Quant research team of the year: Deutsche Bank

Risk Awards 2018: team’s data science insights ‘transformative’ say clients

Left to right: Paul Ward, James Osiol, Caio Natividade, Spyros Mesomeris, Aris Tentes, Jacopo Capra, Jiada Lyu, Andy Moniz
Left to right: Paul Ward, James Osiol, Caio Natividade, Spyros Mesomeris, Aris Tentes, Jacopo Capra, Jiada Lyu, Andy Moniz
Photo: Juno Snowdon

Clients of Deutsche Bank’s quant research team say it has a gift for describing the finer points of complex ideas from machine learning to liquidity risk premia.

“Their research, more so than anything else I have seen, is actually usable from a quantitative standpoint,” says a senior quant portfolio manager at a US pension fund. “It is written in a way that not only describes the ideas but explains precisely what they are doing.”

A quant hedge fund portfolio manager describes chief data scientist Andrew Moniz’s work developing strategies based on unstructured text-based data as “transformative”.

In liquidity risk, the team’s research identified a non-diversifiable liquidity factor that is separate from recognised factors such as value, size or momentum. In work on risk model horizons, the team uncovered risks from mismatches between risk and alpha models that it says investors are overlooking.

Deutsche formed its quant investment solutions team, comprising 18 researchers, in 2009 focusing initially on equity quant strategy. Since then it has expanded into cross-asset quant strategy and data analysis. The team is headed by Spyros Mesomeris, the bank’s global head of quantitative strategy and quant investment solutions research, based in London.

In data science the team published research on using text analysis as a signal for long-only, low-turnover strategies. Its approach allows investors to stitch systematic and fundamental strategies together, providing them with insights not usually gained through one individual strategy.  

“Discretionary managers are increasingly interested in the big data space,” says Mesomeris. “For example, how they can utilise insights from credit card transaction data or geospatial data in their investment process in a way that can enrich their cashflow models or give them a differentiated view of companies.”

A purely systematic strategy may buy a company’s stock because it appears cheap, but such a strategy would be blind to the threats of litigation facing the company. 

Discretionary managers are increasingly interested in the big data space
Spyros Mesomeris, Deutsche Bank

Fundamental investors also gain by drawing information from sources not otherwise accessible in financial accounts. As well as for investigating corporate behaviour, the team has also looked at using natural language processing to gauge the sentiment of central bankers.

The sort of practical guidance Deutsche provides includes suggestions on extracting text from public webpages. Due to the variety of coding options available to website designers, Deutsche proposes a probabilistic method: using a text-to-tags ratio, for example, to identify the largest and most meaningful block of text.

Then there’s measuring the relevance of a specific text to a particular company. The team suggests using a named entity recognition algorithm. This classifies text into categories such as names or places and determines their importance according to frequency and their appearance in, for example, the header or first paragraph.

Mesomeris spent four years as a quantitative analyst at Citi before joining Deutsche Bank, while Moniz joined Deutsche from UBS, where he focused on long-short text-based strategies, having previously been a senior portfolio manager at APG Asset Management. Other senior members of the group include Khoi LeBinh, Asia-Pacific head of global quantitative strategy, Ronnie Shah, head of US quantitative strategy, and Caio Natividade and Paul Ward, directors in European quantitative research. Natividade focuses on futures and options trading using multi-frequency and multi-style models.

On risk models, Mesomeris says investors are overlooking how their choice of model horizon could have a greater impact on performance than their choice of alpha model.

Buy-siders often – wrongly – match the horizon of their risk models with that of their investments or portfolio rebalances, says Mesomeris. “It doesn’t matter so much if you are a long-horizon or short-horizon investor. The most important consideration is what type of portfolio you hold, whether you optimise or not.”

Spyros Mesomeris, Deutsche Bank
Photo: Juno Snowdon
People have become more aware that market-wide liquidity is actually a non-diversifiable source of risk
Spyros Mesomeris, Deutsche Bank

Longer-horizon risk models are better for optimised portfolios due to the models’ lower estimation error of correlations. But shorter-horizon models are typically better for long-only portfolios because longer-horizon models tend to over-forecast market shocks.

Deutsche says investors should, as far as possible, use the same factors in their risk model as in their alpha model. Representing the value factor by cashflow yield in the alpha model but using book-to-price in the risk model, for example, will result in factor misalignment. The consequence is a possible underestimation of risk due to the portfolio’s systemic risk being allocated to risk that is stock-specific.

Liquidity research

On liquidity, Deutsche’s research suggests a liquidity factor exists and is independent of other accepted factors.

“People have become more aware that market-wide liquidity is actually a non-diversifiable source of risk,” says Mesomeris. “There’s another risk factor out there – a macro-level risk factor – that’s non-diversifiable and, depending on your exposure to that factor, your returns will vary depending on changes in the market-wide liquidity environment.”

Deutsche’s research introduces several liquidity metrics and develops strategies to explore each of the dimensions of liquidity: level, volatility, change and beta. Level is a monthly average measure of the proportion of outstanding shares traded every day. Change represents the drift of the day’s traded volume away from the monthly average, while volatility is a measure of the volatility in daily volume. The fourth dimension, beta, is a measure of the stock’s exposure to overall market liquidity.

Many have argued illiquidity premia cannot exist because illiquid stocks cannot be traded, for example, or the premia are merely echoes of other premia.

The research finds level and beta are statistically significant, that is, there is a 90% probability of their explaining excess returns beyond traditional factors in 35% of the months investigated. The statistics are similar for risk premia such as size and quality, and 45% for market beta, value and momentum.

The only significant exposure to other factors for the metrics used in the research was higher liquidity volatility among small-cap stocks.

If you look at asset owners, they’re starting to talk about crisis risk offset trades, which allocate to strategies that seek to perform well in the event of an equity market collapse – some of which involve investments into CTA-type strategies
Spyros Mesomeris, Deutsche Bank

Other research includes linear signal blending. Here the team investigates how uncorrelated factor metrics might be combined to produce a better estimation of factors resulting in a higher Sharpe ratio. Naive signal weighting may not take account of correlations, may overweight volatile signals, or may simply be improved by incorporating alpha. The latter is achieved by allocating risk constraints for a given metric dependent on an investor’s belief in that metric’s returns.

In addition to published research, the team also works with clients on individual projects. One of the most challenging projects, says Mesomeris, was to find a way of managing a client’s equities portfolio risk budget more efficiently. The client had been reliant on equity beta risk but wanted to consider other sources of equity returns.

“We looked at effectively replacing some equity beta risk with “equity-like” factor risk, such as employing short volatility or implied dividend premium strategies,” he says.

Portfolio protection will be the next big topic, says Mesomeris. “If you look at asset owners, they’re starting to talk about crisis risk offset trades, which allocate to strategies that seek to perform well in the event of an equity market collapse – some of which involve investments into CTA [commodity trading adviser]-type strategies.”

Portfolio insurance is expensive and bleeds carry, but Deutsche Bank has recently researched how to hedge portfolios while benefiting from carry returns. The bank concludes that in low-volatility environments, delta-one instruments are a cheaper way of hedging portfolios than options.

As investors prepare for stormy weather, such strategies are likely to be popular.

  • LinkedIn  
  • Save this article
  • Print this page  

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an indvidual account here: