Portfolio Back-testing at One-Minute Level
How Does Portfolio Back-testing Work?
Performing portfolio back-testing is a complicated process that requires powerful software to run. Realistically, it is not something that humans can do manually. Instead, what is needed are carefully coded algorithms that mimic user intent depending on the variables at play.
The following are among the most important elements used to manage a successful financial portfolio back-testing process.
Criteria
A wide variety of criteria is needed to properly conduct such processes. These include:
- Which assets and instruments are to be invested in,
- The allocation size of the said investment and related sizing rules,
- How the evolution of the portfolio valuation and market condition could have changed the individual exposure to an asset,
- How the portfolio volatility and risk parameters could influence the exposure to a single asset,
- How the correlation between assets influences the investment rules; and
- How the liquidity of the asset influences the allocation limits.
From a data perspective, the portfolio back-testing software needs to manage all this information for each single asset strategy being back-tested. It also needs to have the ability to combine them with the aggregated information of the portfolio to give investment managers a clearer idea of how their strategy would have affected their portfolio performance and behaviour in that particular set of circumstances presented during the tested period.
In addition, the aggregated data at portfolio level must be analysed before rules that manage allocation and risk parameters can be applied to it.
Decision-Making
One element of portfolio back-testing is the decision-making process, which basically sees the software conduct automated trading on your behalf based on:
- your strategy,
- your risk controls,
- the historic market prices and data of the period you opt for, and
- any configurations applied.
This sees your financial strategy being turned into a series of mathematical formulae and rules that outlines the parameters’ values, allowing it to sell, buy, or hold on to securities depending on your decision-making process.
Data
For the software to conduct this process, it must be fed copious amounts of historical data, including that for:
- the specific period you are looking to back-test your portfolio in,
- the historical prices of your assets at that point in time, and
- the market data needed for the strategy throughout that period.
For the process to be accurate, investment managers require the highest, lowest, open, and close prices and volumes traded for each asset at various points throughout the day at a one-minute timeframe.
Certain off-the-shelf portfolio back-testing software gives you these at five points of the trading day, namely the start, the end, the top, the bottom, and the total volume mark. Nevertheless, this does not really offer a real-life representation of market prices, especially during periods of high market volatility.
For that reason, one-minute-level historical highest, lowest, open, and close, volume traded for each asset is recommended.
The ideal data source needs to also include the bid-and-ask prices of the market at one-minute timeframes. During difficult market conditions, the spread is typically wildly different to the standard or the average, so working with book prices allows for better back-testing of the strategy behaviour.
The best solution here is to work with the tick data of the traded prices and the market book, but this requires large amounts of data to be processed. Due to this, the computational power required is high, and so are the costs associated with it – both in terms of time and hardware resources.
Machine Learning
The whole idea of portfolio back-testing is to understand how you could tweak your investment strategy to, hopefully, achieve better results in the future for your portfolio and allocations.
Some types of artificially intelligent software come with machine-learning capabilities, which results in the algorithms being automatically modified based on your training set performance criterias.
When machine learning is applied at portfolio level, the software that runs the portfolio back-testing needs to apply the machine-learning test exactly when it is required. This does not just synchronise the whole process, but also helps manage all the strategies according to the information delivered from the machine-learning test.
Monte-Carlo Analysis
The Monte-Carlo analysis – sometimes also referred to as the Monte Carlo Simulation – is a process that helps predict the effects of random variables on a security (2).
During financial portfolio back-testing, this simulation can be used periodically to get a better idea of how unpredictable variables would have affected an investment manager’s portfolio and use the results to amend their investment strategy.
The right software can usually do this automatically and use its computational capabilities to customise its algorithms to an investment manager’s specific portfolio and strategy in the hopes of helping them achieve better results in the future.
As discussed in the section about machine learning, synchronisation and dependencies between processes must be managed accordingly.
