To illustrate the potential use of sentiment data, J.P Morgan constructed a set of simple long-short strategies: going long assets with the top three sentiment scores, and short assets with the bottom three sentiment scores.
Big Data and AI Strategies: White Paper Highlights
Trading Equity Indexes
- A weekly signal performs the best with an annualized return of 5.6% and a Sharpe Ratio of 0.44.
- The daily signal holds its own quite well with a Sharpe Ratio of 0.21.
- The portfolio is uncorrelated with traditional risk factors, such as momentum, value, carry, and volatility.
- It seems interesting to incorporate the strategy into a broader portfolio to reap the offered diversification benefits.
- Filtering based on RavenPack currency news, J.P. Morgan builds a contrarian long-short portfolio where the three currencies with the most negative/positive sentiment have a long/short position vis-à-vis the U.S. Dollar from a basket of 10 currency pairs (using carry-adjusted returns).
- Currencies reacts quickly to news events and hence are susceptible to reversal.
- Similar to what we saw above, the currency portfolio is uncorrelated with traditional risk factors and hence suitable in a broader portfolio.
J.P. Morgan succinctly demonstrates how RavenPack Analytics can be utilized to generate alpha streams that are uncorrelated to known risk factors.
Big Data and AI Strategies: Methodology
At a conference in New York City hosted by J.P. Morgan (JPM) earlier this month, several high-profile speakers from the quant and Big Data space gave their views on the topic. The takeaway from this event: ”alternative data delivers proven alpha”.
To get the ball rolling, JPM’s Global Quantitative & Derivatives Strategy team released a tour de force, titled "Big Data and AI Strategies", in which they put a host of alternative data providers, including RavenPack, under the microscope. RavenPack features prominently in their report and the JPM quant team provides several examples of how to unlock alpha by taking advantage of RavenPack Analytics (RPA) for global macro trading.
As a first step, JPM calculated daily sentiment scores for different asset classes of interest using the RavenPack dataset. The exact methodology is described in the grey box below.
To illustrate the potential use of sentiment data, JPM constructed a set of simple long-short strategies: going long assets with the top three sentiment scores, and short assets with the bottom three sentiment scores.
Trading Equity Indexes
Firstly, the JPM quant team constructed a daily long-short equity portfolio from nine country-specific equity indices using “economic sentiment’ as a confirming indicator to go long/short equity indexes. Sentiment scores were aggregated across various frequencies (e.g. 1 day, 1 week, 1 month, etc.) to create faster or slower moving signals.
The strategy delivers alpha irrespective of the chosen frequency, but the weekly signal performs the best with an annualized return of 5.6% and a Sharpe Ratio of 0.44. However, the daily signal also behaves well, returning a Sharpe Ratio of 0.21. Importantly, the portfolio is uncorrelated with traditional risk factors, such as momentum, value, carry, and volatility. Given the decent Sharpe Ratio expressed in the weekly portfolio together with the low correlation to known risk factors, it seems interesting to incorporate the strategy into a broader portfolio to reap the offered diversification benefits.
As an illustration of the various use-cases of RavenPack Analytics, JPM also considers a developed-market FX example, filtering based on RavenPack currency news. The result is a contrarian long-short portfolio where the three currencies with the most negative/positive sentiment have a long/short position vis-à-vis the U.S. Dollar from a basket of 10 currency pairs (using carry-adjusted returns).
The idea behind the strategy is that currencies reacts quickly to news events and hence are susceptible to reversal. Similar to what we saw above, the currency portfolio is uncorrelated with traditional risk factors and hence suitable in a broader portfolio.
Lastly, the quant team at JPM designed a bond portfolio, which borrows from the two strategies mentioned above in that firstly, it uses only events belonging to Economy and secondly, is contrarian. They report Sharpe Ratios similar to the two other strategies with a maximum of 0.45 over a 1 month aggregation window. Combined with the low correlation to classical risk factors, this portfolio could also be part of a broader portfolio.
Overall, the Quantitative and Derivatives Strategy team at J.P. Morgan demonstrates succinctly how RavenPack Analytics can be utilized to generate alpha streams that are uncorrelated to known risk factors.
Introduction and Overview of Big Data and AI Strategies
Big Data and Machine Learning ‘revolution’: Most records and observations nowadays are captured electronically by devices connected to the internet. This, in principle, allows investors to access a broad range of market relevant data in real time. For instance, online prices of millions of items can be used to assess inflation, the number of customers visiting a store and transacting can give real time sales estimates, and satellite imaging can assess agricultural yields or activity of oil rigs.
Historically, similar data were only available at low frequency (e.g. monthly CPI, weekly rig counts, USDA crop reports, retail sales reports and quarterly earnings, etc.). Given the amount of data that is available, a skilled quantitative investor can nowadays in theory have near real time macro or company specific data not available from traditional data sources.
In practice, useful data are not readily available and one needs to purchase, organize and analyze alternative datasets in order to extract tradeable signals. Analysis of large or unstructured datasets is often done with the use of Machine Learning. Succesful application of Machine Learning techniques requires some theoretical knowledge and a lot of practical experience in designing quantitative strategies.
Datasets and Methodologies: There are two main components of a Big Data investment approach: acquiring and understanding the data, and using appropriate technologies and methods to analyze those data. New datasets are often larger in volume, velocity and variability as compared to traditional datasets such as daily stock prices.
Alternative datasets include data generated by individuals (social media posts, product reviews, search trends, etc.), data generated by business processes (company exhaust data, commercial transaction, credit card data, etc.) and data generated by sensors (satellite image data, foot and car traffic, ship locations, etc.).
In most cases these datasets need a level of analysis before they can be used in a trading strategy. We aim to provide a roadmap to different types of data and assess their relevance for different asset classes as well as their relevance for different investment styles (e.g. macro, equity long-short, etc.). Methods to analyze big and alternative datasets include traditional statistics but also methods of Machine Learning. Machine Learning techniques include Supervised Learning (regressions, classifications),
Unsupervised Learning (factor analysis, clustering) as well as novel techniques of Deep and Reinforcement Learning that are often used to analyze unstructured data and show promise in identifying data patterns in structured data. In this report, we provide theory and practical examples of these Machine Learning methods and assess their efficacy.
Fear of Big Data and Artificial Intelligence: While many traditional investors don’t have a good understanding of the types of data available, and feel uneasy about adopting Machine Learning methods, we want to point out that these are not new concepts. On a limited basis, many investors already deal with...
Click here to continue reading the "Big Data and AI Strategies - Machine Learning and Alternative Data Approach to Investing" White Paper in PDFRequest White Paper