Benjamin Moritz Deutsch

CV, research and teaching

Career and research

“The machine calculates. The people who decide should understand why.”

How do a family’s goals become a portfolio that lasts a lifetime? I look for answers with data, statistics and artificial intelligence.

Download CV (PDF)

40 projects since 2013, most of them with students

with students 32own research and with co-authors and institutes 8

One dot per project in the year it began; without lectures. More: the timeline with all projects

Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

Timeline

Since 2013 I have done research on financial markets, companies and real estate, with data, statistics and AI; since 2016 also together with students. I was closely involved, from the question to the result, in 32 master’s theses, industry projects and internships with more than 50 students from six universities. On top of these come four collaborations with doctoral researchers and institutes, and my lectures. And at work I do research every day, since 2008 at Sal. Oppenheim, HQ Asset Management and FINVIA; the row Work shows when. The figure shows where each project belongs in the investment process.

Projects and teaching by stage and year

with studentswith co-authors, doctoral researchers and instituteslecturesummer schoolseveral years One mark per project or lecture, at the year it started; a line shows how long it ran, an arrow that it is still running. The rows Work and Education show where I worked and studied; the diamonds are summer schools. The grey column on the left sums up 2001 to 2011. Titles of theses and reports in the original language.

Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  1. Summer 2012Lecture, Koblenz University of Applied SciencesMethods and teaching

    Financial Innovations

  2. 2013–2019Working paper, University of CologneExpectations

    Tree-based conditional portfolio sorts

    Tom Zimmermann (opens in a new tab)

    Decision trees predict share returns from recent months’ returns; in the test from 1968 to 2012 they earned about twice the excess return per unit of risk of regressions, before costs.

    Chart of the result

    About twice the excess return per unit of risk

    Information ratio of strategies that buy or sell short US shares by forecast return, against the four-factor model; test 1968 to 2012, estimated each year on the five years before only; before trading costs. Source: working paper of Benjamin Moritz and Tom Zimmermann, tables 1, F.1 and F.2; own chart.
    • The trees find interactions on their own instead of having them specified; the return of the last month matters most.
    • After roughly estimated trading costs a profit remains, but a much smaller one.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

    Best Paper Award 2015

  3. 2016–2020Chapter of the PhD thesis, LMU MunichEnvironment

    The relation between stock market risk and return

    Stefan Mittnik (opens in a new tab)

    The tone of New York Times articles on recessions since 1981 usually brightens when the stock market is at its low and swings the most: for investors it matters more whether the swings are rising or falling than how strong they are.

    Chart of the result

    What matters is the direction of the swings

    US stock market from 2007 to 2010, average return per year on the days of each group. The swings are the standard deviation over 22 trading days; their average was 22.9 per cent, about 16 per cent on calm and 39 per cent on stormy days. Source: PhD thesis of Benjamin Moritz, chapter 2 (with Stefan Mittnik), section 2.4; own chart.
    • Articles that mention “recession” pile up at the end of a recession; their tone, measured with sentiment dictionaries, turns earlier, around the low of the market.
    • Whoever leaves the market when it swings strongly avoids the fall, but often misses the recovery as well.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  4. 2016Master’s thesis, University of ZurichGoals

    Assessing the risk tolerance of retail investors: risk profiling in theory and practice

    Claudio Ambühl

    Online risk questionnaires are not very reliable: the same test client was recommended anything from 23 to 90 per cent in shares.

    Chart of the result

    Same client, 21 online questionnaires: how much in shares?

    rated 5 out of 10rated differently

    Test client: 30 years old, no prior knowledge, wants to invest 15,000 (in the provider’s currency) for 15 years. One bar per online provider: the equity share in the recommended mix, rounded. Even those that rated him the same (dark) recommended between 23 and 61 per cent in shares. Source: master’s thesis of Claudio Ambühl, University of Zurich, table 8; own chart.
    • Eight of the 21 providers recommended the same portfolio to all three test clients: to the 30-year-old as well as to the 60-year-old who retires in six years.
    • Only about four in ten questionnaires met the basic EU rule for investment advice (MiFID), and none asked separately how much risk someone wants to take and how much they can bear.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  5. 2018PhD thesis, supervisor Stefan Mittnik (opens in a new tab), LMU MunichExpectations

    Applications of textual analysis and machine learning in asset pricing

    PLEXUS Advancement Award 2020

  6. 2018Industry project, LMU MunichEnvironment

    Finanzmarktprognose mit Hilfe von Textdaten

    Karem El-Oraby and Raphael Rehms

    Articles from DIE ZEIT make monthly forecasts for German shares, bonds and the Ifo index slightly better: the forecast error falls by up to 9 per cent.

