Weighted compositional data analysis for modeling and forecasting life-table death counts

Han Lin Shang; Steven Haberman

Weighted compositional data analysis for modeling and forecasting life-table death counts

Department of Actuarial Studies and Business Analytics

Research output: Chapter in Book/Report/Conference proceeding › Conference proceeding contribution › peer-review

Abstract

Age-specific life-table death counts observed over time are examples of densities. Non- negativity and summability are two constraints that prevent the direct implementation of standard linear statistical methods. Compositional data analysis presents a one-to-one mapping from constrained to unconstrained space to rectify the constraints. We introduce a weighted compositional data analysis for modeling and forecasting life-table death counts. Our extension assigns higher weights to more recent data and provides a modeling scheme that is easily adapted to allow for constraints. We illustrate our method using age-specific Swedish life-table death counts from 1751 to 2020 and show that the weighted compositional data analytic method improves short-term forecast accuracy compared to their unweighted counterparts.

Original language	English
Title of host publication	Living to 100 Research Symposium
Place of Publication	Orlando, Florida
Publisher	Society of Actuaries
Number of pages	20
Publication status	Published - 30 Nov 2023
Event	Living to 100 Symposia - Orlando, United States Duration: 15 Jan 2023 → 18 Jan 2023

Conference

Conference	Living to 100 Symposia
Country/Territory	United States
City	Orlando
Period	15/01/23 → 18/01/23

Keywords

age distribution of death counts
geometrically decaying weights
centered log- ratio transformation
weighted principal component analysis

Access to Document

https://www.soa.org/4ae322/globalassets/assets/files/resources/essays-monographs/2023-living-to-100-compendium/2023-living-to-100-comp-shang.pdf

Cite this

@inproceedings{3d6ba0e1fd8540e29aec4991dbc1d0a6,

title = "Weighted compositional data analysis for modeling and forecasting life-table death counts",

abstract = "Age-specific life-table death counts observed over time are examples of densities. Non- negativity and summability are two constraints that prevent the direct implementation of standard linear statistical methods. Compositional data analysis presents a one-to-one mapping from constrained to unconstrained space to rectify the constraints. We introduce a weighted compositional data analysis for modeling and forecasting life-table death counts. Our extension assigns higher weights to more recent data and provides a modeling scheme that is easily adapted to allow for constraints. We illustrate our method using age-specific Swedish life-table death counts from 1751 to 2020 and show that the weighted compositional data analytic method improves short-term forecast accuracy compared to their unweighted counterparts.",

keywords = "age distribution of death counts, geometrically decaying weights, centered log- ratio transformation, weighted principal component analysis",

author = "Shang, {Han Lin} and Steven Haberman",

year = "2023",

month = nov,

day = "30",

language = "English",

booktitle = "Living to 100 Research Symposium",

publisher = "Society of Actuaries",

address = "United States",

note = "Living to 100 Symposia ; Conference date: 15-01-2023 Through 18-01-2023",

}

Shang, HL & Haberman, S 2023, Weighted compositional data analysis for modeling and forecasting life-table death counts. in Living to 100 Research Symposium. Society of Actuaries, Orlando, Florida, Living to 100 Symposia, Orlando, United States, 15/01/23. <https://www.soa.org/4ae322/globalassets/assets/files/resources/essays-monographs/2023-living-to-100-compendium/2023-living-to-100-comp-shang.pdf>

TY - GEN

T1 - Weighted compositional data analysis for modeling and forecasting life-table death counts

AU - Shang, Han Lin

AU - Haberman, Steven

PY - 2023/11/30

Y1 - 2023/11/30

N2 - Age-specific life-table death counts observed over time are examples of densities. Non- negativity and summability are two constraints that prevent the direct implementation of standard linear statistical methods. Compositional data analysis presents a one-to-one mapping from constrained to unconstrained space to rectify the constraints. We introduce a weighted compositional data analysis for modeling and forecasting life-table death counts. Our extension assigns higher weights to more recent data and provides a modeling scheme that is easily adapted to allow for constraints. We illustrate our method using age-specific Swedish life-table death counts from 1751 to 2020 and show that the weighted compositional data analytic method improves short-term forecast accuracy compared to their unweighted counterparts.

AB - Age-specific life-table death counts observed over time are examples of densities. Non- negativity and summability are two constraints that prevent the direct implementation of standard linear statistical methods. Compositional data analysis presents a one-to-one mapping from constrained to unconstrained space to rectify the constraints. We introduce a weighted compositional data analysis for modeling and forecasting life-table death counts. Our extension assigns higher weights to more recent data and provides a modeling scheme that is easily adapted to allow for constraints. We illustrate our method using age-specific Swedish life-table death counts from 1751 to 2020 and show that the weighted compositional data analytic method improves short-term forecast accuracy compared to their unweighted counterparts.

KW - age distribution of death counts

KW - geometrically decaying weights

KW - centered log- ratio transformation

KW - weighted principal component analysis

UR - https://www.soa.org/resources/essays-monographs/2023-living-to-100-compendium/

M3 - Conference proceeding contribution

BT - Living to 100 Research Symposium

PB - Society of Actuaries

CY - Orlando, Florida

T2 - Living to 100 Symposia

Y2 - 15 January 2023 through 18 January 2023

ER -

Weighted compositional data analysis for modeling and forecasting life-table death counts

Abstract

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this