In this Insight article we describe our asset allocation changes and performance from December 2007 to September 2009. The returns themselves are quite impressive. How did this come about, if we knew we did not know what was going to happen in the short term?

The information contained in the returns themselves and in the underlying asset allocations is about our investment process itself. So an analysis of our performance is an analysis of our process in action.

The process itself addresses a much trickier question: how do you manage money when the lesson properly learnt from both theory and practice in financial markets is that a manager cannot know what is going to happen in the short term, and their opinion is of limited value to their client? How does a battle-hardened agnostic place their bets and why?

The process is an outcomes-driven approach to goal-based investment portfolios, which we term ‘Defined Outcome Portfolios’. The approach is detailed here on the website.

The key characteristic of this approach is that some bets are avoided or hedged and others tightly and dynamically managed, within a specified ‘risk budget’. The risk budget is expressed in terms of a range of tolerable outcomes, in real terms (after inflation), at the times the money is needed.

With such an approach, the outcomes at shorter horizons are the wrong ones to make bets about. To the extent they are fully matched by risk free assets, matched to the duration of that time horizon, a key insight is that clients can afford to be indifferent to the volatility of these assets. If the volatility has no impact, their short-term performance is also of little import. The volatility is zero for cash but index linked gilts, though risk free in real terms at maturity (because of the government guarantee of inflation uplift), are nonetheless quite volatile in the period prior to maturity. But the volatility has no impact on outcomes, by definition.

Only the volatility of risky assets has a bearing on the probable plan outcomes and therefore the chances of the client achieving their minimum objective. These are the bets, as opposed to the hedges.

This transforms the relevance of the volatile short term return path of the ‘current’ portfolio. It makes a nonsense of comparisons of achieved short-period returns with other managers. Our clients are out of the performance race and in a journey, of their own planning.

Private clients are largely unaware of how to use performance information where planned goal outcomes are partially protected by hedging. They are not alone, however. The advent of ‘liability driven investment’ in the institutional market, which is the closest parallel, has also forced trustees to rethink the significance of reported performance, both for their own asset manager and for comparisons between managers.

Returns
In an earlier post, part way through the market fall, we took two clients as examples. Since all clients have a different combination of explicit time horizons, outcome profiles and risk tolerance, we have to use samples, not a universe. We have updated these two examples in this article.

They both have Defined Outcome portfolios funding retirement spending. One is still ‘in accumulation’, with about 5 years before drawdown starts. The second is already in drawdown, meeting current retirement spending.

The examples were chosen to have similar risk tolerance so that the asset allocation differences, as output by their own models, are explained by the time horizons (although changing profiles for spending outcomes at different stages of retirement also make a difference).

The returns are money-weighted returns. Because the cash flows, mainly new savings or draw, typically arise at month end, there is essentially no difference in the cash-flow adjusted returns whether money-weighted or time-weighted.

The period is from December 2007 through August 2009, as this roughly corresponds to the start of the bear market and the subsequent recovery.

The accumulation portfolio achieved a return over the whole period of -11.3%. The drawdown portfolio achieved a return of -4.1%. The paths of each, indexed to 100 at 31/12/2007, are shown below in Fig 1.

Fig 1: Index returns of two Defined Outcome portfolios

Asset allocation
The building blocks of a Defined Outcome portfolio are divided into two:

• Risk free assets (matched to the nature and duration of the time horizon, such as £x real spending in year y)
• Risky assets (in the form of a mix of equity markets across the world, as represented by broad indices with all but ‘systematic’ risk diversified away).

The two are combined to ensure that the projected outcomes of a dynamically-managed ‘plan’ (not just the current portfolio), counting down to fixed dates, are always consistent with the agreed tolerable outcomes.

The return paths of the building blocks (source: Lipper Hindsight) are shown in Fig 2 below, indexed to 100 at 28/09/2007

• The cash return is a one month deposit rate (the risk free asset for durations below about three years)
• The index linked gilt return is derived from the over 5 years FTSE index (as representative of the ‘medium-dated’ maturities where we typically combine risky and risk free)
• The equity returns are for index-tracking funds provided by Legal & General (as representative of the returns that could have been earned from tracking the risky asset building blocks, after the fund expenses attributable to institutional units).

Fig 2: Indexed total returns of the portfolio building blocks

Attribution
Because the asset allocations output from each model run for each client are implemented using index-tracking funds and individual index linked gilts, the portfolio return and its variability are almost entirely explained by the allocations.

The accumulation portfolio started with an equity allocation of 100% and the drawdown portfolio 65%. The risk free return has been relatively flat due to both low interest rates on cash and moderate volatility and net change in index linked gilt yields over the period.

Considering the difference in equity exposure at the outset, the gap at its widest point between the two portfolios is surprisingly small. The largest drop for the accumulation portfolio reached 28%. This also appears relatively small compared with the returns apparently experienced by most UK-based equity investors, including those investing purely in the UK, which was down 38% at that same point.

As Fig 2 hints at, a large part of the answer is geographical diversification. We tend to hold more in European, Japanese and US markets than most investors. This follows logically from the intention of diversification and is likely to be an output of any portfolio optimisation process that seeks to maximize expected return per unit of risk. However, most investors constrain exposures to foreign markets to suboptimal levels. Diversification benefits reflect the distribution of weights to each market not just the number of markets held.

