{"paper_id":"2310.19992","title":"Robust Estimation of Realized Correlation: New Insight about Intraday Fluctuations in Market Betas","authors":["Peter Reinhard Hansen","Yiyao Luo"],"paper_date":"2023-10-30","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2310.19992v1","page_url":"https://thequant.space/flowcharts/robust-estimation-of-realized-correlation-new-insight-about/","scores":{"empirical_rigor":8,"math_complexity":7,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes a novel subsampled quadrant estimator (QS) that improves efficiency while preserving consistency and robustness for high‑frequency financial data.","quote":"We propose a novel subsampled quadrant estimator that improves efficiency while preserving consistency and robustness.","summary":"Time-varying volatility is an inherent feature of most economic time-series, which causes standard correlation estimators to be inconsistent. The quadrant correlation estimator is consistent but very inefficient. We propose a novel subsampled quadrant estimator that improves efficiency while…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Time‑varying volatility is present and the estimator must remain consistent under it.","type":"statistical","stated":"explicit","quote":"Second, QS is consistent under time-varying volatility.","support":"full"},{"assumption":"The price increments are atomless (no probability mass at zero) when volatility is bounded away from zero.","type":"model","stated":"explicit","quote":"The volatilities are bounded away from zero in both cases (constant with Σ non-degenerate in the first, bounded below by assumption in the second) and σ is independent of W given F0, so that the increments ∆δX and ∆δY are atomless for every δ > 0 and Pr(∆δX∆δY = 0) = 0.5 Zero increments in the data are consequently artifacts of market frictions, primarily price discreteness and stale prices, rather than a feature of","support":"full"},{"assumption":"Volatility process σ is independent of the Brownian motion W (no leverage) for the main asymptotic results.","type":"model","stated":"explicit","quote":"Under leverage, that is, under Assumption 2 in general and under Assumption 3’, the increments are not conditionally Gaussian and we do not know whether they are atomless.","support":"full"},{"assumption":"Zero returns are treated as missing observations and discarded; this does not affect the estimator asymptotically.","type":"data","stated":"explicit","quote":"An explicit convention for zero returns is therefore needed, and we adopt a convention under which they are discarded, ˆτ nz S = 1 N1 N X j=S sgn(∆δX j N ∆δY j N ), with N1 = N X j=S 1{∆δX j N ∆δY j N̸ =0}, and Qnz S = sin(π 2 ˆτ nz S ), such that only the non-zero terms are counted. N1 = N −S + 1 almost surely, ˆτnz S = ˆτS, and Theorems 1 and 3 apply unchanged to Qnz S .","support":"full"},{"assumption":"High‑frequency price processes are martingales with mean‑zero increments.","type":"model","stated":"explicit","quote":"For the high-frequency increments to which QS is applied, zero is the natural center: the price processes of Section 3 are martingales, so the increments have mean zero, and over intervals of a few minutes the median return is negligible relative to the standard deviati","support":"full"}]},"datasets":{"status":"available","items":[{"name":"Small Universe","provider":"TAQ via WRDS","kind":"market-prices","access":"commercial","period":"2015-01-01 to 2021-12-31","universe":"22 stocks (two largest per GICS sector) + SPY","quote":"Our sample runs from January 1, 2015, through December 31, 2021, and contains 1,763 trading days. We will analyze 22 stocks and SPY, an exchange-traded fund that tracks the S&P 500 index, in great detail.","support":"full","known_id":"taq"},{"name":"Large Universe","provider":"TAQ via WRDS","kind":"market-prices","access":"commercial","period":"2015-01-01 to 2021-12-31","universe":"98 S&P 100 stocks (excluding PYPL and RTX)","quote":"A larger panel of 98 stocks is used to identify stock characteristics that are associated with different patterns in intraday market betas. We refer to this data set as the “Large Universe.”","support":"partial","known_id":"taq"},{"name":"Synthetic Heston Model Simulations","provider":null,"kind":"synthetic","access":"unknown","period":null,"universe":null,"quote":"The initial values for volatility σ2 x,0 and σ2 y,0 are drawn from Gamma distributions, Γ(2κx¯σ2 x/s2 x, s2 x/2κx) and Γ(2κy¯σ2 y/s2 y, s2 y/2κy), and the price processes are initialized with X∗ 0 = log(100) and Y ∗ 0 = log(40). The simulated model is a discretized version with 23, 400 increments, which translates to one-second observations over a 6.5-hour period, the length of a typical trading day.","support":"full"}],"known":[{"id":"wrds","name":"WRDS","provider":"Wharton Research Data Services","kind":"other","access":"commercial","quote":"We use NYSE and NASDAQ transaction prices from the TAQ database that were accessed through the Wharton Research Data Services (WRDS) system.","mentions":2,"used_in":1},{"id":"taq","name":"NYSE TAQ","provider":"NYSE","kind":"market-prices","access":"commercial","quote":"We use NYSE and NASDAQ transaction prices from the TAQ database that were accessed through the Wharton Research Data Services (WRDS) system.","mentions":1,"used_in":1}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"tick"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"unclear","findings":[{"finding":"Intraday variation in market betas is primarily driven by intraday variation in correlations rather than relative volatility.","kind":"statistical","quote":"Our results show that intraday variation in betas is primarily driven by intraday variation in correlations. We have documented that relative volatility declines over the day by a similar amount for all stocks, so that the differences across stocks in the intraday movement of their market betas are mainly driven by differences in the intraday movement of their correlations with the market.","support":"full"},{"finding":"Correlations of all stocks with the market increase over the trading day, while relative volatility declines to 50‑75% of its first‑hour level.","kind":"statistical","quote":"We find that the correlations of all stocks in our sample with the market increase over the trading day, whereas their volatilities relative to market volatility decrease over the trading day. The relative volatility during the last hour of active trading is typically 50%–75% of relative volatility during the first hour of trading.","support":"full"},{"finding":"The subsampled quadrant estimator (QS) inherits the robustness of the classic quadrant estimator and is more efficient, using all high‑frequency observations.","kind":"theoretical","quote":"We propose a novel subsampled quadrant estimator that improves efficiency while preserving consistency and robustness. First, it inherits the robustness of the quadrant estimator while being more precise than Q (Section 3.5).","support":"full"},{"finding":"Simulation studies document that QS has lower bias and RMSE than the sample correlation (P) and Kendall’s tau (K) under time‑varying volatility and market microstructure noise.","kind":"statistical","quote":"The lower panels show the resulting bias of the correlation coefficients, P, K, and QS, as a function of the true correlation coefficient, ρ. With low collinearity, the sample correlation, P, has a large bias unless ρ is close to zero, while the bias in K is roughly half that of P. These estimators, P and K, are also biased in the example with high collinearity, but the bias is substantially smaller.","support":"full"},{"finding":"Empirical analysis on ~100 securities over seven years reveals that relative volatility declines similarly across stocks, making differences in beta movements mainly due to correlation dynamics.","kind":"statistical","quote":"The simulations document the finite-sample bias and root mean squared error of the estimators under time-varying volatility and market microstructure noise, and the empirical analysis, based on seven years of high-frequency data for about 100 securities, documents the intraday patterns in correlations and market betas. The empirical analysis documents how market betas vary within the trading day.","support":"full"}],"reported_performance":null,"out_of_sample":{"method":"none","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":18,"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":2,"extracted_at":"2026-11-14T20:00:45Z"},"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/"}