Indexology^® Blog

We previously introduced the S&P 500 Twitter Sentiment Indices and analyzed their performance characteristics. We noted how the indices seek to capture the Tweet-based sentiment premium, with their differentiated performance being driven by differences in index construction and sector weighting. In this blog, we will examine their exposure to standard risk factors and analyze their correlation to other factor indices.

Significant Exposure to the Growth and Quality Factors

We start with a returns-based regression using the five-factor Fama-French model, including momentum as a sixth factor. Exhibit 1 displays the factor loadings, while the complete output of the regression model is shown in Exhibit 2.

Here are some of the key observations.

• A significant negative loading on the value factor indicates that the S&P 500 Twitter Sentiment Indices have had a growth bias.
• Both indices have had a positive loading on quality factors (profitability and investment), well in excess of the benchmark.
• Both indices have exhibited weak and insignificant exposure to the momentum factor, possibly due to the mean reversion that follows highly abnormal social media sentiment.
• For the S&P 500 Twitter Sentiment Select Equal Weight (EW) Index, exposure to small size has been in line with other equal-weight indices.

Variation in Exposure to Momentum and Volatility

Next, we evaluate factor exposures using a holdings-based approach. Following S&P DJI factor definitions, we calculate standardized z-scores for S&P 500 constituents on a monthly basis. Then, we use the constituent weights to evaluate the factor exposure for each index (and the benchmark). As expected, the S&P 500 Twitter Sentiment Select EW Index had the strongest exposure to the sentiment factor (see Exhibit 3). For the remaining factors, the broader S&P 500 Twitter Sentiment Index had a relatively stronger exposure.

Over time, different factors have exhibited different levels of variation in their active exposures (see Exhibit 4). Quality and value exposures appear relatively stable, while momentum and low volatility exposures show a fair amount of variation relative to the benchmark. The active exposure to momentum, in particular, seems to alternate between strongly positive and near zero, supporting the insignificant factor loading seen from the regression analysis (see Exhibit 2).

Active Return Correlations

To round out the characterization of the S&P 500 Twitter Sentiment Indices, we evaluate the correlation between their active returns (relative to the S&P 500) and those of the S&P 500 Factor Indices.

We see that both indices have exhibited low or negative active return correlation to many well-known factors (see Exhibit 5), pointing to potential diversification benefits when targeting multi-factor exposure.

¹ Back-tested information reflects the application of the index methodology and selection of index constituents in hindsight. No hypothetical record can completely account for the impact of financial risk in actual trading. For example, there are numerous factors related to the equities, fixed income, or commodities markets in general which cannot be, and have not been accounted for in the preparation of the index information set forth, all of which can affect actual performance. The back-test calculations are based on the same methodology that was in effect on the index launch date. However, when creating back-tested history for periods of market anomalies or other periods that do not reflect the general current market environment, index methodology rules may be relaxed to capture a large enough universe of securities to simulate the target market the index is designed to measure or strategy the index is designed to capture. The back-test for the S&P Twitter Sentiment Indices is calculated for the period January 2018 to October 2021. S&P Dow Jones Indices designed the sentiment scoring model using data from approximately the same time range. The sentiment scoring model is a natural language processing tool based on linguistic classification of the degree to which a Tweet is likely to be positive or negative. Complete index methodology details are available at www.spdji.com. Past performance of the Index is not an indication of future results. Prospective application of the methodology used to construct the Index may not result in performance commensurate with the back-test returns shown.

S&P® and S&P 500® are registered trademarks of Standard & Poor’s Financial Services LLC. Twitter® is a registered trademark of Twitter, Inc. These marks have been licensed for use by S&P Dow Jones Indices for use with the S&P Twitter Sentiment Index Series. The Indices are meant for informational purposes only and are not recommendations to buy or sell any securities. Any investment entails a risk of loss. Please consult your financial advisor before investing.

Indian equities had a stellar 2021—the S&P BSE SENSEX rose over 23%, outpacing most major emerging markets, though slightly lagging the S&P 500^®, which gained 29%. Information Technology and Financials were the top contributors, adding 9% and 6%, respectively, to the performance of the Indian bellwether. Smaller Indian companies did even better than blue chips; the S&P BSE SmallCap returned over twice as much as the S&P BSE SENSEX, with a thumping 64% total return, while mid caps also provided a bright spot, rising 41% in 2021.