Short-term correlations between different equity markets are highly unstable (and indeed increased dramatically during the bear market). But differences in achieved returns do emerge, implying differences in future returns. With fairly similar long-term real return trends apparent in the historical data, mean reversion operates both between and within markets. Larger international exposure allows this to be exploited more efficiently.

As it turned out, much of the benefit from geographical diversification in this particular period came from currency movements. Sterling fell as asset prices were at their weakest and strengthened as equities themselves rallied.

Currency movements are not normally well correlated with market movements so most of the time these two sources of risk will not combine to increase portfolio risk significantly. What happened here, with currency dampening portfolio volatility, cannot be predicted in advance or relied on.

Asset allocation changes
Fig 3 shows the asset allocation changes for a drawdown model starting with the December 2007 quarter and ending in the two months to August 2009. The actual model runs are monthly but we have conflated the changes to quarterly, to simplify the rebalancing narrative.

The resources at the outset in this model run were £1.8m which puts the changes shown below into context. The range of monthly change in allocations at the highest level, between risk free and risky, is from 4% to 8% of the total portfolio. This is clearly a gradualist approach.

Fig 3: Changes in net equity exposure

The narrative here, even if at the time it needed some explanation and illustration,  is entirely consistent with the portfolio process and so held no surprises for the client.

• Two quarters of additions to equity exposure relative to index linked gilts as the bear market began, as expected equity returns to each planning horizon were then higher than before and higher relative to index linked gilt yields for the same horizon – hence hedges can be marginally lifted
• A small net pull-back in the next quarter as equities rallied and index linked gilts fell, lifting their real yields
• As the US housing bubble started to go wrong and the credit crisis got worse, dragging risk free yields down with it, the next three quarters saw net additions to equities
• From March 2009 the model called for net sales of equities as prices rallied and risk free rates also moved higher.

The net change over the whole period is an addition to equities of £0.25m, equivalent to 14% of the starting portfolio value. Considering the net fall in equities over the whole period, and the relatively small net fall in risk free rates, a constant risk tolerance should be expected to lead to reductions in hedging and an increase in risky exposures. Investors (or their advisers) who instead altered their risk tolerance in response to what they thought was happening in the world would have shown the opposite response, selling risky assets as they declined in price, or might have been frozen into inactivity.

The net changes filter out the rebalancing arising between different equity markets which is also shown in the table. The gross change over the whole period is 42%. In practice, some of the smaller changes indicated by monthly model runs may not be implemented, as the likely improvements in risk-adjusted return are too small relative to the costs of transactions (even when these are minimized by using exchange traded funds on an execution-only stockbroker dealing platform). Many small changes should reverse themselves from month to month. There is an element of judgement in filtering these small changes.

The main changes in country weights are as follows:

• Reductions in Japan, which had started the period at 36% of the risky allocations – high exposure in Japan is a key test of a globally-diversified approach to long-term return return opportunities
• Roughly half of the net addition to equities was in the UK, which is a natural response to increasing real return ranges in all markets, as plan outcomes can then be met with less uncertainty in the home market.

The portfolio model run for August had the following new target allocations.

Fig 4: Closing model allocations

These allocations are entirely logical given the client’s planned profile of the drawdown over the life of the plan; the client’s tolerance of uncertainty in the draw; and our projected return probabilities for the portfolio building blocks, as generated by a new model run in August 2009.

• The 16% cash allocation is matching about three years of spending targets (as an estimated pre-tax equivalent portfolio draw).
• From years 4 to about 8 the outcome targets are fully matched by a schedule of maturing index linked gilts with the same duration as the spending
• The next 6 years are partially matched by index linked gilts but equities are also held against these spending targets
• Only durations later than 15 years are fully backed by equities.

The proportion of the total plan resources that is sensitive to volatility in the portfolio value is represented by the allocations to equities from years 8 to about 15. They are sensitive because the return-generating element of the model does not assume that a fall in equity prices will be offset fully by higher real equity returns within the time frame required for that horizon. There just is not enough time to be that sure. This portfolio proportion, for which short-term performance has an impact on probable outcomes, is about 28%!

The earlier horizons (35% of current resources) are indifferent to volatility because they are fully matched. The later years, beyond 15 years, (the balance of 37%) are far enough off for mean reversion to have offset much of the fall in current value, leaving outcome probabilities little moved by the fall.

Even though there is a middle segment of the plan that is sensitive to volatility, any plan that was fully funded to meet the drawdown targets with a high level of certainty should still meet its original planning targets. This is based on the simulations of the plan, which tested randomly-generated paths for equity markets and exchange rates as well as their interaction with a pre-defined rate of draw in real sterling amounts per annum.

In practice, the progress of the portfolio is likely to feed back to the client’s comfort level in spending. Should this arise because of really bad equity markets globally, implying widespread economic stress, it would not be surprising if the client chose to trim their spending objectives anyway, even if the model stress tests suggested it was not strictly necessary.

TAGS: drawdown, LDI, mean reversion, models, performance