Kam-Fung Cheung
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Kam-Fung Cheung has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
Michael G.H. Bell
5 shared publicationsSupun Perera
4 shared publications- Anthony Man-Cho So
The Chinese University of Hong Kong
3 shared publications - Brian Sing Fan Chan
School of Mathematics and Statistics, The University of Sydney, Sydney, Australia
3 shared publications Jing-Jing Pan
3 shared publications- Nia Chen3 shared publications
Latest Publications
- Open Access
- 17 October 2026
- Published by Elsevier in SSRN Electronic Journal
Time-Occupation Information for Realized Volatility Forecasting
This paper studies whether the calendar-time occupation of the intraday price path contains useful information for realized variance forecasting. We construct occupation-quantile spread (OQS) predictors from symmetric quantile differences of the within-block occupation distribution, so that a price level receives weight according to the time for which the path remains there. At \(\alpha=0\), the daily measure is the sum of squared local high--low ranges and satisfies \(OQS_t(0)=4\log(2)PK_t\); for \(\alpha>0\), local extrema are replaced by interior occupation quantiles. Under a local Brownian approximation, the expected statistic is proportional to integrated variance conditional on local volatility, giving a variance-scale interpretation. We evaluate the predictors in HAR-X models for 30 large-cap U.S. stocks using daily forecasts from monthly updated rolling regressions over February 2015–December 2025. Adding a validation-selected occupation-quantile spread to a HAR model already containing Parkinson and Garman--Klass predictors lowers both QLIKE and HMSE for all 30 stocks, with mean loss ratios of 0.9728 and 0.8969. The results indicate that interior occupation structure adds predictive content beyond standard return- and range-based volatility predictors.- 17 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessTime-Occupation Information for Realized Volatility Forecasting
This paper studies whether the calendar-time occupation of the intraday price path contains useful information for realized variance forecasting. We construct occupation-quantile spread (OQS) predictors from symmetric quantile differences of the within-block occupation distribution, so that a price level receives weight according to the time for which the path remains there. At \(\alpha=0\), the daily measure is the sum of squared local high--low ranges and satisfies \(OQS_t(0)=4\log(2)PK_t\); for \(\alpha>0\), local extrema are replaced by interior occupation quantiles. Under a local Brownian approximation, the expected statistic is proportional to integrated variance conditional on local volatility, giving a variance-scale interpretation. We evaluate the predictors in HAR-X models for 30 large-cap U.S. stocks using daily forecasts from monthly updated rolling regressions over February 2015–December 2025. Adding a validation-selected occupation-quantile spread to a HAR model already containing Parkinson and Garman--Klass predictors lowers both QLIKE and HMSE for all 30 stocks, with mean loss ratios of 0.9728 and 0.8969. The results indicate that interior occupation structure adds predictive content beyond standard return- and range-based volatility predictors. - Open Access
- 10 October 2026
- Published by Elsevier in SSRN Electronic Journal
Time-Occupation Information for Realized Volatility Forecasting
This paper studies whether the calendar-time occupation of the intraday price path contains useful information for realized variance forecasting. We construct occupationquantile spread (OQS) predictors from symmetric quantile differences of the withinblock occupation distribution, so that a price level receives weight according to the time for which the path remains there. At α = 0, the daily measure is the sum of squared local high-low ranges and satisfies OQS t (0) = 4 log(2)P K t ; for α > 0, local extrema are replaced by interior occupation quantiles. Under a local Brownian approximation, the expected statistic is proportional to integrated variance conditional on local volatility, giving a variance-scale interpretation. We evaluate the predictors in HAR-X models for 30 large-cap U.S. stocks using monthly rolling forecasts from February 2015 through December 2025. Adding a validation-selected occupation-quantile spread to a HAR model already containing Parkinson and Garman-Klass predictors lowers both QLIKE and HMSE for all 30 stocks, with mean loss ratios of 0.9728 and 0.8969. The results indicate that interior occupation structure adds predictive content beyond standard return-and range-based volatility predictors.
