Use this URL to cite or link to this record in EThOS:
Title: Dynamic sampling methods for long term wealth management
Author: Go, H. G.
Awarding Body: University of Cambridge
Current Institution: University of Cambridge
Date of Award: 2007
Availability of Full Text:
Full text unavailable from EThOS.
Please contact the current institution’s library for further details.
In finance dynamic stochastic programming traditionally has been applied to institutional pension fund problems and more recently has become usable for more difficult individual wealth management problems. We develop several models to handle specific wealth management issues. We develop a US investment model with an exact tax basis and a rudimentary tax qualified portfolio. We show the ability to solve this model with up to 48 stages in 10 asset classes using an exact tax basis by approximating the solution employing information constraints. For fewer stages we show tractability of solving the full model with at least binary branching at every node of the scenario tree. We also introduce a mortgage model to investigate the effects of interest-only mortgages and their maturity. Modelling maturity selection as a binary decision variable, we find that the interest-only components of a mortgage are of interest when a borrower has a low income initially but expects it to grow. We do not consider the case of investors taking such mortgages to increase their leverage. It is noted that solutions may not be representative of all possibilities because the models reach an upper limit in terms of solvable problem sizes with currently available computing power. Expected value of perfect information (EVPI) calculation capabilities have been added to a modern solver. Given that aggregation is used to decrease solution times of such models we implement for the first time a disaggregator to allow calculation of EVPI subproblems without rereading the problem considered from disk. Aggregation is also found to increase solver speed applied to EVPI subproblems, especially after we reorder nodes. Sequential EVPI importance sampling is shown to be effective for the models introduced here and results improve drastically when mean matching of sampled scenarios is added. We successfully attempt to automate tuning for these algorithms by introducing percentile-based zero thresholds and adjusting these automatically when their current values are found to cause EVPI to fall.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available