with free beer), shareholder bars (London, Berlin—VIP treatment). These perks created experience value beyond financial returns.
Anti-corporate positioning: BrewDog positioned itself as “punk” alternative to corporate beer (Budweiser, Carlsberg owned by AB InBev/Heineken). Founders James Watt and Martin Dickie portrayed themselves as rebellious underdogs fighting Big Beer. This resonated with millennials disillusioned with corporate culture.
Viral mechanics: Each investor told friends/family about their stake. Social proof—“I own part of BrewDog!” Posts on Instagram, Facebook. This created network effects—more investors → more visibility → more customers → more investors.
Comparison to traditional VC: Traditional VC brings capital + expertise but no marketing. BrewDog’s model brought capital + marketing + distribution (shareholders bought more beer). This is strategic genius—align capital with customer acquisition.
Student engagement: “Would you invest in a company just for the brand affiliation and perks, even if financial returns uncertain?”
Transition: “But the BrewDog story has a dark side—controversies about how early investors were treated.”

BrewDog: What Went Wrong?

1. PE dilution (2017) (BrewDog 2017) - TSG Consumer Partners (a US private equity firm) bought 22% of BrewDog for ~£213m - Every existing shareholder’s percentage ownership fell proportionally - “Anti-corporate punk” brand now backed by the very establishment it marketed against

2. Valuation & liquidity - BrewDog set its own price for each new crowdfunding round (£1bn → £2bn valuations) — no independent market test - Secondary market existed on paper but barely traded: retail sellers, no institutional buyers, no agreed price - Result: early investors were locked in — they owned shares valued at £2bn on paper with no route to cash out

3. Culture scandal — “Punks With Purpose” (2021) (Punks With Purpose 2021) - 300+ ex-employees: bullying, unsafe conditions, toxic culture - Brand damage → reputational risk → shareholder value risk

TSG investment controversy (2017): TSG Consumer Partners (US private equity) invested reported £213m for 22% stake, valuing BrewDog at ~£1bn. This was good for founders (cash out, validate valuation) but bad for early crowdfunding investors. Why? Their ownership percentage diluted—if you owned 0.01% before, now you own 0.0078% (22% reduction). Your stake is worth less unless valuation increase compensates. But secondary market didn’t reflect £1bn valuation—investors couldn’t sell at that price.
Anti-corporate hypocrisy: BrewDog built brand on “anti-establishment” punk ethos. Then they took PE money from TSG (which also owns Pabst Blue Ribbon, vitamin water brands—very corporate). Some shareholders felt betrayed—“You sold out to the establishment you claim to oppose.” BrewDog defended it as “necessary for growth,” but it damaged punk credibility.
Secondary market illiquidity: BrewDog offered secondary market (shareholders can sell to other shareholders), but trading was minimal. Why? (1) No transparent valuation—company set prices for new rounds, but secondary market had no liquidity to validate those prices. (2) Information asymmetry—existing shareholders know more than new buyers. (3) Small investor base—most “Equity for Punks” investors were retail, not sophisticated investors who trade. Result: early investors locked in, couldn’t exit.
Overvaluation accusations: BrewDog’s valuation in later crowdfunding rounds (£1bn+, then £2bn in 2020 round) was questioned. Critics argued these were “aspirational” valuations, not market-tested (no public trading, no VC due diligence). Some early investors felt company hyped valuations to attract new investors, but existing shareholders saw no real exit path at those prices.
Punks With Purpose scandal (June 2021): 300+ former and current employees signed open letter alleging: (1) Culture of fear (bullying by management, including founders), (2) Mental health crisis (employees burned out, some hospitalized), (3) Safety issues (bar staff assaulted, company didn’t support), (4) Hypocrisy (marketed as inclusive/progressive but internally toxic). This went viral—major PR crisis.
James Watt’s response: CEO apologized, said “we’re sorry, we got it wrong,” commissioned independent review, hired culture consultants, promised reforms (better HR, mental health support, anonymous reporting). But many saw it as damage control—ex-employees felt vindicated but skeptical of real change.
Investor impact: Brand damage → risk to business → risk to shareholder value. If customers boycott BrewDog due to culture scandals, revenue falls, valuation drops, shareholders lose. This is reputational risk—equity investors bear it.
Broader lesson: Equity crowdfunding ties investors to company culture/values. If company behaves badly, investors complicit (you own it). Traditional VC can exit; retail crowdfunding investors stuck. This is governance risk retail investors often don’t anticipate.
Assessment relevance (if applicable): You may be asked to critically evaluate BrewDog’s Equity for Punks strategy: financial innovation versus marketing exploitation of retail investors (and the governance/liquidity risks that were underplayed).
Student discussion: “If you were an early BrewDog investor, how would you feel about the TSG deal and culture scandals? Would you sell if you could?”
Transition: “BrewDog shows both the power and peril of equity crowdfunding—brilliant community-building, but retail investors bear risks they may not fully understand.”

Equity Crowdfunding: Risks

Risk	Reality
Failure rate	~50–70% of startups fail within 5 years; crowdfunding likely higher (Gompers et al. 2010)
Adverse selection	Best startups choose VCs; crowdfunding gets VC-rejected or control-averse founders
Illiquidity	5–10 year lockup typical; secondary markets thin and slow
Return uncertainty	Too early for robust data; early evidence is bimodal — few big wins, many zeros

Startup failure rates: Academic research (Gompers et al. 2010) shows ~50% of VC-backed startups fail within 5 years, ~70% within 10 years. Note: Exact failure rates vary by study methodology and sample period. Equity crowdfunding companies likely have higher failure rates because: (1) Less vetting (platforms screen lightly vs. VC’s months of due diligence), (2) Adverse selection (best companies go to VCs), (3) No value-add (crowdfunding investors provide capital only; VCs provide networks, expertise, governance).
Adverse selection is central problem: Why would a startup choose crowdfunding over VC? Possible reasons: (1) VC rejected them (adverse selection—company not good enough), (2) Founder wants to avoid VC control/dilution (but this may be suboptimal—VC governance improves outcomes), (3) Marketing/brand-building (legitimate reason—e.g., BrewDog creating customer-owners). Implication: equity crowdfunding attracts mix of VC-rejects and special cases. Investors must screen carefully.
VCs add value beyond capital: Research (e.g., Gornall & Strebulaev 2015) shows VC-backed startups perform better than non-VC startups with similar characteristics. Why? VCs provide: (1) Networks (intro to customers, partners, acquirers), (2) Expertise (strategic advice, operational help), (3) Governance (board seats, performance monitoring), (4) Reputation (VC backing signals quality). Crowdfunding investors provide none of this—just capital.
Illiquidity risk: Investors are locked in until exit (acquisition, IPO) or failure (liquidation). Typical exit timeline: 5-10 years for successful startups, 2-3 years for failures. If investor needs money before exit (medical emergency, job loss, house purchase), they can’t easily sell equity crowdfunding shares. Secondary markets exist (Seedrs) but low liquidity. This is unsuitable for investors with uncertain liquidity needs.
Returns evidence still emerging: Most equity crowdfunding investments made post-2012 (UK) or post-2016 (US). Too early for 10-year return data. Early exits suggest bimodal distribution: few big winners (Monzo 200x), many total losses (100 companies failed → 0x). Expected return highly uncertain—could be 5-10% IRR (decent), could be -50% (loss of half capital). Academic estimates suggest equity crowdfunding returns likely lower than VC returns (20-25% IRR) due to adverse selection.
Comparison to angel investing: Equity crowdfunding is similar to angel investing (investing in early-stage startups). But angels typically: (1) Invest larger amounts (£10k-£100k), (2) Do more due diligence (meet founders, review financials), (3) Provide value-add (mentorship, networks), (4) Diversify across 10-20 startups. Retail equity crowdfunding investors do none of this—invest £1k, read pitch, hope for best. This is high-risk speculation, not professional angel investing.
Regulatory warnings: UK FCA and US SEC require platforms to warn investors: “Investing in early-stage companies is high risk. You may lose all your capital. Do not invest more than you can afford to lose. Diversify across multiple investments.” These warnings are effective—they scare off some retail investors (good—protection works) but sophisticated investors ignore them (they know the risks).
Who should invest in equity crowdfunding?: (1) Sophisticated investors who understand startup risk, (2) Investors with long time horizons (5-10 years), (3) Investors who can afford to lose 100% of investment, (4) Investors diversifying across 10+ deals (not putting all eggs in one basket). Not suitable for: (1) Retail investors with limited savings, (2) Investors needing liquidity, (3) Investors expecting guaranteed returns.
Student reflection: “Given these risks, why does equity crowdfunding exist? What social benefit does it provide?” (Access to capital for startups that VCs reject; democratization of investing; brand-building for companies.)
Assessment (if applicable): You may be asked to evaluate equity crowdfunding as an innovation: does it genuinely democratise investing, or does it expose retail investors to inappropriate risk? Use adverse selection, illiquidity, and returns evidence.
Transition: “Equity crowdfunding is high-risk. Now let’s look at debt-based crowdfunding—marketplace lending—where risk is more measurable.”

