Improving Returns Through Asset Allocation

It is often said that asset allocation is the most important source of performance. The statement is only true to a point. The impact of asset allocation, while profound, is only one variable among many. If anything, potential price changes in stock selection far outpace those of asset classes. Asset allocation decisions simply impact a much larger portion of any portfolio. The benefits of asset allocation alone, when guided by sound principles and executed with the courage of conviction, are illustrated below, using a single-market model (Switzerland). As a provocative comparison, performance of globally investing Swiss Pension funds  is shown along side the model.

Data for the Credit Suisse Pension Fund Index, a proxy for Swiss Pension Funds is published with some delay. As a result CSPF Index data is as per latest official release and temporarily appended by Agathos estimates.

Three Choices, One Objective

Shown are four distinct domestic allocation models, investing in proxies for three asset classes:

  • Cash
  • Bonds
  • Equities

All other possible investments are excluded (commodities, real estate, foreign currencies, etc.). The model is operated with a time frame that seeks to generate value over periods of nine to 18 months forward. It assumes that to be invested leads to a value increase, so the sole objective of the allocation is to contain risk.

Seen through the eyes of a globally investing Swiss pension fund, such a simple concept may appear poorly diversified. Perhaps it is. But the model serves as illustration for potential solutions that apply just as much on a global level. The domestic options here are not intended as the  ‚holy grail’. Even so, the results are quite relevant to the domestic portion of Swiss pension funds, or to any other investor operating across quoted asset classes. Diversification is the means to an end and the name of that destination is value-added. Value may be added by reducing risk, and/or by increasing performance. Diversification should never be self-serving.

Each allocation model carries a designator that indicates maximum permitted equity exposure, in percent: D25, D50, D75 and D100.

The models are restricted to a ceiling in the equity weighting but there is no minimum exposure. Any changes are made on an all-or-nothing basis. The model’s investment choices are:

  • CHF 1 Month LIBOR
  • Synthetic 0% Swiss Federal Bond (10 Years)
  • Swiss Equity Market (SPI)

To put results of the asset allocation into proper context, return data are shown for bonds and equities.

Data is also shown for the Credit Suisse Pension Fund Index. This index is highly representative of Swiss pension funds. Obviously one may argue that is like comparing apples with oranges. Perhaps. But allocation is akin to choosing from a fruit basket and thus the exception would have to be that better choices could be made from a larger basket. If anything, the comparison would be unfair to the domestic model, not vice versa.

Swiss pension funds do not appear to reap any benefits from the vast array of allocation options at their disposal. If through the inclusion of the data in my comparison this becomes transparent, then the much the better.

All time series have a base value of 100 as of 31.12.99.

Risk Probability As Sole Decision Criterion

Across all models, the same underlying mechanic is applied. All decisions to buy, hold, or dispose of assets are driven exclusively by the perceived probability of risk, as either

  • acceptable

or

  • critical

Acceptable refers to probability of gains being greater than for losses, critical assumes a higher probability for losses than for gains. No attempt whatsoever is made to quantify the magnitude of risk, or the magnitude of potential gains. 

The default exposure of the models is to keep the maximum permitted exposure to equities, with the remainder held in bonds (fully invested position).

However, the asset mix is re-arranged whenever either bonds, or equities, show critical risk probability.

Then, proceeds from the disposal of equity exposure are invested into bonds, and from bonds into cash.

If bonds and equities are both found to have critical risk probability, then the allocation will keep 100% cash for as long as either, or both, risk probabilities return to acceptable status.

Graph 3 shows ‘Observed Risk’, a proprietary metric developed by Agathos and used as part of the underlying methodology. Experience shows that observed risk is more sensitive as ex-post risk metric and of superior utility compared to ‘volatility’.

Agathos D75 (a portfolio with up to 75% in equities) usually has risk very near that of government bonds (currently even less than this reference), risk for D100 is somewhat higher than for a bond portfolio but this higher risk is rewarded with substantially improved returns. All of that in the absence of any stock selection within the equity portion (which would likely improve performance further still).

Graph 4 illustrates the result of subtracting observed risk from nominal return.

Allocation Mirrors Investment Methodology

Using a mechanical decision making process to illustrate should not be construed as recommendation, or endorsement of mechanical decision making. On the contrary: I am a sworn sceptic of any algorithm-based investing. It is ultimately self-defeating and I view the current volume of algorithm-based trading & investing in financial markets as a ticking time bomb.

When illustrating the importance of applying judgement mechanical simulations are a necessary evil. Without them, no retrospective calculations would produce acceptable evidence. Only a strict mechanism adds the required amount of retro-active objectivity.

The risk assessment used in the models is not the result of some statistical fishing expedition that eventually settles for the most successful test.

Rather, the mechanism behind the illustration is based on methods of consideration that I have long used as global portfolio manager when taking investment decisions. These were equally valuable when assessing the investment outlook for all quoted investments (asset allocation across asset classes, national markets, and for stock selection around the globe).

If investment performance is expected to consistently be something other than the product of chance, then investment decisions, at any conceivable and permitted level must be based on a specific, and intelligently rather mechanically deployed methodology of sorts. There are numerous perfectly legitimate methodological concepts in existence, all of them have strengths and weaknesses. These materialise under specific circumstances and must be understood fully if they are to function reasonably well at all times.

Excess Returns From Risk Reduction

Table 2 shows some easy-to-understand data summarising monthly rates of return across the full history. In addition, the popular metric ‘volatility’ is shown, together with the magnitude of horizontal distortion of the ‘normal’ distribution which potentially renders all conclusions based on that concept useless.

While typically, the models lag in the sum of gains, this handicap is clearly compensated for by the containment of losses. Proof that the superior returns are mainly, or even exclusively the product of risk management.

Graph 5 shows return and risk relative to equities. All Agathos models exceed returns from the equity market, but with a fraction of the equity market’s risk. This is particularly true for Agathos D100 that clearly out-smarts the market index by virtue of timing alone.

Graph 6 makes the same comparison with Swiss Federal Bonds. Again, all models perform better, even the lowest risk Agathos D25. And only Agathos D100 has risk notably in excess of bonds, but is rewarded with more than twice the return.

Relative Immunity Against Market Declines

Table 3 shows the regression analysis of all Agathos allocation models against the Swiss Equity Market. Given the dynamic asset allocation , it is only natural  when R-squared is extremely low. Statistically speaking, that gives any predictions made with the regression formula a low degree of reliability. Low R-squared readings identify the methodology as source of performance, as it should be.

The allocation is not driven by statistics, it is driven by judgement (in this case a mechanical approximation thereof). And that ‘judgement’, while crude and far from perfect, shows itself superior to ‘chance’, or the return data would not be as shown. As such, it exposes the failure of current pension fund management, which all but ignores risk.

R-squared attempts to quantity how big a portion of investment return is explained by ‘the market’, or ‘benchmark’ ( y-variable), from 0 to 100%. The difference between R-squared and ‘100% ‘ is more or less the impact of investment management decisions. The lower R-squared, the greater that portion of returns attributable to managerial judgement.

Investment performance can and should be measured by statistical means quantitatively, and qualitatively. But it is not generated that way.

Data Sources

All raw data is sourced from the public domain, as published by the institutions listed below.

Most recent data for the Credit Suisse Pension Fund Index may be estimates, due to the delay in the publication of this index.

View Only Exhibits

Contact Author