After exhibiting outsized moves in 2020, equities settled back into a more benign volatility regime; the annualized volatility of daily S&P BSE SENSEX returns dropped by more than half, to below 16% in 2021, and the S&P BSE MidCap and S&P BSE SmallCap volatility also subsided, dropping to 18% and 17%, respectively, from over 28% in 2020.

All S&P BSE sectors and industries were up for the year, with Power leading the way, soaring by 74%.

Despite the well-publicized setback for the Paytm IPO, the S&P BSE IPO surged 56% in 2021, the second-best-performing Indian equity strategy we regularly report on, lagging just 1% behind 2021’s winner, the S&P BSE Enhanced Value Index.

Majority state-owned firms have also had an outstanding year following a dismal 2019 and 2020, as the S&P BSE PSU and the S&P BSE CPSE climbed 48% and 43%, respectively.

Unlike equities, fixed income performance was much more muted. The S&P BSE India 10 Year Sovereign Bond Index edged up 2%, while the S&P India Sovereign Inflation-Linked Bond Index was essentially flat, as demand for inflation protection waned in parallel with a deceleration of inflation pressures in the Indian economy.

Consistent with global trends, indices incorporating environmental, social, and governance (ESG) factors are increasingly gaining in popularity with Indian investors. Their interest might have been raised further by the recent strong run of the S&P BSE 100 ESG Index, which outperformed its parent index for the third year running, returning 29% in 2021 against 27% for the S&P BSE 100.

One group that is having difficulty outperforming the S&P BSE 100 is that of active managers. According to our S&P Index Versus Active (SPIVA) India Mid-Year 2021 Scorecard, over 86% of active Indian equity large-cap managers were beaten by the S&P BSE 100 over the previous 12-month period, and the numbers don’t look much different on a three-  or five-year horizon either, with underperformance rates of 87% and 83%, respectively. Offering an indication of the breadth of opportunity available to “stock pickers,” annualized dispersion (a measure of the spread of stock returns) in the S&P BSE SENSEX declined to a monthly average of 20% this year, having averaged 21% over the 2010-2019 period. Among other consequences, declining dispersion lowers the magnitude of rewards for active managers seeking to beat their benchmarks by over- or underweighting individual names.

We can use volatility and its components dispersion and correlation to analyze stock selection conditions globally. Most active managers run less diversified, more volatile portfolios than their index counterparts. Active managers should prefer above-average dispersion because stock selection skill is worth more when dispersion is high. The role of correlation is more subtle. While counterintuitive, the benefit of diversification is less when correlations are high. Therefore, they should prefer above-average correlations because it will reduce the opportunity cost of a concentrated portfolio. We observe in the U.S. that while 12-month average dispersion declined slightly, we saw a dramatic decline in correlations in 2021 compared to 2020.

Two years ago, we conceptualized the cost of concentration, defined as the ratio of the average volatility of the component assets to the volatility of a portfolio. A higher cost of concentration implies a larger foregone diversification benefit, translating into a higher hurdle for active managers to overcome.

How much higher do returns have to be to justify the additional volatility active managers take on? By multiplying the cost of concentration by a rate of return consistent with the market’s historical performance (e.g., 10% for the S&P 500), we arrive at the required incremental return shown in Exhibit 2 for the S&P 500. Driven by the lower correlations seen in Exhibit 1, this measure almost doubled last year, indicating that active managers gave up a larger diversification benefit in 2021 than in 2020.

Finally, to understand how difficult it is to earn this incremental return, we divide the required incremental return by dispersion to translate the measure into dispersion units. We can interpret a higher number of dispersion units to mean more challenging conditions for active management.

The bars in Exhibit 3 indicate the average value of the required incremental return in dispersion units for several indices across regions. As dispersion and correlations were generally lower, we observe a relatively more difficult environment for active management across most markets.

The challenging environment for stock selection is not unique to the U.S. 2021 stands in stark contrast to 2020, when both dispersion and correlations rose globally. Despite those relatively favorable conditions for stock selection, most active managers still underperformed, proving that true stock selection skill is rare. When SPIVA^® results for 2021 become available, it would not be surprising if we saw dreary global active management performance once again.

With the global economy ramping up from the depths of the COVID-19 pandemic, labor shortages and supply chain bottlenecks are hampering manufacturers.