Marketplace Lending

Mechanism:
Borrowers apply via platform. Platform assesses credit risk, assigns grade (A-F), sets interest rate. Investors browse loans, choose which to fund (or use auto-invest). Platform services loans (collects payments, handles defaults).

Economics:

Platform revenue: Origination fee from borrower (1-5% of loan), servicing fee from investor (1% annual)
Borrower cost: Interest rate 5-15% (vs. 15-25% payday loans, 10-20% credit cards)
Investor return: 3-7% after defaults (vs. 1-2% savings accounts)

Cumulative originations (to ~2020, platform-reported): LendingClub $60bn+ (LendingClub 2007--2020) · Funding Circle £15bn+ · Prosper $20bn+
Note: LendingClub acquired Radius Bank in 2021 and now operates as a regulated bank — the pure marketplace model has evolved.

Marketplace lending as two-sided platform: Investors (capital supply) on one side, borrowers (capital demand) on other. Platform matches, prices, and services. This is Week 3 platform economics applied to credit.
Platform value proposition—borrowers: Lower rates than credit cards/payday loans, faster approval than banks (minutes vs. days), online convenience. Targets prime and near-prime borrowers (FICO 640-780) who banks underprice or reject.
Platform value proposition—investors: Higher returns than savings/bonds, diversification (can spread £10k across 100 loans), transparency (can see loan purpose, borrower profile), control (choose risk level).
Credit risk assessment: Platforms use traditional factors (credit score, income, debt-to-income) PLUS alternative data (digital footprint, bank account data, employment). Berg et al. (2020) show digital footprint data improves prediction, especially for borrowers with limited credit history.
Pricing mechanism: Platform uses statistical model (logistic regression, random forest, neural net) to predict default probability. Assigns risk grade (A = 5% default risk, F = 25% default risk). Sets interest rate to compensate (A = 6%, F = 18%). Investors choose risk-return trade-off.
Default rates: Vary by loan grade (LendingClub 2007--2020). A-grade loans default ~5-10%. E-F grade loans default 20-30%. Overall portfolio default ~8-12%. This is higher than prime mortgages (~2%, industry data) but lower than credit cards (~15-20%, Federal Reserve consumer credit data).
Investor returns: After accounting for defaults and fees, investors earn 3-7% on diversified portfolios. Better than savings accounts (1%) but not as good as stocks (8-10% historical). Risk: loan defaults correlated with economy—recessions hurt badly.
Business model evolution: Initially “peer-to-peer” (retail investors). Now “marketplace” (institutional investors dominate—hedge funds buy 70%+ of loans, industry estimates). Why? Institutions can analyze better, bear risk better, demand lower returns (accept 4% vs. retail expecting 7%).
Securitization: Platforms bundle loans and sell to institutional investors as asset-backed securities. This is traditional finance repackaged. LendingClub issued $10bn+ in securitizations before IPO (SEC filings, pre-2014).
COVID-19 impact: Many platforms restricted lending in March 2020 (uncertainty spike). Some failed (Lendy, Collateral). Survivors focused on prime borrowers. Lesson: marketplace lending is procyclical—amplifies credit cycles.
Student task: “Would you lend £1,000 via a marketplace platform? Which loan grade? Why?”
Assessment: This is core Week 6 content. You must understand platform mechanics, pricing, and risks.
Transition: “Now let’s analyse marketplace lending using the two-sided market framework.”

Part III — Marketplace Lending as a Two-Sided Platform

Two-Sided Markets

Marketplace lending = classic two-sided platform

Two sides:

Demand side: Borrowers (need capital)
Supply side: Investors / Lenders (provide capital)
Platform: Matches, prices, services the loans

Applying the two-sided market framework:

Cross-side network effects (positive feedback loops)
Same-side congestion (quality vs. quantity tradeoffs)
Asymmetric pricing (who pays what?)
Platform governance (screening, monitoring, enforcement)

Week 3 connection: This week applies platform economics concepts from Week 3 to credit markets. Marketplace lending demonstrates two-sided markets, network effects, and pricing dynamics in financial services context.
Two sides clearly defined: Borrowers (demand capital) and investors/lenders (supply capital). Platform sits in middle, providing matching, underwriting, servicing. Unlike banks (which intermediate using balance sheet), platforms don’t take principal risk—they connect parties.
Cross-side effects critical: More investors → lower borrowing rates (supply increases) → attracts more borrowers. More high-quality borrowers → higher investor returns → attracts more investors. These positive feedback loops are key to platform growth.
Same-side congestion risk: Too many low-quality borrowers → high defaults → investors leave (lemons problem). Too many investors → competition drives rates down → lower platform revenue. Platforms must balance volume and quality on both sides.
Asymmetric pricing: Platforms charge borrowers more (1-5% origination fee + interest) than investors (1% servicing fee). Reflects elasticities—borrowers have fewer alternatives (need money), investors have many (stocks, bonds, savings).
Platform governance essential: To avoid adverse selection (bad borrowers driving out good), platforms must screen aggressively, price risk accurately, and monitor performance. This is costly (30-40% of revenue) but necessary for survival.
Student engagement: “Why would you use a marketplace platform instead of a bank? As a borrower? As an investor?”
Transition: “Now let’s break down the cross-side network effects and same-side congestion in detail.”

Cross-Side Network Effects

Investor → Borrower

More capital available
Interest rates fall
Attracts high-quality borrowers
Better borrower mix

Borrower → Investor

Lower default rates
Higher investor returns
Attracts more investors
More capital deployed

Result: Positive feedback loop — each side’s participation makes the platform more valuable to the other

Cross-side effects are positive feedback loops: Each side’s participation makes the platform more valuable to the other side. This is classic two-sided market dynamics (Rochet & Tirole 2003).
Investor to borrower transmission: If platform attracts 1,000 investors willing to lend £10M total, and borrower demand is £5M, then platform has excess supply. Result: interest rates fall (borrowers bid down rates). Lower rates attract more borrowers, especially high-quality ones who can get bank loans but prefer platform convenience.
Borrower to investor transmission: If platform attracts high-quality borrowers (FICO 720+, stable employment, low debt-to-income), default rates are low (5-8%). Investors earn good returns (6-8% after defaults). Word spreads, more investors join. This increases capital supply further, benefiting borrowers.
Virtuous cycle example—Funding Circle early years: Platform attracted small business borrowers rejected by banks post-2008. Screened aggressively (10% approval rate). Investors earned 6-7% returns. As reputation grew, more investors joined, rates fell slightly, more borrowers applied. Platform grew from £10M originated (2011) to £1bn+ (2016). This is positive feedback in action.
Condition for virtuous cycle: Platform must maintain quality—screen borrowers, price risk accurately, service loans properly. If quality slips (defaults rise), cycle reverses: investors leave → capital dries up → rates rise → only desperate borrowers remain → defaults rise further → death spiral.
Comparison to bank lending: Banks don’t have same cross-side effects because depositors and borrowers don’t interact. Bank sets rates unilaterally based on cost of deposits + target spread. Marketplace platforms’ rates determined by investor-borrower supply-demand balance.
Student discussion: “What could break this virtuous cycle? What risks does a marketplace platform face?” (Credit cycle turns, defaults spike, reputation damaged, regulation changes.)
Assessment (if applicable): You may be asked to explain cross-side network effects in marketplace lending with examples.
Transition: “But positive feedback can turn negative—let’s examine same-side congestion risks.”

