Content uploaded by Roger G Ibbotson
Author content
All content in this area was uploaded by Roger G Ibbotson
Content may be subject to copyright.
A preview of the PDF is not available
Survivorship bias in performance studies
Abstract
Recent evidence suggests that past mutual fund performance predicts future performance. We analyze the relationship between volatility and returns in a sample that is truncated by survivorship and show that this relationship gives rise to the appearance of predictability. We present some numerical examples to show that this effect can be strong enough to account for the strength of the evidence favoring return predictability.
... All quantities with a c hat denote observed rather than true values. The estimator Ω contains survivorship bias [42]-a type of sampling bias stemming from the difference betweenF and Φ, becauseF is only measurable for observed species. For intuition, observed species (those that 'survived' sampling) will probably have higher abundances than the average of truly present species because rare species tend to be missed, so the mean observed abundance is higher than the true mean abundance of all species: ...
... We recommend obtaining a variety of estimators when assessing community diversity and detecting change, with Ω being the best abundance-based estimator and overall best at detecting richness differences, Ω 0 being overall best under low spatial heterogeneity conditions, and Chao2 being the least biased but requiring more replicated observations. Ω-type estimators contain survivorship bias [42] because observed species tend to have higher abundance and higher occupancy than unobserved species. If survivorship bias can be corrected (as revealed by the Taylor expansion version Ω T )-and this may very well be possible by studying how spatial and abundance statistics change with subsampling and downsampling experiments-the proposed approach would come very close to being the perfect estimator (see the idealized Ω C ). ...
Species richness is an essential biodiversity variable indicative of ecosystem states and rates of invasion, speciation and extinction both contemporarily and in fossil records. However, limited sampling effort and spatial aggregation of organisms mean that biodiversity surveys rarely observe every species in the survey area. Here we present a non-parametric, asymptotic and bias-minimized richness estimator, Ω by modelling how spatial abundance characteristics affect observation of species richness. Improved asymptotic estimators are critical when both absolute richness and difference detection are important. We conduct simulation tests and applied Ω to a tree census and a seaweed survey. Ω consistently outperforms other estimators in balancing bias, precision and difference detection accuracy. However, small difference detection is poor with any asymptotic estimator. An R-package, Richness , performs the proposed richness estimations along with other asymptotic estimators and bootstrapped precisions. Our results explain how natural and observer-induced variations affect species observation, how these factors can be used to correct observed richness using the estimator Ω on a variety of data, and why further improvements are critical for biodiversity assessments.
This article is part of the theme issue ‘Detecting and attributing the causes of biodiversity change: needs, gaps and solutions’.
... Survivorship bias can also explain why many people underappreciate the role of luck in determining their current and future incomes (see, e.g., Brown et al., 1992). Although initial luck-whether in resources or opportunity-almost always determines people's selection into a path where efforts have a higher probability of guaranteeing survivorship and future economic successes (e.g., Chetty et al., 2014), survivorship bias suggests that the survivors tend to learn only selected information about their own successes and not failures. ...
... Those study samples were incomplete because the set of surviving funds are typically truncated by survivorship bias (Brown et al., 1992). When examining an unbiased sample of funds, Grinblatt and Titman (1989) and Connor and Korajczyk (1991) found that smaller funds, maximum capital gain, and growth funds perform worse than expected, contrary to previous findings. ...
How do we persuade people to part with money they feel they have rightly earned? We conducted a dyadic experiment (N=1,986) where luck determined which of the players’ performance counted toward winning the game. Despite luck playing a large part, we found strong evidence of justified deservingness among the winners. The better they performed in the task, the less they redistributed to their nonwinning partner. However, in treatments where performance was transparent, winners significantly increased redistribution to nonwinners who performed similarly well. We find that transparency can effectively alter redistributive preferences even when people feel fully deserving of their income.
... It is challenging to identify alpha signals or build smart beta indexes using noisy datasets. • Survivorship bias of historical market data: Survivorship bias is caused by a tendency to focusing on existing stocks and funds without consideration of those that are delisted [9]. It could lead to an overestimation of stocks and funds, which will mislead the agent. ...
... For educational purposes, we provide Jupyter notebooks as tutorials 9 to help newcomers get familiar with the whole pipeline. • Multi-agent RL for liquidation strategy analysis [7]: We reproduce the experiment in [7]. ...