    Chart of the result

    How much smaller the forecast error became with newspaper text

    Forecasts for the following month, on the last third of the data up to November 2017, each made only with what was known at the time. Compared with a model that uses only the past of the series itself; for each series the best of 220 variants, so closer to an upper bound. Source: project report of Karem El-Oraby and Raphael Rehms, LMU Munich, 2018, section 5.3; own chart.
    • The basis: 469,137 articles of DIE ZEIT since 1946, turned into word counts month by month. A statistical method (LASSO) picked, out of 225,528 words, the few that helped the forecast.
    • Many of the words chosen had nothing to do with markets in terms of content: moods and topics beyond the economy carry information too.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  7. 2018Master’s thesis, LMU MunichExpectations

    Forecasting financial time series with deep learning

    David Milewski

    Neural networks called the direction of the US stock market right in up to 74 of 100 months; always saying “up” was right in 69.

    Chart of the result

    How often the direction was right: 108 months of US shares

    Share of months in which the forecast for the following month had the right sign, November 2008 to October 2017, US total market index; the networks were trained on monthly returns from 1959 to 2008. Source: master’s thesis of David Milewski, LMU Munich, 2018, section 3.2.5; own chart.
    • Because shares rose in more than two of three months, simply saying “up” is hard to beat. The best network got it right in 74 of 100 months, many others did worse.
    • Had one been invested only when the best network expected rising prices, the money would have grown by 168 per cent instead of 130 per cent with the index. But only 95 of the 648 networks beat the index, and which one is best is known only afterwards.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  8. 2020Master’s thesis, LMU MunichExpectations

    A comprehensive analysis of the use of deep learning models for forecasting the cross-section of stock returns

    Cem Öztürk

    Tested over 43 years on large US shares, two of nine neural networks earned more per unit of risk than a simple basket; the networks also showed which indicators matter.

    Chart of the result

    Nine neural networks: return per unit of risk

    Sharpe ratio of the tenth of shares each network rated best, bought only, January 1976 to December 2018, before costs; US large caps. Source: master’s thesis of Cem Öztürk, LMU Munich, 2020, table 3.7; own chart.
    • Tested over 43 years on large US shares, retrained every quarter with 66 indicators; which indicators mattered could be read from the networks: above all the book-to-market ratio.
    • Two of the nine came out ahead of the simple basket. The thesis shows how hard a broad basket is to beat and how much depends on the choice of network.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  9. 2020Industry project, UC BerkeleyExpectations

    Analysis of stock return predictors using profitability metrics

    Alex Bietrix and Paul-Noël Digard

    Models that voted each month on the direction of the S&P 500 cut its largest loss from −53 to −22 per cent, with the same return per unit of risk but less return.

    Chart of the result

    Timing the S&P 500: less return, much smaller losses

    January 1992 to January 2019, monthly: invested in the S&P 500 or in cash, depending on the vote of the models; long-only variant with a growing data window. Return per unit of risk (Sharpe ratio) 0.58 for both. Source: report of Alex Bietrix and Paul-Noël Digard, UC Berkeley, 2020, table 3; own chart.
    • Judged only by profit, not by forecast statistics: each month the models that had earned most over the last two years voted on whether the index would rise.
    • None of the combined variants beat the index; of 23 single indicators one did (nine-month momentum). Beating the market consistently is hard, whatever the method.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  10. 2020Industry project, UC BerkeleyEnvironment

    Business news topic modeling

    Zacharie Bouhnik and William Lambert

    The topics of the New York Times explain about 40 per cent of the swings of the US stock market and 16 per cent of its monthly returns, after the fact.

    Chart of the result

    How much of the US stock market the news topics explain

    Share of the month-to-month variation explained by the attention the New York Times gave to five topics in the same month (R²; volatility about 40 per cent). Explained afterwards, not forecast. Source: report of Zacharie Bouhnik and William Lambert, UC Berkeley, 2020, section 4; own chart.
    • The basis: 219,162 articles from the business section of the New York Times, 1998 to 2020, sorted by a topic model into 40 topics such as energy, pensions or crisis.
    • Attention to the topic “crisis” rose steeply as early as February and March 2007, months before the financial crisis broke. More attention to recession, loan problems and energy went with falling prices.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  11. 2020Industry project, UC BerkeleyAllocation

    Interpretable models for portfolio management

    Jiayin Guo and Yvonne Zhu

    Explainable regressions with chosen interactions built better long-short portfolios than decision trees: about half the largest loss and more return per unit of risk.