Same-Side Congestion

Borrower-side (lemons problem)

Too many risky applicants approved
Defaults soar: 8% → 15%+
Investors flee → capital dries up
Death spiral: only desperate borrowers remain

Investor-side (excess capital)

Too many investors, too few good loans
Rates fall → returns drop 6% → 3%
Investors exit for equities/bonds
Capital supply shrinks

Platform response: Screen aggressively · Risk-based pricing · Diversification tools

Same-side congestion defined: When one side of platform grows too fast or attracts wrong participants, it hurts that side’s value proposition and potentially breaks the platform. Akerlof’s (1970) lemons problem applied to credit markets.
Borrower-side congestion is biggest risk: If platform doesn’t screen, low-quality borrowers flood in. They know they’re risky (information asymmetry) and traditional lenders rejected them (adverse selection). Result: default rates spike from 8% (platform target) to 15-20%. Investors lose money, stop funding new loans, platform revenue collapses.
Lemons death spiral: Bad borrowers → high defaults → investors leave → capital dries up → rates rise (to compensate remaining investors) → only most desperate borrowers apply (even worse quality) → defaults rise further → more investors leave → platform fails. This happened to Lendy (UK, 2019 collapse), thousands of Chinese P2P platforms (2018-2020).
Example—LendingClub 2016 crisis: Platform discovered some loans were mis-sold (borrower information inaccurate, loans didn’t meet investor criteria). Scandal broke, CEO resigned, investors panicked, platform stock fell 60%, lending volume dropped 30%. Took 2 years to recover trust. This shows how fragile platform reputation is.
Investor-side congestion less severe but real: If too many investors compete for loans, they bid down rates. Borrowers pay 8% → 6%. Borrowers benefit, but investor returns fall from 6% (after defaults) → 4%. At 4%, investors earn less than corporate bonds (5%) or stocks (8% long-run). They leave platform, capital supply shrinks.
Platform screening imperative: To prevent borrower-side congestion, platforms must screen aggressively. Funding Circle has publicly cited approval rates of roughly 10% for SME applications (press statements, approximate). LendingClub’s consumer loan approval rate was reported as roughly 10% in SEC filings pre-2020. ⚠️ Both figures are approximate and vary over time and credit cycle — use as illustrative order-of-magnitude only.
Risk-based pricing: Platforms assign risk grades (A-F) and charge higher rates to riskier borrowers (A = 6%, F = 18%). This compensates investors for default risk and deters some high-risk borrowers (who can’t afford 18%). But pricing is imperfect—if platform underestimates risk, investors still lose money.
Diversification as investor protection: Platforms mandate diversification—spread £10k across 100+ loans (£100 each). If one loan defaults (lose £100), others perform (earn interest). Expected return positive if defaults < 10%. But diversification fails in recessions—correlated defaults (many borrowers default simultaneously).
Comparison to bank regulation: Banks must hold capital (reserves) against loan losses. Marketplace platforms don’t—investors bear all losses. This regulatory arbitrage allows platforms to operate with lower costs but creates systemic fragility (if defaults spike, platforms can fail quickly).
Student task: “If you were a marketplace lending platform, what screening criteria would you use? How strict would you be?”
Assessment (if applicable): You may be asked to explain same-side congestion and how platforms mitigate it, or to compare marketplace-lending fragility to bank regulation.
Transition: “Congestion risks are real. Platforms use trust mechanisms to manage information asymmetry and maintain quality.”

Trust Mechanisms

The problem: Borrowers know their repayment ability; investors don’t.
Traditional fix: Relationship lending, collateral, covenants — none scale to a platform.

How platforms respond:

Credit scoring — statistical models predict default probability
Alternative data — employment, cash flow, digital footprint
Transparency — investors see loan purpose and borrower profile
Diversification — spread £10k across 100 loans (reduces idiosyncratic risk)
Reputation — platform brand signals screening quality
Skin in the game — some platforms co-invest to align incentives

Information asymmetry is THE problem in credit markets: Borrowers know if they can repay; lenders don’t. This creates adverse selection (bad borrowers apply more) and moral hazard (borrowers don’t try to repay).
Traditional banks solve via relationships: Branch managers know local customers, monitor over time, can threaten to cut off future credit. But relationship lending doesn’t scale—platforms serve strangers.
Collateral helps: If borrower defaults, lender seizes asset. But most marketplace lending is unsecured consumer credit (no collateral except future income). Business lending may have asset backing, but valuation hard.
Credit scoring—traditional: FICO score (300-850) based on payment history, utilization, credit age, types, inquiries. Predicts default reasonably well for prime borrowers. But 45 million US adults have no credit score (thin file) or low score due to youth/poverty, not risk.
Alternative data improves prediction (Berg et al. (2020)): Berg et al. study digital footprint variables from a German e-commerce platform — device type (smartphone model), email provider (e.g. Gmail vs Yahoo), and shopping behaviour (timing, basket size). Digital footprint alone achieves AUC of 69.6% vs 68.3% for credit bureau score alone — a 1.3 percentage point improvement. Most valuable for the 6% of borrowers who had no credit bureau score at all. ⚠️ Berg does NOT study education, employment tenure, or bank-account cash flow — those are separate data types used by other platforms.
Transparency as trust mechanism: Unlike banks (black box), platforms show investors loan purpose, borrower income, loan amount, interest rate. This builds trust—investors feel in control. But most investors don’t analyze deeply (behavioral bias—overconfidence).
Diversification: Platform advice: spread £10k across 100+ loans (£100 each). If one loan defaults (lose £100), other 99 perform (earn interest). Expected return positive. This works IF defaults are independent. But recessions create correlated defaults—diversification fails when needed most.
Platform reputation: Funding Circle and LendingClub built brand over 10+ years. Investors trust them to screen borrowers. But new platforms face cold-start problem—no reputation, hard to attract investors.
Skin in the game: Some platforms (e.g., Funding Circle initially) co-invested 5-10% of each loan. This aligned incentives—platform suffers if defaults high. But as platforms scaled, co-investment became expensive. Most now don’t co-invest.
Moral hazard problem: Once borrower receives money, incentive to repay may be weak (especially unsecured loans). Platforms use collections, report to credit bureaus (threat of bad credit score), and occasionally sue. But enforcement costly.
Failure of trust: When platforms fail to screen (e.g., Lendy in UK, which funded property developments without proper due diligence), investors lose big. Lendy collapsed 2019, investors lost £165m. This is why governance matters.
Student task: “Rank these trust mechanisms by effectiveness. Which would you rely on as an investor?”
Assessment (if applicable): You may be asked to explain how platforms address information asymmetry.
Transition: “Let’s see how alternative data and machine learning improve credit assessment.”