- 10 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessTime-Occupation Information for Realized Volatility Forecasting
This paper studies whether the calendar-time occupation of the intraday price path contains useful information for realized variance forecasting. We construct occupationquantile spread (OQS) predictors from symmetric quantile differences of the withinblock occupation distribution, so that a price level receives weight according to the time for which the path remains there. At α = 0, the daily measure is the sum of squared local high-low ranges and satisfies OQS t (0) = 4 log(2)P K t ; for α > 0, local extrema are replaced by interior occupation quantiles. Under a local Brownian approximation, the expected statistic is proportional to integrated variance conditional on local volatility, giving a variance-scale interpretation. We evaluate the predictors in HAR-X models for 30 large-cap U.S. stocks using monthly rolling forecasts from February 2015 through December 2025. Adding a validation-selected occupation-quantile spread to a HAR model already containing Parkinson and Garman-Klass predictors lowers both QLIKE and HMSE for all 30 stocks, with mean loss ratios of 0.9728 and 0.8969. The results indicate that interior occupation structure adds predictive content beyond standard return-and range-based volatility predictors.
- Open Access
- 9 October 2026
- Published by Elsevier in SSRN Electronic Journal
Time-Occupation Information for Realized Volatility Forecasting
This paper studies whether the calendar-time occupation of the intraday price path contains useful information for realized variance forecasting. We construct occupation-spread predictors from symmetric quantile differences of the within-block occupation distribution, so that a price level receives weight according to the time for which the path remains there. At α = 0, the daily measure is the sum of squared local high–low ranges and satisfies DQ_t(0) = 4 log(2) PK_t; for α > 0, local extrema are replaced by interior occupation quantiles. Under a local Brownian approximation, the expected statistic is proportional to integrated variance conditional on local volatility, giving a variance-scale interpretation. We evaluate the predictors in HAR-X models for 30 large-cap U.S. stocks using monthly rolling forecasts from February 2015 through December 2025. Adding a validation-selected occupation spread to a HAR model already containing Parkinson and Garman–Klass predictors lowers both QLIKE and HMSE for all 30 stocks, with mean loss ratios of 0.9728 and 0.8969. The results indicate that interior occupation structure adds predictive content beyond standard return- and range-based volatility predictors.
- 9 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessTime-Occupation Information for Realized Volatility Forecasting
This paper studies whether the calendar-time occupation of the intraday price path contains useful information for realized variance forecasting. We construct occupation-spread predictors from symmetric quantile differences of the within-block occupation distribution, so that a price level receives weight according to the time for which the path remains there. At α = 0, the daily measure is the sum of squared local high–low ranges and satisfies DQ_t(0) = 4 log(2) PK_t; for α > 0, local extrema are replaced by interior occupation quantiles. Under a local Brownian approximation, the expected statistic is proportional to integrated variance conditional on local volatility, giving a variance-scale interpretation. We evaluate the predictors in HAR-X models for 30 large-cap U.S. stocks using monthly rolling forecasts from February 2015 through December 2025. Adding a validation-selected occupation spread to a HAR model already containing Parkinson and Garman–Klass predictors lowers both QLIKE and HMSE for all 30 stocks, with mean loss ratios of 0.9728 and 0.8969. The results indicate that interior occupation structure adds predictive content beyond standard return- and range-based volatility predictors.
- Open Access
- 12 August 2023
- Published by Elsevier in Transportation Research Part A: Policy and Practice
- Vol. 176
A system dynamics analysis of the trans-Pacific trade lane during the pandemic
- 12 August 2023
- Published by Elsevier in Transportation Research Part A: Policy and Practice
- Vol. 176
Open AccessA system dynamics analysis of the trans-Pacific trade lane during the pandemic
- 1 October 2021
- Published by Elsevier in European Journal of Operational Research
- Vol. 294 (1) , 353-364
Improving connectivity of compromised digital networks via algebraic connectivity maximisation
- 1 October 2021
- Published by Elsevier in European Journal of Operational Research
- Vol. 294 (1) , 353-364
Improving connectivity of compromised digital networks via algebraic connectivity maximisation
- Top Cited Papers
- 1 June 2021
- Published by Elsevier in European Journal of Operational Research
- Vol. 291 (2) , 471-481
Attacker–defender model against quantal response adversaries for cyber security in logistics management: An introductory study
- 1 June 2021
- Published by Elsevier in European Journal of Operational Research
- Vol. 291 (2) , 471-481
Top Cited PapersAttacker–defender model against quantal response adversaries for cyber security in logistics management: An introductory study
- 4 September 2020
- Published by Taylor & Francis in Transportmetrica A: Transport Science
- Vol. 17 (4) , 648-674
Connectivity evaluation of large road network by capacity-weighted eigenvector centrality analysis
The methods to evaluate the robustness of a network have been extensively studied. Such methods often require obtaining traffic equilibrium conditions or solving mathematical problems, and these methods can only be applied to a network of limited size. On the other hand, nowadays detail road network data can be downloaded freely, and such data may provide different insights on network robustness evaluation. This paper applies the capacity-weighted eigenvector centrality method to identify the strongly and weakly connected parts of large networks. The eigenvector centrality is one of the evaluation methods based on network topology with a small computational load. This method can be applied to directed networks and does not require their adjacency matrices to be symmetric. Several numerical examples showed that the capacity-weighted eigenvector centrality analysis can identify the strongly and weakly connected parts of the network, and it can be used to evaluate connectivity of network for robustness.
