{"paper_id":"2506.07928","title":"Predicting Realized Variance Out of Sample: Can Anything Beat The Benchmark?","authors":["Austin Pollok"],"paper_date":"2025-06-09","paper_type":"dataset-benchmark","paper_url":"https://arxiv.org/pdf/2506.07928v1","page_url":"https://thequant.space/flowcharts/predicting-realized-variance-out-of-sample-can-anything-bea/","scores":{"empirical_rigor":6,"math_complexity":7.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes forecasting firm‑level daily realized variance using high‑dimensional machine‑learning models and low‑dimensional factor models, and evaluating the forecasts via equity option straddle portfolios.","quote":"We consider this phenomenon at the daily frequency using high-dimensional machine learning models, as well as low-dimensional factor models.","summary":"The discrepancy between realized volatility and the market’s view of volatility has been known to predict individual equity options at the monthly horizon. It is not clear how this predictability depends on a forecast’s ability to predict firm-level volatility. We consider this phenomenon at the…","strategy":"PCA‑HAR nested model (forecast of realized volatility)"},"assumptions":{"status":"available","items":[{"assumption":"Forecasts are generated using data available before 3:55 pm ET, avoiding look‑ahead bias.","type":"data","stated":"explicit","quote":"Because we estimate our model between the close of period t −1 and before 3 :55pm ET in period t, we cannot include the observation (xt−1, yi,t) in our estimation set."},{"assumption":"Firms that enter or exit the S&P 500 during the sample are dropped to keep a balanced panel.","type":"data","stated":"explicit","quote":"Because we are working with an unbalanced panel at each point in the sample (11), we choose to apply a filter which drops any firms that enter or exit the estimation set."},{"assumption":"Realized variance is measured using 5‑minute returns from NYSE TAQ data, assuming sufficient liquidity and price quality.","type":"data","stated":"implied","quote":"which is computed as the sum of all squared 5‑minute returns within a single day."},{"assumption":"Option implied volatilities are taken from OptionMetrics and are assumed to be accurate representations of market expectations.","type":"market","stated":"implied","quote":"and Greeks for all U.S.-listed index and equity options."},{"assumption":"Trading can be executed at the daily close without significant market impact.","type":"execution","stated":"implied","quote":"We cannot include the observation (xt−1, yi,t) in our estimation set. We do not have access to yi,t, which we wouldn’t have access to until after the close of period t."},{"assumption":"The sampling frequency (5‑minute) is appropriate for the liquidity of S&P 500 constituents.","type":"other","stated":"implied","quote":"If updated price observations are only available intermittently throughout a trading day, then the effective sampling frequency required to generate the high frequency returns is lower than the frequency intended to compute the realized variance."}]},"datasets":{"status":"available","items":[{"name":"Realized variance data (5‑minute returns)","provider":"NYSE TAQ","kind":"market-prices","access":"commercial","period":"1993‑01 to 2019‑06","universe":"S&P 500 constituents","quote":"We use the NYSE Trades and Quotes (TAQ) database... For every 5‑minute interval between 9 :30am and 4pm ET, we use the last recorded transaction price","known_id":"taq"},{"name":"Option data","provider":"OptionMetrics","kind":"options","access":"commercial","period":"1996‑01 to 2019‑06","universe":"Optionable S&P 500 stocks","quote":"and Greeks for all U.S.-listed index and equity options. Our options sample starts in January 1996 and ends in June 2019.","known_id":"optionmetrics"},{"name":"Equity fundamentals and returns","provider":"CRSP","kind":"fundamentals","access":"commercial","period":"1993‑01 to 2019‑06","universe":"S&P 500 constituents","quote":"For the study of equity returns, we use the CRSP database, which is the source of stock prices, returns, trading volume, market capitalization...","known_id":"crsp"}],"known":[{"id":"taq","name":"NYSE TAQ","provider":"NYSE","kind":"market-prices","access":"commercial","quote":"In the theoretical world of continuous time price evolution, sending ∆→0, which corresponds to 2Such a dataset can be constructed from the NYSE TAQ (Trades and Quotes) database.","mentions":3,"used_in":3},{"id":"crsp","name":"CRSP","provider":"Center for Research in Security Prices","kind":"market-prices","access":"commercial","quote":"We compare the option’s bid or ask to the end-of-day bid or ask on the underlying stock, which is obtained from CRSP.","mentions":4,"used_in":2},{"id":"optionmetrics","name":"OptionMetrics IvyDB","provider":"OptionMetrics","kind":"options","access":"commercial","quote":"The construction of the call and put options data set from OptionMetrics allows us to create a subset of data containing S&P 500 at-the-money, delta-neutral straddles.","mentions":4,"used_in":2}],"sample":{"start":"1993-01","end":"2019-06","market":"S&P 500 stocks","asset_class":"options","frequency":"daily"},"data_resolution":"daily"},"transaction_costs":{"status":"available","modeled":"no","headline_net_of_costs":false,"detail":null,"quote":null,"source":"front_matter","rule":"R1","candidates":{"front_matter":"no","evidence_card":"unclear","evidence_card_quoted":true},"card_detail":null,"card_quote":"... understanding the effects of transaction costs and margin requirements as documented in Goyal and Saretto (2009).","meaning":"a trading or execution result is reported without costs"},"results":{"status":"available","evaluation_type":"trading-pnl","findings":[{"finding":"Traditional forecast error rankings show little to no consistent out‑of‑sample advantage over benchmark models.","kind":"predictive-accuracy","quote":"Viewed through traditional forecast rankings, it is hard to significantly and unambiguously beat the benchmark models."},{"finding":"Alternative volatility forecasts improve option‑straddle portfolio performance, increasing Sharpe ratio by about 20% and average daily return by roughly 0.6% (60 bps).","kind":"performance","quote":"volatility forecasting models. It beats the benchmark by increasing the Sharpe ratio by 20% as well as improving by average daily return by approximately 0.6% (or 60 basis points)."},{"finding":"An equally‑weighted average forecast ensemble yields similar improvements over the benchmark.","kind":"performance","quote":"The equally‑weighted average forecast ensemble has similar improvements over the benchmark."},{"finding":"LASSO‑based portfolios deliver lower daily average returns but also lower return volatility, resulting in Sharpe ratios comparable to the rolling‑SD benchmark.","kind":"performance","quote":"While the portfolios formed from the LASSO model have lower daily average returns, the model also produces a drop in portfolio return volatility."},{"finding":"Conditional straddle portfolios exhibit higher mean returns, lower standard deviation, lower skewness, and substantially lower kurtosis compared to unconditional portfolios.","kind":"statistical","quote":"the conditional straddle portfolios display an increase in mean, a decrease in standard deviation, a decrease in skewness, and an especially large decrease kurtosis."}],"reported_performance":{"strategy":"PCA‑HAR nested model (forecast of realized volatility)","sharpe":0.8044,"sharpe_benchmark":null,"annual_return_pct":null,"total_return_pct":null,"max_drawdown_pct":null},"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":53,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T17:36:05Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}