Part IV — Credit Risk, Adverse Selection, and Alternative Data

Traditional Credit Scoring

FICO score (US, 300-850) based on five factors:

Payment history (35%): Have you paid bills on time?
Amounts owed (30%): How much debt do you carry?
Credit history length (15%): How long have you had credit?
New credit (10%): Recent applications/accounts
Credit mix (10%): Types of credit (cards, loans, mortgage)

UK equivalents: Experian (0-999), Equifax (0-700), TransUnion (0-710)

What it captures: Past credit behavior over 2-7 years (same in UK)

Dominance: 90% of US lenders use FICO; UK lenders use bureau scores

FICO vs UK systems: FICO is US-specific (Fair Isaac Corporation, created 1989). UK doesn’t use “FICO scores”—instead, three main credit reference agencies (Experian, Equifax, TransUnion) each have their own scoring systems with different ranges. Experian uses 0-999 (fair: 721-880, good: 881-960, excellent: 961-999). Equifax uses 0-700. TransUnion uses 0-710. But conceptually, they work the same way—all use payment history, utilization, credit age, etc. So we’ll use FICO as the teaching example, but UK students should understand their credit scores work on similar principles.
Why different scoring ranges? Each agency uses proprietary algorithms and weights factors differently. This is confusing for consumers—you can have “good” score with Experian but “fair” with Equifax. Unlike US (where FICO is standard), UK has no single dominant score. Banks may use one or all three agencies. This fragmentation is a UK-specific issue.
FICO dominance in US: FICO used by 90% of US lenders. Banks love it because it’s standardized, auditable, regulatory-approved. UK lenders similarly rely on bureau scores (Experian/Equifax/TransUnion) for same reasons—regulatory compliance, risk management, standardization.
Five factors explained: Payment history (35%) weighs most heavily—late payments, defaults, bankruptcies hurt score most. Amounts owed (30%) measures utilization (debt / credit limit)—high utilization signals distress. Length of credit history (15%) rewards time—longer history = more data = more reliable. New credit (10%) penalizes “credit shopping” (multiple applications suggest desperation). Credit mix (10%) rewards diversity (mortgage + card + auto loan better than just cards). These factors are essentially the same in UK scoring systems.
Backward-looking by design: FICO and UK scores use 2-7 years of history. This is both strength (lots of data, hard to game) and weakness (past doesn’t always predict future, especially after life changes). This limitation applies equally to all traditional credit scoring—US and UK.
Standardization value: FICO is standardized across all US lenders—640 means same thing at Bank of America as at Chase. UK has less standardization (three different scores), but each agency’s score is standardized across lenders using that agency. This enables comparison, benchmarking, regulatory oversight. But standardization also limits innovation.
Student engagement: “How would you rate your credit behavior on these five factors? Do you think your credit score (whether FICO or Experian/Equifax/TransUnion) would accurately reflect your ability to repay a loan?”
Transition: “Credit scores work well for people with credit history. But what about the 45 million Americans—or millions of UK residents—with no score?”

Who Gets Excluded?

The thin file problem: (Consumer Financial Protection Bureau 2015)

US: 26 million credit invisible + 19 million unscorable (45M total)
UK: Similar issue—millions lack credit history or have thin files
Youth: Haven’t had time to build credit history (US and UK)
Immigrants: No domestic credit history despite good finances (US and UK)
Cash-preferring households: Never borrowed, often low-income (US and UK)

Backward-looking bias:

Past financial trouble (medical emergency, job loss) penalizes for years
Current ability to repay not captured
Recovery from hardship takes 5-7 years to reflect in score

Key omissions (US and UK systems):

Income/employment: Credit scores ignore salary, job stability
Education: College graduates default less (not in scoring models)
Cash flow: Steady income & expenses not visible to bureaus

Thin file exclusion is systemic (US and UK): Consumer Financial Protection Bureau (2015) reports 26 million US adults are “credit invisible” (no credit history with nationwide bureaus) plus 19 million have unscorable records (total 45M). UK faces similar issues—millions lack credit files or have insufficient data. This isn’t random—disproportionately affects: (1) Young adults 18-25 (haven’t had time to build history), (2) Recent immigrants (no domestic credit history despite possibly excellent finances in home country—applies to EU migrants to UK, US migrants to UK, etc.), (3) Low-income households who avoid formal credit (rely on cash, prepaid cards, informal lending). Same patterns in US and UK.
UK-specific context: Unlike US (where FICO dominates), UK has three separate scoring systems (Experian, Equifax, TransUnion). This creates additional confusion—you might have “thin file” with one bureau but not another. Some UK lenders use multiple bureaus to cross-check. But the underlying problem is identical: no credit history → can’t get credit → can’t build history (catch-22).
Exclusion consequences are severe: If you have no score or low score (US FICO < 640; UK Experian < 560), traditional banks won’t lend. Options: (1) Payday lenders (400% APR in US; UK capped at 0.8% daily = ~292% annual but still predatory), (2) Rent-to-own (100%+ effective rates), (3) Borrow from family/friends (strains relationships), (4) Go without (can’t buy car, can’t move for job opportunity). This perpetuates poverty—lack of credit access limits economic mobility. Same cycle in US and UK.
Backward-looking bias example: You lose job in 2018 (recession), miss credit card payments, FICO drops from 720 to 580. You get new job in 2019, income now £40K, can afford payments. But FICO still 580 in 2020 (late payments stay on report 7 years). Banks reject you despite current ability to repay. Your score won’t recover to 700+ until 2024-2025. This 5-7 year lag punishes past hardship even after recovery.
Income omission is puzzling (US and UK): Credit scoring models don’t use income directly. Why? Historical reason—credit bureaus report credit behavior but not income. Income data comes from lenders (reported separately). Scoring models were designed around available bureau data. But this is suboptimal—ability to repay depends on income! Someone with £100K income and £10K debt is safer than someone with £30K income and £10K debt, even if both have similar credit histories. Applies to FICO (US) and Experian/Equifax/TransUnion (UK).
Education predicts default (universal finding): Research shows college graduates default less than non-graduates, controlling for income and credit score (academic studies, though note: Berg et al. 2020 studied digital footprint, not education directly). Why? (1) Higher lifetime earnings, (2) Better financial literacy, (3) Lower unemployment risk, (4) Cultural/behavioral factors (planning, delayed gratification). But credit scoring models (US and UK) ignore education because bureaus don’t collect it. Note: Specific percentages vary by study and context—consult original research for precise estimates.
Employment stability matters (universal): Same job for 3+ years associated with lower default risk than frequent job changes (industry studies). Why? Income stability, less likely to have employment gaps. But credit scoring models ignore this—bureaus don’t track employment duration (US and UK). Note: Specific percentages vary by platform and borrower segment—alternative finance platforms report this correlation but exact magnitude varies.
Cash flow analysis (new data source): If you have steady £2,000/month income and £1,500/month expenses, you have £500 surplus to repay loans. This is “ability to repay” in real time. But traditional credit scores don’t see bank account data. Platforms can—using Plaid (US) or TrueLayer (UK) with permission, they analyze deposits (income), withdrawals (expenses), balance trends. This real-time ability to repay is more predictive than 5-year-old credit card payment history. This innovation applies in both US and UK markets.
Racial/ethnic disparities: FICO scores correlate with race/ethnicity (legacy of redlining, income inequality, systemic barriers). Federal Reserve and CFPB data consistently show significant average score gaps across racial groups — the direction and existence of the gap are well-established. ⚠️ A specific “50-60 points lower” figure is often cited in policy discussions but the exact gap varies by dataset, year, and methodology; do not present it as a precise verified statistic. The key policy point is that FICO is not legally discriminatory (doesn’t use race) yet still perpetuates historical inequalities for groups with limited credit history.
Regulatory protection creates lock-in: Regulators (OCC, FDIC, Fed) require banks to use credit scores for underwriting and to document lending decisions. FICO is regulatory-approved, auditable, defensible. Alternative models (even if better) face regulatory uncertainty—“Will regulators accept this?” This creates lock-in—banks stick with FICO despite limitations.
Student engagement: “If you were designing a credit score from scratch, what would you include? Why?” (Income, employment, education, cash flow, rent payments, utility payments, etc.)
Assessment (if applicable): You may be asked about limits of traditional credit scoring (and how this creates exclusion), and/or to evaluate trade-offs between standardisation and innovation (alternative data).
Transition: “Traditional credit scoring has fundamental limitations. Enter alternative data—using new signals to predict creditworthiness.”

Alternative Data: What Can We Add?

