also consider the likely effects of some variants of the new system. For this analysis,
we use new lifetime earnings simulations of graduates, set out in Dearden et al (2006),
and summarised in the Appendix, to estimate the repayment profiles of different
graduates for any given level of debt, as well as the likely per-student cost to the
taxpayer of fee deferral. These simulations provide us with estimates of the future
distribution of graduate earnings, on the assumption that the present observed
structure of earnings dynamics amongst graduates continues into the future, and that
in addition to any simulated change in earnings due to age or experience, earnings
increase by 2% in real terms each year due to underlying future productivity growth in
the economy.9 The simulations represent one of the only published empirical
characterisations of the distribution of lifetime earnings for graduates in the UK. The
only other work that we are aware of that attempted a similar exercise for the UK is
that of Barr et. al. (1995), who developed a dynamic cohort microsimulation model
based on one year of cross-sectional earnings data, to investigate the potential effects
of different ICL schemes. In contrast to their work, we use panel data to account for
dependence in earnings across time and movements into and out of the work force.
This allows us to characterise more realistically the distribution of earnings.
i. Lifetime earnings simulations
This section provides a very brief summary of our methods for simulating lifetime
earnings (a more detailed description is contained in the Appendix). The lifetime
earnings distributions are constructed using functions known as copulas, the name
being a reference to the fact that they ‘couple together’ k univariate distributions to
form a k-variate distribution. In our case, the construction of k-variate distributions is
simplified by assuming that wages follow a first-order Markov process, meaning that
only the most recent draw of wages affects the distribution of the next draw. This
allows us to model the dynamics of earnings over the lifecycle from just two wage
9 This is in accordance with HM Treasury, who use this as the rate of long-term earnings growth.
Please note that our previous simulations in working versions of this paper, and others, did not allow
for this economy wide real earnings growth, making our simulations of the effects of policy rather
different: in particular by simulating lower absolute levels of earnings going forward, we in general
projected a greater cost to the taxpayer in loan subsidies, and a lower cost to the individual; the length
of time we simulated that it would take graduates to repay their loans was also generally considerably
longer than with our current earnings simulations.