The Benefit of Alternative Data for Macro Strategies

Differentiated dataset implementation into macro investing strategies is not something new. Data sources like satellite and shipping data have been used by macro and commodity trading strategies for decades – long before the term “#AlternativeData” was coined and saw its acceleration in adoption 5 years ago.

However, despite this validation of certain data sources, we have still observed a lag in broader adoption within the macro trading and asset allocation community relative to corporate-focused equity and credit strategies. This has started to change in the last 2 years and we believe the adoption of new data sources for systematic macro, discretionary macro, and multi-asset groups will continue on a positive trajectory.

What has served as a catalyst? The commencement of the Q2 2020 pandemic, government lockdowns, and supply chain disruption created a significant challenge to company and economic forecasting models.

Nowcasting, the practice of using higher frequency data sources to anticipate economic measures with a significant lag (like #GDP ), became more prevalent within the Federal Reserve Banks, Central Banks, and by economic researchers more broadly. However, such models are not without their challenges and the recent dynamic of Central Banks falling ‘behind the curve’ on inflation forecasting shows that such models have not delivered with the accuracy required by policymakers. Do we have the scope to go further by using more alternative data in macro research?

We have noticed a meaningful change in demand for inflation, pricing, and supply chain insights within our advisory team since 2020. This has been prevalent with the largest multi-strategy and multi-asset groups where data adoption is advanced. We now believe we will see further adoption with more macro funds exploring new data sources. ?For several years, we have enjoyed using our data hackathons as a means of validating the real-world benefits of data sources and allowing practitioners to be assessed by experienced judges using dataset combinations and frameworks developed over multi-week periods. This reflects the reality of multiple dataset implementation alongside portfolio processes.?In the context of Eagle Alpha clients looking for more macro insights, in April 2021 we hosted a data hackathon with the specific challenge of creating a ‘real time’ macro framework that could improve on accuracy and timeliness versus the #NewYork Fed Nowcasting index. Thank you to our competing teams including Professor Sudip Gupta of Fordham University , Seth Leonard of System2 , and Professor Ben Lourie of University of California, Irvine - The Paul Merage School of Business .

Competing teams presented a framework that was assessed on dataset blending, transparency, economic intuition, explanation, and a breakdown of any signal composition. In short, trying to avoid blind machine learning and ‘black boxes’.?The winning team of this hackathon was led by Professor Sudip Gupta The team used weekly transaction data, daily sentiment from news articles, weekly app usage data, daily satellite data, monthly real estate, and daily social media insights. There were a lot of positives in that data was granular but also challenges were that the data was too granular, varying time frequencies, noisy, and approaches typical to standard econometric or blind machine learning models were not suitable. For the data science or statistical reader, the team addressed this with sparse group lasso and mixed data sampling to arrive at a GDP forecast combining traditional and alternative data inputs. These various alternative datasets were grouped alongside traditional economic time series for housing, manufacturing, surveys, consumption, income, and trade.

This was a particularly important time to host such a data hackathon, the New York Fed models having had limited error versus actual GDP prior to the pandemic, started to see significant deviations in 2020 and 2021. Professor Gupta’s team wished to solve these errors and to see if alternative data could reduce the error term in a GDP forecasting model. ?

So did alternative data work? Did the alternative data help outperform the dynamic factor models from the New York Fed which only uses traditional economic data series? In summary, as the image below shows, alternative data was significant in explaining the error term and you can see how an “#Altcast” model tracked better versus “#Nowcast”. This was all out of sample forecasting and an excellent contribution to our data hackathon series. Congratulations again to the winning team and Professor Gupta.

Image 1: “Altcast” versus “Nowcast” – what tracks Real GDP better??

These exercises show the value added by researchers incorporating new data sources into their economic models. This type of analysis is not in isolation - there is extensive research around the value of new data sources published by IMF, ECB, FRB Chicago, OECD, and other macro research groups. Academic papers have been published referencing #SatelliteData , #GoogleTrends , #TransactionData , #Geolocation (#Mobility ), and #Sentiment .

More recently, in October, we hosted a macro discussion panel with SESAMm , Macrobond Financial , Outra , and S&P Global Market Intelligence at our London Conference. One of the data companies (SESAMm) discussed that they were seeing a moderation in sentiment using natural language processing around news articles and web-sourced data in relation to the topic of “inflation”. Similar improvements were also observed when analyzing Central Bank statements and interviews with central bankers. On 10th November 2022, core CPI in the United States increased 0.3% for the month and 6.3% on an annual basis, both lower than expectations. Prices declined for medical care services, used vehicles, and apparel. These latter components are all areas where we have supported granular pricing data insights with alternative data for the last two years. Two-year yields fell 0.3% and US equity markets rallied 7.5%, the largest gain since March 2020.