New data sources platforms can use:

1. Digital footprint (Berg et al. 2020)

Device type (iPhone > Android on default rates—income proxy)
Email provider (Gmail vs. Yahoo patterns differ)
Shopping behavior (purchase timing, basket patterns)

2. Bank account data (cash flow analysis)

Income deposits, expense patterns, balance trends
Real-time ability to repay (current income vs. expenses)
Overdraft frequency, regular bill payments

3. Education & employment (used by some platforms)

Degree level, job tenure, industry sector
Verification via payslips or LinkedIn

4. Psychometric testing

Personality/cognitive tests (“Would you return a lost wallet?”)
Controversial and used mainly in emerging markets

Alternative data defined: Information not in traditional credit reports but predictive of credit risk. Includes: digital footprint, bank account data, utility payments, rent payments, education, employment, psychometric traits.
Berg et al. (2020) digital footprint findings: Paper analyzes German e-commerce platform (270k purchases, 2015-2016). Digital footprint variables (device type, email provider, shopping behavior) achieve AUC of 69.6% vs 68.3% for credit bureau score alone—1.3 percentage point improvement. Key insight: digital footprint alone nearly as predictive as traditional credit scores. Most valuable for “unscorable” customers (6% of sample) who lack credit history.
Device type as income proxy: iPhone users default less than Android users (statistically significant). Paper interprets this as proxy for income/wealth rather than causal effect. Gmail users default less than Yahoo users. These digital signals correlate with creditworthiness through income/wealth channels.
Borrowers with no credit history benefit most: For customers with no credit bureau score (6% of sample), digital footprint provides predictive power where traditional data doesn’t exist. This is the inclusion story—alternative data makes lending viable to previously “unscorable” population.
Cash flow analysis (separate from Berg): Some platforms use bank account data (via Plaid/TrueLayer with permission) to analyze real-time ability to repay. Example: Two borrowers both have FICO 680. Borrower A earns £3K/month, spends £2.5K (£500 surplus). Borrower B earns £3K/month, spends £2.9K (£100 surplus, frequent overdrafts). Cash flow analysis reveals A is safer. This is complementary to digital footprint.
Education & employment (separate from Berg): Some platforms ask for education level and employment tenure, though Berg et al. (2020) doesn’t study these variables. Academic research in other contexts suggests education and stable employment predict lower default, but specific percentages vary by study and context.
Psychometric testing controversial: Some platforms (e.g., Lenddo in Philippines, EFL in emerging markets) use personality/cognitive tests. Questions like: “Would you return a lost wallet with £100?” “Are you impulsive?” “Do you plan ahead?” Evidence shows moderate predictive power (~5% improvement). But concerns: (1) Invasive (feels like interrogation), (2) Gameable (people can lie), (3) Culturally biased (questions assume Western norms). Regulators skeptical—hard to audit, potential for discrimination.
Privacy-accuracy tradeoff central: Alternative data improves prediction but raises privacy concerns. Where’s the line? Most people accept: education, employment (publicly verifiable). Many accept: bank account data (opt-in with permission). Few accept: social media, browsing history, psychometric tests (feel invasive). Regulators haven’t decided—unclear what’s allowed under fair lending laws.
Discrimination risk: If alternative data correlates with protected characteristics (race, gender), using it may violate equal credit laws (even if correlation is proxy, not causation). Example: Zip code predicts default (income/wealth proxy) but also correlates with race (redlining legacy). Is zip code allowed? US regulators say “depends on context”—if zip code proxies for race, it’s illegal; if it proxies for income, it’s legal. But how do you distinguish? This is unsettled legal area.
Student discussion: “Would you give a lender access to your bank account data for better rates? What about your social media? Where’s your line?”
Assessment (if applicable): You may be asked what alternative data sources can improve credit-risk prediction, and/or to discuss fairness trade-offs—does alternative data expand inclusion or create new forms of discrimination?
Transition: “Alternative data helps. But does it work? Let’s see the evidence.”

Berg et al. Evidence

Berg et al. (2020) foundational study (Berg et al. 2020):

Analyzed German e-commerce platform with 270,000 purchases
Compared models: traditional data only vs. traditional + alternative data

Key findings:

1.3 percentage point improvement in AUC (69.6% vs 68.3% for credit score alone)
Digital footprint alone nearly as predictive as traditional credit scores
Biggest gains for borrowers with no credit history (6% of sample had no traditional credit score)
Device type, email provider, shopping patterns all predictive of default risk

Implication: Digital footprint data enables lending to previously “unscorable” borrowers (those without credit history) whilst maintaining prediction accuracy

Berg et al. (2020) foundational study: Analyzed German e-commerce platform with 270k purchases (2015-2016). Compared models using only traditional credit bureau score vs. digital footprint variables. Key finding: digital footprint alone achieves AUC of 69.6% vs 68.3% for credit score alone—1.3 percentage point improvement. This is remarkable because digital footprint performs nearly as well as traditional credit data.
What Berg actually studies: Device type (smartphone model), email provider (Gmail vs Yahoo), shopping patterns (purchase timing, basket size), account age, delivery preferences. These “digital footprint” variables proxy for income and creditworthiness. Paper does NOT study education, employment, or cash flow directly.
Limited credit history insight: 6% of sample had no credit bureau score (“unscorable”). For these customers, digital footprint provides predictive power where traditional data doesn’t exist. This is the key inclusion benefit—alternative data enables lending to previously excluded population.
Income proxies: Paper finds device type and email provider are likely proxies for income/wealth. iPhone users and Gmail users have lower default rates. Shopping behavior (timing, patterns) also correlates with creditworthiness.
Psychometric testing: Some platforms (e.g., Lenddo in Philippines, EFL in emerging markets) use personality/cognitive tests. Evidence shows moderate predictive power. ⚠️ “~5% improvement” cited elsewhere in the deck is an approximate characterisation from practitioner literature, not a verified academic estimate — treat as illustrative only. Concerns: invasive, gameable, culturally biased.
Digital footprint signals: Device type (smartphone model), email provider, browser, time-on-application, shopping timing. Berg finds these proxy for income and creditworthiness. Raises fairness questions — should owning an iPhone affect credit access?
Gains are largest for unscorable borrowers: Berg reports the 1.3pp AUC improvement for the full sample, but notes the effect is especially valuable for the 6% with no credit bureau score. ⚠️ Specific improvement percentages like “10% for FICO borrowers, 40% for no-FICO borrowers” are NOT precise figures from Berg’s paper — do not present these as verified statistics.
Machine learning vs. traditional regression: Logistic regression (traditional) works well with 10-20 features. ML (random forests, gradient boosting, neural nets) works better with 100+ features (can find complex interactions). Berg et al. show ML improves prediction by 5-10% beyond logistic regression when using alternative data.
Privacy-fairness tradeoff: Using granular data improves prediction but raises concerns. Should lenders use Facebook likes? Browsing history? Where’s the line between predictive and invasive? Regulators haven’t decided.
Discrimination risk: If alternative data correlates with protected characteristics (race, gender), using it may violate equal credit laws (even if correlation is proxy, not causation). Example: Zip code predicts default (income/wealth proxy) but also correlates with race (redlining legacy). Is zip code allowed? Unclear.
Student discussion: “Would you give a lender access to your bank account data for better rates? What about your social media?”
Assessment (if applicable): You may be asked to explain how alternative data improves credit assessment and to discuss fairness trade-offs.
Transition: “Alternative data helps but isn’t magic. We need rigorous validation—that’s where statistical science from Week 1 comes in.”

🔬 In-Class Exercise A: Meet the Data

~5 minutes — open Lab 6, run Sections 0 & 1

Important

We use the UCI German Credit dataset — 1,000 real loan applications with 20 features.
This is the same data structure a marketplace lending platform would use.

Run the setup and explore cells. Then answer:

What is the default rate in this dataset?
Which features are numeric? Which are categorical?
Do borrowers who default tend to have longer or shorter loan durations?

Statistical Approach

Connection to Week 1 foundations:

Alternative data improves prediction—but how do we know if our model works?

