Saturday 16 October 2021

Technology and Human Bias (1) -- Robo-advising reduces culture biases!

In this new post-corona era, there is a rising interest in technology’s impact on biases in human decisions. Do few or no human interactions reduce and even eliminate culture, racial, and other taste-based biases? In this series of blogs, I hope to provide more understanding on this question by discussing several recent research papers.

The first paper to discuss is D'Acunto et. al. (2020) where they find that automated robo-advising lending tools can reduce culture biases in the peer-to-peer lending process. Moreover, the reduction of cultural biases actually provides sizable economic returns to investors! Discriminating lenders face 32% higher default rates and about 11% lower returns on the loans they issued.

As their study uses a leading P2P platform in India, Faircent, let's first have a brief idea of the cultural bias in India.

https://www.culturalsurvival.org/publications/cultural-survival-quarterly/ethnic-and-religious-conflicts-india

https://www.pewresearch.org/fact-tank/2021/06/29/key-findings-about-religion-in-india/

https://en.wikipedia.org/wiki/Shudra


Their main findings are two folds. 

Before investors use robo-advising, they detect two spiking patterns that are indications of culture-based discrimination in the lending process.

  • Both Hindu and Muslim lenders tend to favor borrowers of their same religious relative to borrowers of the other group.
  • Lenders are less likely to lend to borrowers who tend to be considered as Shudra borrowers.
After adopting robo-advising, these patterns of cultural discrimination are largely reduced. The share of borrowers of different religions are equated for Hindu and Muslim lenders. Similarly, the share of Shudra borrowers increases substantially.

You might wonder whether it is a good thing to have a more equalized lending process? Maybe it is not cultural discrimination, but just a rationalized choice that gives them a higher return? In the economic theories, this is called "Statistical Discrimination" which can at least date back to Arrow (1971) and Phelps (1972). However, the study shows that it is not true at least in their setting. Before the adoption of robo-advising, borrowers with the same region are more likely to default. Shudra borrowers' loans do not perform worse than other loans. If anything, they are less likely to default.


Reference
Kenneth Arrow, 1971. "The Theory of Discrimination," Working Papers 403, Princeton University, Department of Economics, Industrial Relations Section.

D'Acunto, F., Ghosh, P., Jain, R., & Rossi, A. G. (2020). How Costly are Cultural Biases?. Available at SSRN 3736117.

Phelps, E. S. (1972). The statistical theory of racism and sexism. The American Economic Review, 62(4), 659-661.

Tuesday 20 March 2018

Cryptofinance I – Campbell R. Harvey -- How bitcoin works?


This blog is based on the lecture notes of Professor Campbell R. Harvey in Fuqua School of Business, Duke University.

Bitcoin is the leading cryptocurrency, with a market capitalization 30 times as large as the second one (Ripple). Actually, there are over 500 other crypto-currencies and the first one came out in the 1980s. What’s special about bitcoin? How it works?

Decentralization
Conventional financial intermediates are centralized and for-profit business. In contrast, bitcoin and others are for peer-to-peer (P2P) transactions. They don’t need financial intermediates and the transactions are done directly between the involved two parties.  The conventional financial intermediates build up (also control) the trust system. In the world of crypto-currency, the trust system is built upon cryptographic algorithms, not the governments, nor corporations.

Double Spend Problem and Triple-Entry Accounting
The early cryptocurrencies failed because of the “double spend problem”. Think, when we transmit a document, we are not actually transmitting the document itself, but a copy of it. This doesn’t matter for documents. However, this would be a nightmare for transmit assets, such as currency, intellectual property rights and other rights. The meaning of transmit a sum of money from me to you is that I cannot use it anymore and cannot do a same transaction later.
The way bitcoin solves this “double spending problem” is the “triple-entry accounting”. In addition to the traditional double entry accounting, there is a third entry. Every transaction goes into a repository of common knowledge, which is called the “blockchain”. It could be thought as a public ledger. Blockchain is highly secure and maintained by everyone on the network.

How is new bitcoin created?
Every time someone wants to send bitcoins to some else, a transaction occurs and it will be recorded in the blockchain. Every 10 minutes, a new block contained transactions in the preceding 10 minutes is created. Miners compete in solving cryptographic “proof of work” to add a new block to the chain. Only the first miner who solves the problem get new bitcoins. As the difficulty of solving the code increases with more codes being broken, the production speed of bitcoins will be slowed down.

Security
It is blockchain network that validate transactions, not any tradition intermediates. Suppose Alice wants to buy something from Bob using 1 bitcoin. The network will first check if Alice has 1 bitcoin to transfer. If so, the transaction occurs and it is recorded to the blockchain. Each block and its previous blocks are sealed by miners with s computational code.  Contrary to traditional recording keeping, it is the transactions, not the balances that are kept in the blockchain.

Privacy and Transparency
How could we maintain privacy and transparency at the same time? Each transaction involves a public+private key. For the Alice-Bob example above, Bob sets up a digital public key, say a QR code, and sends it to Alice. Alice signs the transaction by using a private key that is unforgeably tied to the public key. The linkage is to be solved by the miners. After it is solved, the block is sealed and added to the blockchain, which becomes common knowledge on the networks. This common knowledge will be used when next time Bob wants to spend this 1 bitcoin.