Fortunately, there is relief on the way. Digitalization is sweeping through manufacturing plants and transforming today’s sleepy mills into the smart factories of tomorrow. The catalysts include several of the technological forces driving the Fourth Industrial Revolution: exponential computing power, Big Data, artificial intelligence (AI), and machine learning (ML). The factory of tomorrow will be increasingly autonomous, self-optimizing, and sustainable.

The S&P Kensho Smart Factories Index seeks to track the companies that are enabling this manufacturing revolution. The index includes companies from across the full innovation ecosystem with an emphasis on the following four key technologies.

• Digital Manufacturing Solutions (52.6% of index weight¹): Covers the software enabling connected, integrated digitalization of manufacturing activities, including the equipment used for environment sensing and monitoring, advanced process control, and predictive maintenance. Led by cloud adoption, digital transformation with remote capabilities has gone from nice-to-have to need-to-have.
  • Per an Intel research report,² manufacturers experience up to 800 hours of unscheduled downtime annually (30% of it unplanned), while their skilled workers are aging, leaving 2 million jobs at risk of not being filled.
  • Cost savings and waste reduction are additional catalysts to adoption. A McKinsey study³ found that AI-enhanced predictive maintenance of industrial equipment can generate a 10% reduction in annual maintenance costs, up to a 20% downtime reduction, and a 25% reduction in inspection costs.
• The Industrial Internet of Things (IIoT) (22.9% of index weight): Represents the interconnection of big, ubiquitous data, sensors, instruments, and other devices networked together with computers’ industrial applications. It is key to facilitating manufacturers’ ability to connect, automate, track, and analyze industrial activities.

The IIoT provides real-time asset performance monitoring solutions, enabling operators and supervisors to visualize trends graphically from anywhere in the world on any connected device.

• Industrial Machine Vision (17.2% of index weight): Includes technology that combines sensors, cameras, computers, and ML/AI with visual data to identify product defects and model and predict equipment processes and product results.
• Digital Twins Technology (7.3% of index weight): Provides a virtual replica of any physical product, piece of equipment, or asset. For example, a digital twin could be a virtual version of a manufactured product, the entire production line, an entire factory, or network of plants.

Big results are expected from digital twins. Gartner⁴ predicts that organizations will save USD 1 trillion per year in maintenance costs by using digital twins.

In Exhibit 1, we can see the five-year forward growth outlook⁵ for each key technology and 2021 performance of companies in each segment.

The digitalization of factories promises to upend established manufacturing practices. The S&P Kensho Smart Factories Index offers niche market exposure that is well differentiated from the more traditional emphasis of legacy industrial-focused indices.

¹ As of Dec. 31, 2021. Index weight totals are based on companies’ product segment mappings.

² Intel Research. https://s21.q4cdn.com/600692695/files/doc_downloads/intelligent-factory-infographic.pdf

³ McKinsey Research. https://www.mckinsey.com/~/media/mckinsey/industries/semiconductors/our%20insights/smartening%20up%20with%20artificial%20intelligence/smartening-up-with-artificial-intelligence.ashx

⁴ Gartner Digital Twin Forecasts. https://www.ey.com/en_au/advanced-manufacturing/how-digital-twins-give-automotive-companies-a-real-world-advantage

⁵ Digital Manufacturing Solutions Growth. https://www.mordorintelligence.com/industry-reports/digital-transformation-market-in-manufacturing

IIoT Market Forecast. https://www.prnewswire.com/news-releases/industrial-iot-market-worth-106-1-billion-by-2026–exclusive-report-by-marketsandmarkets-301325443.html

Industrial Machine Vision CAGR. https://www.marketsandmarkets.com/PressReleases/industrial-machine-vision.asp

Digital Twins CAGR. https://www.mobinius.com/blogs/digital-twin-trends/

In a previous blog, we introduced the S&P 500^® Twitter Sentiment Index and the S&P 500 Twitter Sentiment Select Equal Weight (EW) Index, highlighting their objective and construction.

At the core of these indices is a model-based sentiment score that captures the direction and magnitude of the sentiment associated with a given company. S&P DJI derives these sentiment scores daily by analyzing the relevant corpus of Tweets containing $cashtags for each company.

Persistent Spread between Quintiles

Using equal-weighted quintiles created from S&P 500 constituents sorted by aggregate sentiment scores each month, we observe a gradual decline in performance from the top quintile to the bottom quintile (see Exhibit 1). We also see that the Q1-Q5 spread was fairly persistent and stable over time (see Exhibit 2). The spread was also generally robust to the choice of parameters used for calculating the sentiment score, demonstrating that Tweet-based sentiment partially satisfies the criteria for being considered a viable systematic factor.