Statistical challenges from Week 1:

Bias-variance tradeoff: More features reduce bias but risk overfitting (high variance)
Validation strategy: Single train/test split unreliable (Week 1, §0.6)
Base rate fallacy: Accuracy is useless for rare events (Week 1, §0.8.3)
Model uncertainty: Point estimates hide uncertainty (Week 1, §0.2)

What we’ll demonstrate:

Cross-validation (proper validation, not single split)
Regularization (manage bias-variance tradeoff)
Model diagnostics (calibration, ROC curves, precision-recall)

Real-world data in this deck

Evidence slides use real studies: Berg et al. (2020) 270k German e-commerce; CCAF global/UK/US volumes; CFPB thin-file 45M; platform-origination stats (LendingClub, Funding Circle, Zopa). Code demos below use illustrative synthetic data (same structure as Lab 6) so they run without external datasets; for real analysis use LendingClub public data, FCA P2P statistics, or course materials.

Connection to Ch 01: Statistical Foundations

This is Week 1 applied: bias-variance tradeoff, cross-validation, bootstrap uncertainty quantification, base rate fallacy. Credit scoring demonstrates why statistical foundations matter in practice.

Cross-Validation

Problem: Single random split is unreliable — results vary by chance

Example: Credit scoring with 5,000 loans, 10% default rate (illustrative synthetic data; in practice use platform data e.g. LendingClub public data or lab/course sample matching Berg et al. structure).

Key insight: Cross-validation provides stable estimates + quantifies uncertainty

Connection to Week 1, §0.6: Model Selection & Validation

Single train/test split gives one realization of random data splitting. 5-fold CV gives five independent estimates → more reliable performance assessment and uncertainty quantification.

🔬 In-Class Exercise B: Build and Validate

~8 minutes — Lab 6, Sections 3–5

Work through:

Section 3 — fit numeric-only logistic regression. What is the AUC?
Section 4 — add categorical features. How much does AUC improve?
Section 5 — run 5-fold cross-validation. Report: mean ± std

Discuss with a neighbour:

“Your CV mean is lower than your single-split AUC. What does that tell you?”

Regularisation

Bias-variance tradeoff (Week 1, §0.2):

High bias: Model too simple (underfits) — misses patterns
High variance: Model too complex (overfits) — learns noise

Credit scoring challenge: 50+ features (credit score, income, DTI, education, employment, cash flow, digital footprint) → overfitting risk

Solution: Regularization (L1 Lasso, L2 Ridge)

Key insight: Regularization reduces overfitting → better out-of-sample performance

Connection to Week 1, §0.2: Bias-Variance Tradeoff

L1 (Lasso): Feature selection — sets weak coefficients to zero (reduces variance, increases bias slightly)
L2 (Ridge): Shrinkage — reduces all coefficients (smooth tradeoff)

Both manage complexity → prevent overfitting when you have many features relative to samples.

Model Diagnostics: Calibration Plots

Question: Does “20% default probability” actually mean 20% of borrowers default?

Calibration check: Compare predicted probabilities to observed frequencies

Well-calibrated model: Points lie on diagonal (predicted = observed)
Poorly calibrated: Model overconfident or underconfident

Why calibration matters for credit scoring: Platforms use predicted probabilities to set interest rates. If model says 15% default risk but true risk is 25%, investors lose money!

ROC & Uncertainty

ROC curve: True Positive Rate vs False Positive Rate at all thresholds

But: Single ROC curve hides uncertainty from finite sample

Solution: Bootstrap resampling (Week 1, §0.2) to quantify uncertainty

Key insight: AUC = 0.72 ± 0.03 is more honest than AUC = 0.72 (hides uncertainty)

Connection to Week 1, §0.2: Bootstrap Uncertainty

Bootstrap resampling creates 100 “plausible” test sets → 100 AUC estimates → confidence interval. This quantifies model uncertainty from finite data.

Precision-Recall

Problem: Credit defaults are rare events (~10% of borrowers)

ROC curves can be misleading for imbalanced data (Week 1, §0.8.3: Base Rate Fallacy)

Precision-Recall curves: Focus on positive class (defaults)

Precision: Of borrowers we flag as risky, what % actually default?
Recall: Of borrowers who default, what % do we catch?

Key insight: For rare events (defaults, fraud), precision-recall more informative than ROC

Connection to Ch 05: Type I/II Error Tradeoff

High precision: Few false alarms (reject only truly risky) → more loans approved (inclusion)
High recall: Catch most defaults → protect investors

Platform must choose cost-sensitive threshold based on business costs of each error type.

Selection Bias: What We Don’t See

Critical problem: Credit models trained on approved loans only

Missing data: Rejected applicants (who platform judged too risky)

Selection bias consequence: Model underestimates default risk

Below: illustrative simulation; same pattern appears in real platform data (approved vs full applicant pool).

Why this matters:

Platform relaxes lending standards (approves more borderline cases)
Default rate jumps from 14% to 18% (approaching true 20%)
Investors surprised, lose money, platform fails

Connection to Week 2, §2.3: Selection Bias

Models trained on selected samples (approved loans) don’t generalize to full population (all applicants). This is survivorship bias in credit markets—we only see repayment behavior of borrowers we approved.

Part IV Summary

What we demonstrated:

Cross-validation (Week 1, §0.6): Stable estimates + uncertainty quantification
Regularization (Week 1, §0.2): Manage bias-variance tradeoff with many features
Calibration: Predicted probabilities match observed frequencies (critical for pricing)
Bootstrap uncertainty (Week 1, §0.2): Honest confidence intervals for AUC
Precision-recall (Week 1, §0.8.3): Focus on rare events (base rate fallacy)
Selection bias (Week 2, §2.3): Missing data (rejected applicants) distorts estimates

Key lesson: Alternative data improves prediction (Berg et al. (2020)), but rigorous validation (Week 1 foundations) separates good models from dangerous ones.

Next: Adverse selection — even with better data, bad borrowers still apply more

🔬 In-Class Exercise C: Investor Returns

~8 minutes — Lab 6, Section 7

Run the risk grades and investor returns cell. Then answer:

Grade	Interest Rate	Your Net Return
A	6%	?
B	10%	?
C	15%	?
D	22%	?

Which grade gives the best net return for investors?
Why do Grade D loans disappoint despite the highest interest rate?
What does this imply about “high-yield” lending as a strategy?

Adverse Selection

Why it happens: Bad borrowers know their risk; good borrowers have better options.

Evidence

Platform defaults: 8–12% vs banks 2–5% (LendingClub 2007--2020)
Platform avg FICO: 680 vs prime bank: 760
Akerlof’s “lemons” — bad drives out good

Platform responses

Screen hard (approve 10–20% only)
Risk-based pricing (charge higher rates)
Diversification tools
Co-investment (rare now)

Adverse selection classic problem (Akerlof 1970, “Market for Lemons”): When sellers know more than buyers, bad products drive out good. In credit: bad borrowers know they’re risky, apply more; good borrowers know they can get traditional loans, avoid platforms. Result: platform gets adverse mix.
Empirical evidence: Marketplace lending default rates 8-12% (LendingClub 2007--2020). Traditional bank unsecured loans default 2-5% (industry data). Why the gap? Mix of borrowers—platforms get near-prime and subprime (FICO 640-720), banks get prime (FICO 760+).
Cherry-picking by traditional lenders: Banks skim the cream—best borrowers with high FICO, low debt-to-income, stable employment. Platforms get leftovers—borrowers banks rejected or underprice. This is selection bias.
Behavioral sorting: Good borrowers are risk-averse—they prefer bank brand, FDIC insurance, branch access. Bad borrowers are desperate—they accept higher platform rates because they have no alternative. This behavioral difference amplifies adverse selection.
Platform screening: To combat adverse selection, platforms must screen aggressively. Funding Circle approves ~10% of small business applications (platform reported data). LendingClub approved ~8-10% of consumer applications pre-2020 (platform data, before tightening). This filters out worst borrowers but also reduces volume (costs money to process rejections).
Risk-based pricing: Platforms charge 5-15% interest to compensate for default risk. But if risk is underestimated (model error), investors lose money. LendingClub repriced risk grades in 2016 after realizing defaults higher than predicted—investors angry.
Diversification mandate: Platforms tell investors to spread £10k across 100 loans. This is Diversification 101—reduce idiosyncratic risk. But it doesn’t eliminate systematic risk (economy-wide shocks). In COVID-19, many loans defaulted simultaneously—diversification didn’t save investors.
Co-investment declined: Early P2P platforms co-invested (Funding Circle put £100k of own money alongside investors). This signaled confidence and aligned incentives. But as platforms scaled, co-investment became expensive (ties up capital, hurts ROE). Most platforms stopped. Result: incentive misalignment—platform profits from origination fees even if loans default.
Death spiral risk: If defaults exceed expectations, investors lose money, word spreads, new investors stop joining, existing investors flee, capital dries up, only desperate borrowers remain, platform fails. This happened to Lendy (UK), Funding Circle’s 2019 troubles (stock fell 90%).
Regulatory response: In some jurisdictions (e.g., UK FCA), platforms must warn investors: “Capital at risk. Default rates may be higher than disclosed. No FSCS protection.” This is disclosure-based regulation—inform but don’t prohibit.
Student task: “If you were a marketplace lending platform, how would you screen borrowers to avoid adverse selection without rejecting too many?”
Assessment (if applicable): You may be asked to analyse adverse selection in alternative finance and evaluate mitigation strategies.
Transition: “Adverse selection is one challenge. Now let’s discuss inclusion and regulation.”