This month we saw the latest job report based on BLS statistics and non-farm payrolls. The government data pointed to a better-than-expected jobs market and short-term US interest rate expectations increased. However, this data has a lag and is based on household surveys between the 6th and 12th of November.

For a more up-to-date read on job market conditions, one could consider, ? RIWI (Real Time Interactive Intelligence) -?a data vendor that specializes in the rapid capture of online opt-in-based intelligence. Their data is showing a 4% increase, in the weeks subsequent to 12th November, of respondents who reported a layoff of a peer in their close circle.

One could also monitor LinkUp data, which gathers job postings from company websites as to job demand by the 10,000 Global Employers with the most job openings in the US. As the image shows, job openings have been on a steady trajectory down since Q1 2022.

Image 2: Job opening in the US by Global Corporate Employers (source: LinkUp)

With differentiated and more timely insight and if academia and central banks are all seeing value – what reservations do investment strategies have??We believe there are 4 challenges to address. ?

Challenge 1: Data products are often presented by data vendors with an equity bias in their marketing, promotion, and use case applications. No economic researcher wants to hear about single-company revenue forecasts.

Our view: Data vendors need education on how their data can be aggregated, mapped, or presented in a way that makes the data suitable for macro research.

Challenge 2: Economic researchers do not believe that there are data categories besides satellite, shipping, and capital market flow data that apply to their research.

Our view: Consider data categories like in the image below (image 2) that can be used for research into specific macro variables. Areas like employment, housing, supply chain, inflation, and mobility are ripe for alternative data-driven research.

Challenge 3: There is not sufficient history in alternative data for validation and backtesting in economic research models.

Our view: We filtered 1,000 data products within the macro-relevant data categories - our platform profiling shows that 40% of these products have 10 years or more of history – this is a typical criterion for macro-driven backtesting.

Challenge 4: Alternative data is not of sufficient quality for macro models.

Our view: We believe this can be a very broad-based assumption and will vary significantly by data category and product. We have ingested, and written data quality reports and worked with most types of data sources. There is significant variation in data products. Hackathon teams, central bank research groups, and academics have been able to present research conclusions involving a wide variety of data products.

Image 3: Aligning Alternative Data categories with macro research topics and categories (source: Eagle Alpha)

Once researchers can get comfortable with these points. There are additional angles to alternative data that macro strategies should be capitalizing on. For example, corporate data owners may not provide data to the investment community that reveals brand, ticker, or company insights due to commercial sensitivities with underlying customers and stakeholders. However, if you asked that corporate data owner if they would provide that data at a City or Country level the response could be very different. Going further, when data is aggregated to a country or regional level the cost of the data can be less than if provided at a granular corporate or brand level. A macro researcher could be benefiting from data sources that are both differentiated and reasonable in price. For corporate data owners, there could be new demand triggers in the monetization journey and this is before one considers research towards government and social policy – more on this at our January conference.

In summary, we continue to see advanced users and experienced practitioners use alternative data to anticipate inflection points in relation to consumer sentiment, inflation, housing, and employment market conditions. However, adoption and usage are a fraction of where they should be. Gaining that edge or conviction on inflection points in the rate cycle, the US economy, or growth differentials is key to all strategies and performance. It is time for macro researchers, asset allocators, and multi-asset teams to step up their examination of alternative data sources through a macro lens.

We now connect data products on demand to our platform (generating coverage, data dictionary, and data samples) if you want a true sense of the suitability of the underlying dataset for macro research before entering a full trial data delivery process. Our advisory team has been dealing with macro-related research questions for many years and can help prioritize data sources versus your research topics and hypotheses. We look forward to discussing the merits of data for macro research with any of our readers.

This will be our last letter of 2022 - we hope you have found the series insightful – you can see the others in the series and letters below. Please do share with any colleagues who wish to subscribe.

Finally, we look forward to seeing many of you in New York on 19th January 2023 for our next in-person Conference. 500 attendees will benefit from an action-packed schedule, 70 data vendors attending for 1-to-1 meetings, a subset of new-to-market data products, academic perspectives, differentiated keynotes, and thought leadership across 4 key break out areas covering Buyside, Private Equity, Compliance, and Corporates.

You can see the full agenda and registration here.

If we do not speak to you prior to January, best wishes to you and your family. Enjoy the festivities and holidays.

Yours Sincerely,

Niall Hurley

CEO