Generating Synthetic Histories for Backtesting Tactical Asset Allocation Strategies

Most ETFs used in tactical asset allocation strategies only possess recent inception dates. This has implications for backtesting these strategies under various macroeconomic regimes. In this article we outline some of the problems associated with extending backtests and present some solutions.

Recently we introduced systematic tactical asset allocation strategies and presented a backtest of the well-known 60/40 static allocation benchmark.

Before we present more examples of tactical asset allocation (TAA) strategy backtests in future articles we thought it would be instructive to highlight some of the issues that are often faced when dealing with TAA implementations. In this article we are going to take a look at one such issue, namely, the need to generate 'synthetic histories' in order to extend our backtests well into the past.

Travelling into the Past

Many TAA strategies only provide new allocation weights on a monthly basis. Due to the transaction costs associated with portfolio turnover they tend not to be rebalanced more frequently than this. Hence performance is often only tracked at monthly intervals (although our QSTrader software tracks performance on a daily basis for these sorts of strategies). Thus in a ten year period there will only be 120 data points available to examine a strategy's performance. Compared to higher frequency strategies this does not provide much information on which to base portfolio allocation decisions.

In addition to the infrequent sampling of performance is the fact that macroeconomic market regimes can often last a long time. Such regimes can last for a decade or more. Hence if a TAA strategy is to be evaluated for 'all seasons' it is necessary to backtest strategies over a long history to truly gauge their performance.

For institutional funds the above two concerns present less of a challenge than they do for the typical retail quant trader. Funds have access to expensive, long-history datasets for a wide variety of instruments. It is possible for these funds to make up their allocations to various asset classes using derivatives such as futures or swaps. Hence they are better positioned to historically evaluate allocation strategies.

Retail traders often do not possess this luxury. More modest account sizes mean the budget for datasets is much smaller. In addition the margin requirements and transaction costs required to trade derivative instruments are usually prohibitive for the retail trader looking to implement long-term TAA strategies.

Low-Cost ETFs to the Rescue?

This is where low-cost Exchange Traded Funds (ETFs) come in. They have effectively 'democratised' the availability of large scale asset allocation to the retail trader with much lower overall costs. The small total expense ratios, large assets under management (AUM) and extensive liquidity make systematic TAA strategies a viable proposition for retail quant traders.

The disadvatange of this approach is that ETFs for some of the desired asset classes only possess a recent inception date, typically post 2000, with many post 2007/2008. This limits the ability to backtest TAA strategies with a realistic impementation further into the past.

If retail traders desire more realistic backtests over longer timeframes it is necessary to augment the ETF pricing data with 'synthetic' simulated data. This comes with its own set of problems.

Synthetic History

In order to extend TAA backtests beyond the inception date of the implementation ETFs it is necessary to utilise 'synthetic' or 'proxy' data.

Since many ETFs used for TAA often track market indices one approach to the problem is to simply prepend the market index returns values to the ETF returns prior to the ETF inception dates and apply a proxy for expense ratios.

For instance the SPY ETF tracks the S&P500 market-cap weighted US stock market, but only has data to 1993. Hence the returns data for the S&P500 itself, with suitable costs subtracted, can be used prior to this date to extend a TAA strategy that has a US large-cap allocation.

While this method of extending asset class history is relatively straightforward to implement it does comes with its own issues. For instance, with daily OHLCV data it is possible to carry out rebalancing logic for a typical TAA strategy after the market close for submission of rebalancing orders at the next market open. This is the default behaviour in the new development version of QSTrader.

However, long-term index series available to retail traders usually only have daily closing values. This means that in order to extend an ETF returns series beyond their inception date it is necessary to utilise a proxy for the missing open values. One (unrealistic) method is to forward-fill the missing opening value from the previous day's closing value.

An alternative is to simulate trading by backtesting at the close, which also presents its own set of difficulties. One never knows the close price until it has actually occurred, at which point the market is shut and no further orders can be sent. A live trading implementation of this approach relies on carrying out rebalances near to the closing time of the exchange and using Market-On-Close orders to apply the necessary rebalances.

Despite all of these issues we at QuantStart generally believe it is better to have an indicative guide to performance of such long-term strategies across various macroeconomic regimes than not. Since we believe backtesting is essentially a filtering process for choosing strategy implementations possessing this knowledge will always be useful in aiding decision making.

Next Steps

Generating synthetic history accurately is only one of a myriad of issues that occur when attempting to backtest TAA strategies. In future articles we will consider the effect of corporation action handling as well as when and how to rebalance your allocations.