Part V — Inclusion, Regulation, and Python Implementation

Inclusion: Evidence

Optimistic

15M+ borrowers banks rejected (platform data)
Rates 5–10% below credit cards (15–20%)
Credit history built via bureau reporting
Alternative data unlocks thin-file borrowers (Berg et al. 2020)

Skeptical

Still requires: smartphone, bank account, digital literacy
Poorest excluded — no income to repay
8–12% default rate worsens financial position (LendingClub 2007--2020)
Digital divide: skews urban, educated, younger

Inclusion narrative central to FinTech: Alternative finance platforms market themselves as democratizing credit—serving excluded borrowers. Is this true?
Optimistic evidence—access: LendingClub, Prosper, Funding Circle collectively served 10M+ borrowers in US/UK (2006-2020, platform cumulative data). Many had FICO 640-720 (near-prime)—banks would reject or charge 20%+. Platforms charged 8-15% → savings of 5-10 percentage points. For $10k loan over 3 years, that’s $1,500 saved (illustrative calculation).
Optimistic evidence—rates: Marketplace lending rates (8-15%) much lower than payday loans (400% APR) or rent-to-own (100%+ effective rates). For borrowers with no alternatives, this is genuine improvement.
Optimistic evidence—credit building: Marketplace loans reported to credit bureaus. If borrower repays on time, credit score improves (adds payment history). Some borrowers “graduate” to prime (FICO > 760) and qualify for bank credit. This is mobility.
Alternative data benefit: Berg et al. (2020) show alternative data most helps borrowers with no traditional credit history. Without alternative data, these borrowers rejected; with it, some become lendable. This is inclusion at the margin. Note: The 6% of sample with no credit score gained most predictive value from digital footprint data.
Skeptical evidence—still excludes poorest: You need income to repay loans. Households with no income (unemployed, disabled, retired with no savings) can’t borrow—platforms reject them too. Alternative finance doesn’t solve poverty, just shifts margin slightly.
Skeptical evidence—digital divide: Marketplace lending requires internet, smartphone, bank account, email, digital literacy. Rural, elderly, low-education borrowers may lack these. Studies show marketplace lending usage higher in urban, educated, younger demographics—replicates existing inequality.
Skeptical evidence—default harms: 8-12% of marketplace borrowers default. For them, alternative finance made things worse—they borrowed, couldn’t repay, now have bad credit, debt collectors, stress. Is it better to be excluded or included and harmed? Depends on counterfactual (what would they have done otherwise—payday loan? No borrowing?).
Behavioral concerns: Marketplace platforms use behavioral nudges (e.g., “pre-approved, just click to accept”). Some borrowers overborrow (don’t fully understand cost, risk). This is predatory lending lite—not illegal, but exploitative.
Comparison to microfinance: Microfinance (Grameen Bank, etc.) promised inclusion for poor in developing countries. Mixed evidence—some borrowers benefited, many didn’t (debt traps). Alternative finance may follow similar path—helps some, harms others, overhyped.
Student discussion: “Is access to credit always good? Could exclusion sometimes protect people?”
Assessment (if applicable): You may be asked to critically evaluate inclusion claims: use evidence, and discuss who benefits versus who remains excluded.
Transition: “Inclusion benefits exist but are limited. Regulators face tradeoffs.”

Regulation

UK FCA

Light-touch 2011–2018 (encourage innovation)
Disclosure-based: warn investors, disclose defaults
2019 crackdown after Lendy collapse (£165m lost)
Tighter: appropriateness checks, £10k exposure cap

US SEC

Reg CF: equity raises up to $5M/year
Investor limits: £2,200 or 5% of income
State-by-state variation for P2P lending
CFPB oversight reduced post-2020

Core tension: innovation vs. consumer protection · disclosure vs. prohibition

Innovation-protection tradeoff central: Regulators want innovation (competition, efficiency, inclusion) but must protect consumers (asymmetric info, behavioral biases, fraud risk). How to balance?
UK light-touch rationale (2011-2018): FCA wanted to encourage FinTech. Approach: minimal entry barriers, disclosure-based regulation (warn investors, let them decide), self-certification (investors claim to be sophisticated). Result: rapid growth (£10bn+ lent 2011-2019).
UK failures exposed weaknesses: Lendy collapsed 2019 (£165m lost), Collateral collapsed 2019 (£40m lost), Funding Circle struggled (stock fell 90%). Problem: inadequate due diligence, borrower quality poor, platforms prioritized volume over risk management. Investors blamed FCA for light touch.
UK tightening (2019-2020): FCA imposed: (1) Stronger risk warnings, (2) Appropriateness checks (test investor knowledge), (3) Limit exposure (£10k max unless sophisticated), (4) Stress testing (platforms must show they can survive downturn). Result: some platforms exited, industry consolidated.
Disclosure vs. prohibition debate: FCA chose disclosure (warn investors, let them choose). US often chooses prohibition (ban products deemed too risky—e.g., payday loans in some states). Which is better? Disclosure preserves choice but assumes rationality (behavioral economics says people ignore warnings). Prohibition protects but reduces access (even for sophisticated investors).
Bank-like regulation question: Should platforms be regulated like banks? Banks must: hold capital (reserves), get FDIC insurance, submit to prudential regulation, follow fair lending laws. Platforms argue: we don’t take balance-sheet risk (investors bear losses), we shouldn’t face bank capital requirements. Regulators’ concern: platforms are systemically important (if they fail, credit dries up), should face commensurate oversight.
Fair lending laws apply: US Equal Credit Opportunity Act (ECOA) and Fair Lending Act apply to marketplace lenders. Can’t discriminate by race, gender, age. But enforcement weak—hard to prove discrimination when decisions are algorithmic. Proxy discrimination (using zip code, education) legal gray area.
Equity crowdfunding highly regulated: Selling securities requires SEC registration. Reg CF (2016) created exemption: companies can raise $5M/year from public, must disclose financials, investors limited to $2,200 or 5% income (whichever greater). This is compromise—allow innovation, limit harm.
State variation (US): Some states (e.g., Arkansas) ban payday lending; others (e.g., Utah) allow it. Marketplace lending faces similar patchwork—platforms must get licenses in each state. Costly, slows growth.
International divergence: China had massive P2P boom (2013-2018), then bust (2018-2020)—thousands of platforms failed, millions of investors lost money. Government cracked down, now heavily regulated. Europe has diverse approaches—UK liberal, Germany restrictive.
Student task: “If you were a regulator, would you allow marketplace lending with light touch (disclosure) or ban high-risk products (prohibition)? Why?”
Assessment (if applicable): You may be asked to evaluate regulatory approaches in alternative finance: discuss innovation versus protection trade-offs (and differences across jurisdictions).
Transition: “Now let’s implement credit scoring and platform economics in Python.”