Capturing the Sentiment Premium

The indices capture the sentiment premium in different ways. The broader S&P 500 Twitter Sentiment Index takes a more diversified approach by using a higher stock count (200), while the S&P 500 Twitter Sentiment Select EW Index only selects 50 “high conviction” names. Using two-year rolling windows, we see that both have historically outperformed the S&P 500 over time to varying degrees (see Exhibit 3).

The differences in index construction influence other quantitative metrics as well (see Exhibit 4). The diversified S&P 500 Twitter Sentiment Index has about half the tracking error compared with the S&P 500 Twitter Sentiment Select EW Index, resulting in about twice the information ratio. It has historically beaten the benchmark over 50% of the time in up- and down-market periods, leading to a better overall batting average. Both indices have had relatively high turnover due to the dynamic nature of social sentiment and generally higher volatility of high-sentiment stocks. Due to lower stock count and higher turnover, the S&P 500 Twitter Sentiment Select EW Index exhibited a significantly higher active share.

Sector Attribution

Exhibit 5 shows the results of a multi-period Brinson-Fachler attribution that helps explain the outperformance of the S&P 500 Twitter Sentiment Indices. The allocation effect determines whether over/underweighting a sector relative to the benchmark contributes positively/negatively to excess return. The selection effect measures the contribution of individual securities based on their relative weights. The interaction effect measures the combined impact of allocation and selection.

For the S&P 500 Twitter Sentiment Index, IT was the largest contributor to excess return, with a total effect of 4.99%. The average weight was 3.65% more than the benchmark, and this overweight contributed 1.91% in terms of allocation effect, while superior stock selection contributed 2.64%. Financials underperformed the benchmark over the period examined, and the index benefited from a smaller allocation to it. Though the selection of stocks had a negative effect in this sector, the interaction resulted in a positive contribution, adding 1.72% in total.

The S&P 500 Twitter Sentiment Select EW Index was underweight in seven sectors relative to the benchmark, and with the exception of IT, all contributed positively to excess return. Health Care had a higher weight compared with the benchmark, and the companies in this sector had the most volatile sentiment over the past few years due to frequent COVID-19-related announcements. Though the attribution was sometimes favorable, the compound effect over time resulted in a negative performance impact. Nevertheless, positive contributions from Financials, Industrials, and Real Estate led to a healthy excess return.

Looking at total sector attribution, the outperformance of each index interestingly came from different effects. The allocation effect contributed to the bulk of the S&P 500 Twitter Sentiment Index’s outperformance, while the selection effect was the main driver for the S&P 500 Twitter Sentiment Select EW Index. The interaction effect contributed positively in both cases.

¹ Back-tested information reflects the application of the index methodology and selection of index constituents in hindsight. No hypothetical record can completely account for the impact of financial risk in actual trading. For example, there are numerous factors related to the equities, fixed income, or commodities markets in general which cannot be, and have not been accounted for in the preparation of the index information set forth, all of which can affect actual performance. The back-test calculations are based on the same methodology that was in effect on the index launch date. However, when creating back-tested history for periods of market anomalies or other periods that do not reflect the general current market environment, index methodology rules may be relaxed to capture a large enough universe of securities to simulate the target market the index is designed to measure or strategy the index is designed to capture. The back-test for the S&P Twitter Sentiment Indices is calculated for the period January 2018 to October 2021. S&P Dow Jones Indices designed the sentiment scoring model using data from approximately the same time range. The sentiment scoring model is a natural language processing tool based on linguistic classification of the degree to which a Tweet is likely to be positive or negative. Complete index methodology details are available at www.spdji.com. Past performance of the Index is not an indication of future results. Prospective application of the methodology used to construct the Index may not result in performance commensurate with the back-test returns shown.

S&P® and S&P 500® are registered trademarks of Standard & Poor’s Financial Services LLC. Twitter® is a registered trademark of Twitter, Inc. These marks have been licensed for use by S&P Dow Jones Indices for use with the S&P Twitter Sentiment Index Series. The Indices are meant for informational purposes only and are not recommendations to buy or sell any securities. Any investment entails a risk of loss. Please consult your financial advisor before investing.