Why Daily Returns Still Matter for Volatility Forecasting

For quantitative traders, selecting the right dataset to predict future market movements is arguably the most critical decision they make. Historically, the closing or opening price of a given stock or commodity was a good starting point to analyze the trading pattern of that specific security.

But today, as trades are settled in microseconds by advanced IT systems and a lot of trading volume is created by high-frequency “bots”, the data from such market activity is often preferred.

A new study is suggesting that high-frequency market data has not made daily returns obsolete. Using a new realized-volatility model, it reveals that combining both signals can materially improve crude-oil volatility forecasts, risk limits, and hedging decisions.

The study was done by three researchers at the Indian Institute of Technology Guwahati, and published in Finance Research Open¹, under the title “Do returns still matter? A complete asymmetric volatility model with realized measures in financial markets”.

A Brief Overview Of Risk Models

From the 1980s, researchers in economics and traders started to incorporate a new metric in the risk model they used to predict market behaviors: time-varying volatility of asset returns. This allowed the model to better reflect real markets, where asset risks fluctuate over time, with periods of high and low turbulence clustering together rather than remaining constant, as was the case in previous models.

Later on, high-frequency data was preferred for such models, as it was considered to be a superior data set for this application:

“High-frequency data allow finer granularity, facilitating the separation of jump and continuous components and providing a more detailed account of volatility dynamics.”

This led to the creation of the GARCH model (generalized autoregressive conditional heteroscedasticity), which was later refined in further advanced models incorporating extra elements like the different effects of positive and negative shocks and other datapoints.

Over time, GARCH-type models have expanded considerably, with applications spanning multiple asset classes, including equity, commodity, cryptocurrency, and derivative markets.

One of these updated models is the GJR-GARCH, a type of model that takes into account financial volatility by giving more weight to bad news (negative returns) than good news, capturing the stock market “leverage effect.”

Mixing High-Frequency Data And GARCH

This study proposes to mix high-frequency, minute-by-minute data, especially “realized variance,” with the GJR-GARCH model, creating the RGJR-GARCH.

Realized variance is a metric that measures the actual volatility of a financial asset over a specific time window by summing up high-frequency intraday price returns.

This differs from traditional daily variance, which sees zero change if the end of price is the same as the start, even if volatility was high intra-day.

By doing so, this new model can both integrate detailed high-frequency data and the accuracy of GJR-GARCH models.

Testing The Model With Oil Trading

Testing For Volatile Markets

In order to validate their model, the researchers tested it using USO (USO ), the United States Oil Fund, an ETF that tracks the price movements of West Texas Intermediate (WTI) light sweet crude oil since 2006.

This was chosen as crude oil markets are characterized by pronounced intraweek volatility driven by macroeconomic announcements, inventory shocks, and geopolitical developments. The recent Russia–Ukraine and USA-Iran wars have provided an additional case of surprise volatility impacting trading returns and financial models’ efficiency.

To compare to a more “normal” market, they also tested their model with SPY, the most actively traded ETF tracking the S&P500 index.

In both cases, they collected data from January 1, 2010, to April 30, 2020.

The researchers also noted that other potentially interesting markets, like gold and forex markets, do not typically provide reliable tick-by-tick data over long horizons, and cryptocurrency price data is generally available at most at a 1-minute interval, making unavailable the high-frequency data the model needs to perform correctly.

Models Comparison

The researchers used the data to test with different models 35 one-day-ahead forecasts, which were subsequently aggregated into weekly horizons.

They found that the RGJR-GRCH model had the highest forecast accuracy for all numbers of weeks ahead tested, up to 7 weeks later.

Source: Finance Research Open

More importantly, this difference in performance meant that RGARCH started to underperform by week 3 prediction and suffer negative performance by week ahead #4, RGJR-GRCH kept making accurate predictions for periods as far as 6 weeks ahead, and only very moderate errors for week #7.

By looking at what drove this superior performance, the researchers demonstrate it was indeed the use of high-frequency trading data.

“The superior performance of the RGJR-GARCH model relative to the GARCH and GJR-GARCH models is attributed to the effective use of high-frequency data in modeling volatility dynamics.”

The effect was especially pronounced for oil-linked assets such as USO, where volatility exhibits pronounced weekly regularities. This is important for practical applications, as accurate weekly volatility forecasting can inform dynamic hedging and contract pricing for energy-sector participants, such as commodity traders and producers.

This also meant that for forecasting volatility, daily returns are important as well, not just the high-frequency data. Both datasets are intermingled and should be processed as one.

Investing In High-Frequency Trading

CME Group Inc.

As better forecasting is created by an improved financial model, the value of accurate, long-dated, and high-frequency datasets increases accordingly. This is especially true for high-volatility, geopolitically sensitive securities and assets like crude oil. So this makes the platform able to provide these high-frequency data and actionable trading securities likely to benefit from such types of academic research.

CME’s NYMEX marketplace is central to WTI crude oil price discovery, futures trading, and hedging. The company is also active in all sorts of trading covering all commodities (agricultural, energy, metals), as well as carbon credits, treasuries, foreign exchanges, indexes, equities, cryptocurrencies, etc.

The company has quickly grown its revenue from around ~$3B in 2015 to ~7B expected in 2026.

It is also quickly internationalized, with non-US activity growing 10% CAGR and a sales presence in 12 countries, covering ~13,000 clients worldwide. Overall, this growth pattern can be expected to hold and the company to benefit from many financial innovations, from blockchain to carbon trading and U.S. mortgage futures.

Source: CME

Latest CME Group (CME) Stock News and Developments

Study Referenced

^{1. Prakash Raj, et al. Do returns still matter? A complete asymmetric volatility model with realized measures in financial markets. Finance Research Open. Volume 2, Issue 3, September 2026, 100139. https://doi.org/10.1016/j.finr.2026.100139}