Lab 6: Credit Risk Scoring — Complete the Lab

Using the UCI German Credit dataset (1,000 real loans):

Section	Task	Key skill
0–1	Load & explore	Data literacy
2	Encode features	Categorical handling
3–4	Baseline → richer model	Model comparison
5	Cross-validation	Honest evaluation
6	Calibration	Probability trustworthiness
7	Risk grades & returns	Platform economics
8	Fairness (extension)	Critical reflection

Deliverable: Complete sections 3–7, answer the embedded reflection questions (150–250 words each).

Lab structure: Five tasks mirroring lecture content. Build credit scoring model, evaluate, connect to economics/policy.
Task 1 (load data): Use sample marketplace lending dataset (LendingClub public data or course sample). Features: credit score, income, loan amount, purpose, employment, education.
Task 2 (baseline model): Logistic regression using only traditional features (credit score, income). Measure performance (AUC ~0.65-0.70 typical).
Task 3 (alternative data): Add employment stability, education, loan purpose. Re-train model. Measure performance improvement (AUC ~0.72-0.75). This demonstrates Berg et al.’s finding.
Task 4 (economics analysis): Calculate investor returns across risk grades (A-F). Plot default rates, interest rates, net returns. Show risk-return tradeoff. Connect to platform pricing logic.
Task 5 (inclusion reflection): Who benefits from improved credit scoring? Who’s still excluded? Discuss fairness—is it okay to use education if it correlates with socioeconomic status?
Timing: 60-90 minutes in lab; 60-90 minutes directed learning extensions.
Assessment link: This prepares you for questions that combine platform economics, credit risk, and evidence-based evaluation.
Student support: TAs will circulate. Provide starter code and hints.
Transition: “Start with Task 1 in lab. Bring questions to seminar.”

Quick Check-In Questions

Q1: What is adverse selection in marketplace lending, and how do platforms address it?
Q2: Name one way alternative data improves credit risk prediction.
Q3: Why might equity crowdfunding be riskier for investors than rewards crowdfunding?

A1: Adverse selection = bad borrowers apply more (they know they can’t get traditional loans). Platforms address via screening (approve only 10-20%), risk-based pricing (charge high rates), diversification (spread risk), co-investment (rare now).
A2: Accept any one well-explained example: (a) Digital footprint — Berg et al. (2020) show device type, email provider, and shopping patterns achieve AUC of 69.6% vs 68.3% for credit bureau score alone, with the biggest gains for borrowers with no credit history; (b) Cash flow analysis — real-time bank account data captures ability to repay better than historical credit scores; (c) Employment stability — tenure at current employer correlates with lower default risk (general finding, not Berg specifically). ⚠️ Do NOT cite specific percentages (30%/20%/15%) for education/employment — these are not verified findings.
A3: Equity crowdfunding involves buying shares in startups—high failure rate (~50%+ fail within 5 years), illiquidity (can’t sell easily), valuation uncertainty (hard to price). Rewards crowdfunding is pre-purchase—risk is product non-delivery (~9% failure rate), but limited financial loss (smaller amounts, no equity stake).
Why ask these?: Reinforces key concepts, checks understanding.
Follow-up: If students struggle, briefly recap. If they answer well, affirm.
Timing: 2-3 minutes.

Key Takeaways

1. Alternative finance includes crowdfunding (rewards, equity, debt) and marketplace lending

2. Marketplace lending operates as two-sided platform—investors ↔︎ borrowers with cross-side effects and congestion risks

3. Information asymmetry and adverse selection are central challenges; platforms use credit scoring, alternative data, diversification to mitigate

4. Alternative data (education, employment, cash flow) improves credit prediction, especially for borrowers with limited credit history

5. Inclusion benefits exist but limited—alternative finance helps near-prime borrowers, not poorest; default rates higher than banks

6. Regulation evolves—UK light-touch then tightening, US patchwork, key tension is innovation vs. consumer protection

Recap structure: Six key takeaways mapping to learning objectives.
Takeaway 1: Taxonomy—understand different models and risk profiles.
Takeaway 2: Platform economics applies: cross-side effects, same-side congestion, pricing asymmetry.
Takeaway 3: Information asymmetry is THE problem in credit. Trust mechanisms vary by model.
Takeaway 4: Berg et al. (2020) evidence is crucial—alternative data works, especially for excluded borrowers.
Takeaway 5: Inclusion narrative is partly true—helps some, not all. Critical evaluation needed.
Takeaway 6: Regulators face hard tradeoffs—no perfect solution.
Return to learning objectives: “Look back at objectives. Can you explain each?”
Next week preview: “Week 7 bridges linear models to machine learning thinking, and previews the three CW2 scaffolds.”
Final engagement: “Would you invest in marketplace lending? Back a Kickstarter? Why or why not?”
Closing: “Alternative finance is a mixed bag—genuine innovation helping some, overhyped and risky for others. Critical thinking essential.”

References

See chapter bibliography for full citations.

Core readings:

Mollick (2014) — Kickstarter success factors and social proof
Berg et al. (2020) — Alternative data and credit risk assessment
LendingClub, Funding Circle data (public filings)

Mollick 2014: “The Dynamics of Crowdfunding” in Journal of Business Venturing. Analyzed 48k Kickstarter projects. Key findings: early momentum, social proof, reputation predict success.
Berg et al. 2020: “On the Rise of FinTechs—Credit Scoring using Digital Footprints” in Review of Financial Studies. German P2P lending data. Shows alternative data reduces prediction error 15-25%.
LendingClub data: Public company filings (S-1, 10-Ks) available on SEC website. Detailed loan-level data (2007-2018) publicly available. Default rates, interest rates, borrower characteristics.
Additional reading: Iyer et al. (2016) on herding in P2P lending, Balyuk (2019) on financial innovation and borrower outcomes, Philippon (2016) on FinTech and finance costs.
Accessing papers: All in reading list on Moodle or via library.
Student task: “Start with Mollick (crowdfunding), then Berg et al. (credit risk).”

Berg, Tobias, Valentin Burg, Ana Gombović, and Manju Puri. 2020. “On the Rise of FinTechs: Credit Scoring Using Digital Footprints.” Review of Financial Studies 33 (7): 2845–97. https://doi.org/10.1093/rfs/hhz099.

BrewDog. 2009--2023. “Equity for Punks: BrewDog Crowdfunding Campaigns.” Company data and platform disclosures.

———. 2017. “TSG Consumer Partners Investment in BrewDog.” Press releases and financial media reports.

Cambridge Centre for Alternative Finance. 2018. “The Global Alternative Finance Market Benchmarking Report.” Industry Report. Cambridge Judge Business School, University of Cambridge. https://www.jbs.cam.ac.uk/faculty-research/centres/alternative-finance/publications/the-global-alternative-finance-market-benchmarking-report/.

———. 2020. “The Global Alternative Finance Market Benchmarking Report: Trends, Opportunities and Challenges.” Industry Report. Cambridge Judge Business School, University of Cambridge. https://www.jbs.cam.ac.uk/faculty-research/centres/alternative-finance/publications/.

Consumer Financial Protection Bureau. 2015. “Data Point: Credit Invisibles.” Government Report. Consumer Financial Protection Bureau. https://www.consumerfinance.gov/data-research/research-reports/data-point-credit-invisibles/.

Gompers, Paul, Anna Kovner, Josh Lerner, and David Scharfstein. 2010. “Performance Persistence in Entrepreneurship and Venture Capital.” Journal of Financial Economics 96 (1): 18–32. https://doi.org/10.1016/j.jfineco.2009.11.001.

Kickstarter. 2024. “Kickstarter Stats.” Platform public data. https://www.kickstarter.com/help/stats.

LendingClub. 2007--2020. “LendingClub Statistics.” SEC filings and company data. https://www.lendingclub.com/info/download-data.action.

Mollick, Ethan. 2014. “The Dynamics of Crowdfunding: An Exploratory Study.” Journal of Business Venturing. https://doi.org/10.1016/j.jbusvent.2013.06.005.

Punks With Purpose. 2021. “Open Letter: Punks with Purpose.” Public letter signed by 300+ former and current BrewDog employees.