- 4 September 2020
- Published by Taylor & Francis in Transportmetrica A: Transport Science
- Vol. 17 (4) , 648-674
Connectivity evaluation of large road network by capacity-weighted eigenvector centrality analysis
The methods to evaluate the robustness of a network have been extensively studied. Such methods often require obtaining traffic equilibrium conditions or solving mathematical problems, and these methods can only be applied to a network of limited size. On the other hand, nowadays detail road network data can be downloaded freely, and such data may provide different insights on network robustness evaluation. This paper applies the capacity-weighted eigenvector centrality method to identify the strongly and weakly connected parts of large networks. The eigenvector centrality is one of the evaluation methods based on network topology with a small computational load. This method can be applied to directed networks and does not require their adjacency matrices to be symmetric. Several numerical examples showed that the capacity-weighted eigenvector centrality analysis can identify the strongly and weakly connected parts of the network, and it can be used to evaluate connectivity of network for robustness.
- 1 February 2020
- Published by Elsevier in Transportation Research Part B: Methodological
- Vol. 132, 15-28
An entropy maximizing approach to the ferry network design problem
- 1 February 2020
- Published by Elsevier in Transportation Research Part B: Methodological
- Vol. 132, 15-28
An entropy maximizing approach to the ferry network design problem
- 26 July 2019
- Published by Taylor & Francis in Maritime Policy & Management
- Vol. 46 (8) , 957-966
Connectivity analysis of the global shipping network by eigenvalue decomposition
Maritime shipping necessitates flexible and cost-effective port access worldwide through the global shipping network. This paper presents an efficient method to identify major port communities, and analyses the network connectivity of the global shipping network based on community structure. The global shipping network is represented by a signless Laplacian matrix which can be decomposed to generate its eigenvectors and corresponding eigenvalues. The largest gaps between the eigenvalues were then used to determine the optimal number of communities within the network. The eigenvalue decomposition method offers the advantage of detecting port communities without relying on a priori assumption about the number of communities and the size of each community. By applying this method to a dataset collected from seven world leading liner shipping companies, we found that the ports are clustered into three communities in the global container shipping network, which is consistent with the major container trade routes. The sparse linkages between port communities indicate where access is relatively poor.
- 26 July 2019
- Published by Taylor & Francis in Maritime Policy & Management
- Vol. 46 (8) , 957-966
Connectivity analysis of the global shipping network by eigenvalue decomposition
Maritime shipping necessitates flexible and cost-effective port access worldwide through the global shipping network. This paper presents an efficient method to identify major port communities, and analyses the network connectivity of the global shipping network based on community structure. The global shipping network is represented by a signless Laplacian matrix which can be decomposed to generate its eigenvectors and corresponding eigenvalues. The largest gaps between the eigenvalues were then used to determine the optimal number of communities within the network. The eigenvalue decomposition method offers the advantage of detecting port communities without relying on a priori assumption about the number of communities and the size of each community. By applying this method to a dataset collected from seven world leading liner shipping companies, we found that the ports are clustered into three communities in the global container shipping network, which is consistent with the major container trade routes. The sparse linkages between port communities indicate where access is relatively poor.