Optimal calibration of a ML classifier based on business knowledge
Clasificación óptima bajo decisiones de negocio Optimal calibration of a classifier based on business knowledge Versión en Español Calibrating a predictive algorithm for production in a business environment requires not only consideration of the algorithms' performance, underlying data, and related statistics, but also an economic evaluation of the related business-related actions that the algorithm will trigger. In my experience, this is a highly relevant topic but one that is not frequently considered or discussed. As a result of this, in many applications classifiers are configured without adequate consideration of business trade-offs, which is why I decided to write this post. To exemplify, consider a financial institution which is implementing a classifier (such as Logistic Regression classifier) to prevent fraudulent transactions. Of course, a fraud involves costs that the financial institution seeks to reduce. The classifier algorithm decides if each transaction that takes place in the system should be flagged as a possible fraud. Typically, such a flag triggers a series of actions that will be taken by the company, and that will also carry associated costs. What we will see next is that such costs need to be taken into account in order to adequately calibrate a predictive model. Suppose, to begin…
Note on AMMs “picked-off” risk
AMMs picked-off risk Note on AMMs "picked-off" risk It has been popularized the term "Impermanent loss" (IL) to refer to costs incurred by liquidity providers (LPs) of an AMM pool in the case relative market prices change, and those changes are profited out by arbitrageurs. This Twitter thread by @AnthonyLeeZhang and @guil_lambert discuss that a more appropriate term for this loss is "picked-off" risk. In my understanding (thanks to discussions with Javier Garcia Sanchez), IL is not the best term and below are my notes of why. IL is also referred as an "opportunity cost", meaning that LPs would have been better staying out of the AMM in such a case. I tend to think that the term "opportunity cost" idea is neither adequate. It is my understanding that "picked-off risk" or "picked-off loss" are right terms. As mentioned, this situation takes place when: i) the relative prices of the assets of interest (i.e., those in the pool) change outside the AMM (in the "market" or centralized exchange of reference), and ii) an arbitrageur takes advantage of the price differential (between the one provided by the AMM and market) to obtain a benefit. My understanding is that this is indeed…
Bienes públicos, Gitcoin, y financiamiento cuadrático
english Gitcoin – blog post Bienes públicos, Gitcoin, y financiamiento cuadrático1 Hay algo que tienen en común las inversiones que se realizan en las ciudades, en la actividad de ciertas startups, y el software open source gratuito. La inversión en la generación de un espacio verde, un museo, un espacio de arte, reporta un beneficio no solo para los beneficiarios inmediatos, sino que también vuelven más atractivo el barrio, la ciudad en cuestión, etc. Típicamente, este fenómeno puede verificarse en el incremento del valor de las propiedades cercanas. En el caso de las startups, si la inversión en una de ellas, por ejemplo, hace que ésta pueda innovar exitosamente en un modelo de negocio, este será luego copiado por muchas empresas, cuyos valores también se incrementarán en consecuencia. El último ejemplo es la inversión en código abierto (gratuito). Se estima que el 99% de las aplicaciones desarrolladas por las compañías actualmente contienen software open source, alcanzando hasta un 70% del código de esas aplicaciones (Synopsis, 2020). Esto implica un ahorro muy importante por parte de las compañías gracias a la utilización de este tipo de código, que fue posibilitado gracias a la inversión de compañías o grupos de desarrolladores pioneros2.…
An indebtedness atlas for Argentina
The indebtedness atlas An indebtedness atlas for Argentina The level of indebtedness of the general population has recently been signaled by national authorities [as one of the most pressing problems in Argentina1. While the country ranks relatively low in terms of financial inclusion, for example, with less than 10% of adults borrowing from a traditional bank in 2017 (Demirguc-Kunt, et. al, 2018), there is also an active debate on the role of high-interest rates loans, such as payday loans2. While the literature has shown that neighborhoods matter for upward income mobility or in shaping children outcomes (Chetty et al. 2020), the relationship between neighborhoods and access to credit or indebtedness has remained relatively unexplored. Which neighborhoods in Argentina are the most problematic in terms of financial distress of their population? Which neighborhoods exhibit signals to be credit constrained? Does the physical proximity of credit suppliers play a role? Are there neighborhoods which tend to finance at higher costs, for example, due to a higher participation of payday borrowing? What has been the effects of national policies providing low cost to the population on their indebtedness and financial distress? In this project we are building fine-grain maps of the indebtedness of…
Efectos de la ley de Alquileres CABA
Efectos de la Ley de Alquileres Noviembre Aquí encontrarán una primera versión1 de un documento de trabajo que analiza los efectos de la ley de alquileres de CABA, promulgada en 2017. Bienvenidos los comentarios. La ley de CABA de 2017 introdujo algunas modificaciones a la operación del mercado, incluyendo la transferencia de la obligación del pago de la comisión del inquilino al propietario, entre otras medidas que mencioné en un post anterior. En el documento, me propuse medir los efectos de la ley de alquileres de la Ciudad de Buenos Aires de 2017 en los valores de alquiler. La principal motivación detrás de la metodología propuesta es evitar confundir la medición del efecto con otros factores que pudieran afectar los valores de manera contemporánea, como podría ser el efecto de la inflación, las modificaciones en el valor que ocurren a raíz de las tendencias de transformación urbana, y evitar sesgos de selección en la comparación con otras ciudades. Todo el análisis está basado en datos abiertos, de ofertas de alquiler publicadas en Properati. El código completo también estará próximamente disponible. El análisis es posible gracias a algunas particularidades de la Ciudad de Buenos Aires, más precisamente el hecho de estar…
Efectos ley de alquileres CABA #1
[latexpage]In englishComparto aquí un análisis preliminar de los efectos de la ley de alquileres, sancionada en 2017, sobre los valores de alquiler. Esta ley, declarada inconstitucional en Mayo de este año, libró a los inquilinos del pago de la comisión inmobiliaria, al tiempo que también impuso una comisión máxima a ser pagada a las inmobiliarias.
Identification with DAGs: Introduction with simple simulations
En español
In this post I want to share with you some introductory ideas on how Directed Acyclical Graphs (DAGs) are used for causal identification. I am also sharing a few (Stata based) numerical simulations (here), that can be illustrative of their use in a regression application.
The DAG approach has been around for at least a decade now, and is described in extent in the excellent book by Pearl and Mackenzie (2018)’s “The Book of Why”. There’s so much going on in the book that I will be writing more about it in a future post.
Notes on Matching in Entrepreneurial Finance Networks
español We are glad our paper (with Virginia Sarria Allende and Gabriela Robiolo) went out in Venture Capital : A International Entrepreneurial Finance Journal. A working paper version is available here. Here are some brief comments on the ideas, the econometrics, and data approach: In the paper we study the “matching” between investors and startups in the entrepreneurial finance market. Broadly speaking, we are concerned with the question of who will invest in who, and in the role played by (social, professional) networks in the explanation. Specifically, we show evidence on a simple idea: due to information related frictions in the entrepreneurial finance market, being closer in the network of connections actually matters for matching. Being closer increases not only the attractiveness of a prospective match, but also makes observable attributes more attractive. But “being closer” has a particular interpretation here. Our measured networked connections, are not the typical social (or follower) style of connections. We recognize a link if there is information that you have worked, invested, mentored, etc. a common startup or organization in the past. So we could say that these are really costly (or “signally meaningful”, in Spence’s sense) connections. For founders (or prospective investors alike)…