Periodic Reporting for period 2 - FIDAI (Financial Information, Data Abundance, and Investment)
Período documentado: 2023-01-01 hasta 2024-06-30
Recent years have witnessed a massive growth in the volume and variety of data ("Data Abundance"). In parallel, thanks to progress in computing power, new techniques (“Artificial Intelligence”) have emerged to harness the information in these data. How does this evolution affect the quality and nature of information produced in financial markets? Is there a risk that it induces investors to trade on noisier signals about economic fundamentals? How does it change the horizon at which information is produced? How does it affect the signal-to-noise ratio in securities prices and capital allocation? Addressing these questions is of utmost importance for society given the vital role played by financial markets for the allocation of capital to investment projects (e.g. those required to cope with long term risks, such as climate change risks) and their role for enabling investors to share risks and save.
The goal of FIDAI is to study these unanswered, yet fundamental, questions using a combination of analytical and empirical analyses. To this end the project is divided in 4 different work packages.
The first work package develops a theory of optimal data mining to disentangle the effect of Data Abundance (DA) from the effect of Artificial Intelligence (AI) on the quality of predictors used by asset managers. The second work package tests some predictions of the theory. In particular, it tests whether the availability of alternative data has a negative effect on the performance of asset managers lacking the skills to extract information from this data.
The third and fourth work packages study theoretically and empirically the effecst of data abundance and AI on agents’ allocation of their efforts for information production between two tasks: forecasting short-term cash-flows and forecasting long-term cash-flows and (ii) the consequence of these effects for the maturity of corporate investments. They show that the rise of alternative data is associated with a decline in the quality of long-term forecasts and that this decline can negatively affect the capital invested in long-term investment projects (whose cash-flows are realized more slowly).
The goal of FIDAI is to study these unanswered, yet fundamental, questions using a combination of analytical and empirical analyses. To this end the project is divided in 4 different work packages.
The first work package develops a theory of optimal data mining to disentangle the effect of Data Abundance (DA) from the effect of Artificial Intelligence (AI) on the quality of predictors used by asset managers. The second work package tests some predictions of the theory. In particular, it tests whether the availability of alternative data has a negative effect on the performance of asset managers lacking the skills to extract information from this data.
The third and fourth work packages study theoretically and empirically the effecst of data abundance and AI on agents’ allocation of their efforts for information production between two tasks: forecasting short-term cash-flows and forecasting long-term cash-flows and (ii) the consequence of these effects for the maturity of corporate investments. They show that the rise of alternative data is associated with a decline in the quality of long-term forecasts and that this decline can negatively affect the capital invested in long-term investment projects (whose cash-flows are realized more slowly).
The work performed so far has followed the time line proposed in the original grant vproposal. This has led to the following main results, presented in 4 different papers.
1/ The paper "Equilibrium Data Mining and Data Abundance", accepted for publication in the Journal of Finance and corresponding to Work Package 1 in the ERC proposal,provides a theory of data mining by quant funds. It predicts that as new datasets become available, the dispersion of quant funds' performance increases and the capital allocated to quant funds eventually decreases.
2/ The paper "Displaced by Big Data: Evidence from Active Fund Managers", corresponding to Work Package 2 in the ERC proposal, tests one implication of the theory developed in "Work Package 1. Namely, this theory implies that the introduction of new alternative data should make stock prices more informative about future cash-flows and therefore reduces the performance of asset managers who lack the skills to use this data (because their investment style relies more on qualitative analysis and judgement). Consistent with this prediction, we show that the introduction of new alternative data (from satellite images) covering retailers in the U.S. has a negative effect on the performance of funds holding these stocks and especially so when funds rely on traditional sources of expertise (e.g. geographical location or industry expertise).
3/ In the paper "Does alternative data improve financial forecasting? The horizon effect", accepted for publication in the Journal of Finance and corresponding to Work Package 3 in the ERC proposal, we show empirically that the rise of alternative data (e.g. social media data, geolocation data, or satellite imagery data) has triggered an improvement in the quality of security analysts' short-term earnings' forecasts and a decline in the quality of their long-term earnings' forecasts. The paper also proposes an economic mechanism (based on the optimal allocation of efforts between the tasks of forecasting short-term and long-term outcomes) that explains this observation.