Finance is a particularly difficult playground for deep reinforcement learning. However, establishing high-quality market environments and benchmarks for financial reinforcement learning is challenging due to three major factors, namely, low signal-to-noise ratio of financial data, survivorship bias of historical data, and model overfitting in the backtesting stage. In this paper, we present an openly accessible FinRL-Meta library that has been actively maintained by the AI4Finance community. First, following a DataOps paradigm, we will provide hundreds of market environments through an automatic pipeline that collects dynamic datasets from real-world markets and processes them into gym-style market environments. Second, we reproduce popular papers as stepping stones for users to design new trading strategies. We also deploy the library on cloud platforms so that users can visualize their own results and assess the relative performance via community-wise competitions. Third, FinRL-Meta provides tens of Jupyter/Python demos organized into a curriculum and a documentation website to serve the rapidly growing community. FinRL-Meta is available at: https://github.com/AI4Finance-Foundation/FinRL-Meta
... There are style factor strategies that can generate alpha in the cross-section of cryptos documented by Liu, Tsyvinski, and Wu (2019) and Liu, Liang, and Cui (2020), such as a market crypto factor, size, and momentum, making an allocation to crypto more attractive. On the other hand, given the relatively short histories of crypto returns, survivorship bias may not be accurately measured (see Brown et al. 1992). It might be that the true probability of a bliss regime is lower than the empirical estimates we find, or lower than in the hypothetical comparative statics exercises we examine. ...
... The resulting base sample has a total of 187 equity funds (representing over 98% of the total net assets [TNA] of Taiwan equity funds and balanced funds), 31 fund companies, 9 fund categories, and approximately 24,056 fund-month observations over the sample period. To minimize the survivorship bias proposed by Brown, Goetzmann, Ibbotson, and Ross (1992), all of the available funds that existed during the sampling period were included in the data set, and only the funds with less than six months of monthly data were eliminated. ...
... • People are more willing to talk about their successful, completed projects than about failed projects. This leads to survivorship bias, a term more commonly used in finance [BGIR92]: there is a lot more information and stories about successful projects than about failed projects. • Although many features are consistent across industries, some practices are field specific, making comparison challenging [BH08]. ...
Crowdfunding through platforms like Kickstarter is a common way to finance development projects for innovative products. These projects have a risk of failure that is not always communicated clearly to the people providing the money (backers). In this paper data from 35 crowdfunded projects is collected and analysed. It is used to estimate the probability that projects deliver the promised rewards to the backers , also known at the delivery rate or fulfillment rate. The results show that approximately 30 percent of projects are fulfilled or delivered on time or with less than six month delay. 40 percent of projects fail to delivery anything. This is in line with data about other IT projects. There are no significant relations between known project characteristics and the delivery status. The paper concludes with recommendations for crowdfunding platforms on how to improve transparency.
... This creates a survivorship bias, in which students who were retained for that long period are the ones represented in the study. Survivorship bias is notoriously known to cause a more optimistic view of the outcome and it is rightly so (Brown, Goetzmann, Ibbotson, & Ross, 1992;Carpenter & Lynch, 1999). In other words, our study gives an overview of the program completers rather than a holistic picture of all program entrants and their pathways. ...
Research has repeatedly demonstrated that students with effective learning strategies are more likely to have better academic achievement. Existing research has mostly focused on a single course or two, while longitudinal studies remain scarce. The present study examines the longitudinal sequence of students' strategies, their succession, consistency, temporal unfolding, and whether students tend to retain or adapt strategies between courses. We use a large dataset of online traces from 135 students who completed 10 successive courses (i.e., 1350 course enrollments) in a higher education program. The methods used in this study have shown the feasibility of using trace data recorded by learning management systems to unobtrusively trace and model the longitudinal learning strategies across a program. We identified three program-level strategy trajectories: a stable and intense trajectory related to deep learning where students used diverse strategies and scored the highest grades; a fluctuating interactive trajectory, where students focused on course requirements, scored average grades, and were relatively fluctuating; and a light trajectory related to surface learning where students invested the least effort, scored the lowest grades, and had a relatively stable pathway. Students who were intensely active were more likely to transfer the intense strategies and therefore, they were expected to require less support or guidance. Students focusing on course requirements were not as effective self-regulators as they seemed and possibly required early guidance and support from teachers. Students with consistent light strategies or low effort needed proactive guidance and support.
... The distribution of bootstrapped t-statistics in the tails is likely to exhibit better properties than the distribution of bootstrapped alpha estimates. For one it scales the alpha by its standard error (which tends to be larger for shorter-lived funds and for funds that take higher levels of risks), and also because it is related to the Treynor and Black (1973) appraisal ratio, which is prescribed by Brown et al. (1992) for mitigating the survivorship bias problem (Kosowski et al. (2006). In panel A of Table 3, we have reported the cross-section of ranked funds based on the four-factor alpha values. ...
ResearchGate has not been able to resolve any references for this publication.