4/ In the paper "The horizon of investors’ information and corporate investment", corresponding to Work Package 4 in the ERC proposal, we show empirically that the dynamics of security analysts' forecasts found in Work Package 3, can lead to a decline in the share of long-term corporate investment in the economy. Indeed, we show empirically, that firms with long-term investment projects reduce their investment when the quality of security analysts' long term earnings' forecasts declines. We also show that firms allocate relatively less capital to divisions investing in long-term projects when the quality of security analysts' long-term forecasts declines. These observations are consistent with a theory in which the information produced by the stock market about firms' cash-flows at different horizons influences managers' investment decisions.
1/ The paper "Equilibrium Data Mining and Data Abundance", accepted for publication in the Journal of Finance and corresponding to Work Package 1 in the ERC proposal,provides a theory of data mining by quant funds. It predicts that as new datasets become available, the dispersion of quant funds' performance increases and the capital allocated to quant funds eventually decreases.
2/ The paper "Displaced by Big Data: Evidence from Active Fund Managers", corresponding to Work Package 2 in the ERC proposal, tests one implication of the theory developed in "Work Package 1. Namely, this theory implies that the introduction of new alternative data should make stock prices more informative about future cash-flows and therefore reduces the performance of asset managers who lack the skills to use this data (because their investment style relies more on qualitative analysis and judgement). Consistent with this prediction, we show that the introduction of new alternative data (from satellite images) covering retailers in the U.S. has a negative effect on the performance of funds holding these stocks and especially so when funds rely on traditional sources of expertise (e.g. geographical location or industry expertise).
3/ In the paper "Does alternative data improve financial forecasting? The horizon effect", accepted for publication in the Journal of Finance and corresponding to Work Package 3 in the ERC proposal, we show empirically that the rise of alternative data (e.g. social media data, geolocation data, or satellite imagery data) has triggered an improvement in the quality of security analysts' short-term earnings' forecasts and a decline in the quality of their long-term earnings' forecasts. The paper also proposes an economic mechanism (based on the optimal allocation of efforts between the tasks of forecasting short-term and long-term outcomes) that explains this observation.
4/ In the paper "The horizon of investors’ information and corporate investment", corresponding to Work Package 4 in the ERC proposal, we show empirically that the dynamics of security analysts' forecasts found in Work Package 3, can lead to a decline in the share of long-term corporate investment in the economy. Indeed, we show empirically, that firms with long-term investment projects reduce their investment when the quality of security analysts' long term earnings' forecasts declines. We also show that firms allocate relatively less capital to divisions investing in long-term projects when the quality of security analysts' long-term forecasts declines. These observations are consistent with a theory in which the information produced by the stock market about firms' cash-flows at different horizons influences managers' investment decisions.
The project has led to several new results that go beyond the existing literature. First, it shows that the effect of alternative data on the informativeness of long horizon forecasts is negative and that this effect can be detrimental to investment in long term projects (see "Does alternative data improve financial forecasting? The horizon effect" and "The horizon of investors’ information and corporate investment"). Moreover, it has led to the development of a new theory of information production by investors that allows studying how the big data revolution affects (i) the allocation of capital between quantitative asset managers (using tools from data science to produce information) and (ii) discretionnary asset managers (using judgemental analysis to produce information) and (ii) the heterogeneity in the precision of signals used by asset managers for making their portfolio allocation decisions (see "Equilibrium data mining and data abundance"). One prediction of this theory is that the rise of alternative data should reduce the performance of asset managers lacking skills required to process large amount of data. The project has found supporting evidence for this prediction by considering the effect of the availability of new alternative data (based on satellite imagery) on the performance of active mutual funds' managers (see "Displaced by big data: Evidence from active asset managers"). We expect to be able to find additional support for this prediction by considering the introduction of other types of data. We also expect, based on the results obtained so far, to provide a better understanding of the horizon at which asset prices provide information about future cash flows.