    Chart of the result

    Long-short portfolio: largest loss from a peak

    US shares 1980 to 2019, rebalanced monthly: bought the tenth with the best forecasts, sold short the tenth with the worst, equal weights. Return per unit of risk (information ratio): regressions 0.98 and 0.96, random forest 0.61, LightGBM 0.24. Source: report of Jiayin Guo and Yvonne Zhu, UC Berkeley, 2020, figures 3 and 4; own chart.
    • The aim was models one can explain: every year the indicators and their interactions were chosen anew, and Shapley values showed why a model picked a share.
    • For portfolios that only buy, the decision trees were ahead; for long-short portfolios the simple regressions, and interactions made the falls smaller.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  12. 2020Industry project, UC BerkeleyExpectations

    LSTM for stock return prediction and portfolio construction

    Xuanyi Ji and Shimai Zhang

    A neural network (LSTM) ranked US shares by expected return: very accurate in training, much less so in the test.

    Chart of the result

    How well the forecast matched the later return (correlation)

    Correlation between forecast and actual return of US shares; 1 would be a perfect match, 0 none. Test January 2017 to March 2019. Source: report of Xuanyi Ji and Shimai Zhang, UC Berkeley, 2020, appendix; own chart.
    • A typical picture with financial data: a network learns its training data almost by heart, and only a small part of that carries over to new months.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  13. 2021Working paperSelection

    Modern asset management

    Christian Maschner and Martin Schmitz

    Asset management has turned from a seller’s into a buyer’s market: investors want better results, lower costs and solutions of their own. Five levers help, from working culture to AI, and using them takes four kinds of know-how.

    Chart of the result

    What makes a modern asset manager

    What investors ask for today

    1. Better results
    2. Cheaper processes
    3. Solutions for each investor

    Five levers

    1. Working together, not in silos
    2. Scalable technology
    3. New data, such as texts
    4. Artificial intelligence
    5. Clear, simple communication

    Four pillars

    1. Finance
    2. Statistics
    3. Computer science
    4. Years of practice
    Summary of the paper; own chart. Source: Christian Maschner, Benjamin Moritz and Martin Schmitz, Modern asset management, April 2021.
    • The key question is not active or passive, but who evaluates the numbers best and most systematically, at low cost and fitting each investor.
    • Technology alone brings no edge; what counts is the team that applies it to the investment question.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  14. 2021Industry project, UC BerkeleyExpectations

    Forecasting stock market returns with deep learning

    Alec Madayan

    Across 1,728 settings of a neural network with memory (LSTM) it became clear how much forecasts depend on settings; none beat the market reliably.

    Chart of the result

    Return per unit of risk, depending on the settings of the network, 1980 to 2019

    Ranges over the settings tested (layers, neurons, learning rate and others), read off the charts of the report and rounded; before costs. Source: report of Alec Madayan, UC Berkeley, 2021, pp. 6–9; own chart.
    • Depending on its settings the network came close to the market or stayed clearly below: with neural networks the choice of settings matters a great deal.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  15. 2021Industry project, UC BerkeleyExpectations

    Forecasting stock market returns – focus: AutoML

    Ayush Gupta

    Automated AI (AutoML) found forecasting models on a par with carefully built regressions; without leverage none of them beat the market from 2000 to 2020.

    Chart of the result

    What became of 1 dollar, 2000 to 2020

    S&P 500, monthly; models chosen with data up to 2000, then retrained every month on the last 20 years; before costs and without the cost of borrowing for leverage. Source: report of Ayush Gupta, UC Berkeley, 2021, p. 7; own chart.
    • AutoML kept pace with a carefully built ridge regression: automation saves work, but it does not conjure up returns.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  16. 2021Industry project, UC BerkeleyExpectations

    Forecasting stock market returns – focus: ensemble

    Kunal Chakraborty

    Each combination of forecasting models did a little better than the single model of its kind; the best kept pace with the market from 1998 to 2020 although it was sometimes in cash.

    Chart of the result

    Invested or in cash, depending on the forecast: return a year

    S&P 500, test 1998 to 2020: invested in the month after a positive forecast, otherwise in cash; before costs. Single models on a rolling window of 60 months; the combinations learned on the first 60 per cent of the data and were tested on the rest. Source: report of Kunal Chakraborty, UC Berkeley, 2021, pp. 14–15; own chart.
    • Bets on falling prices paid off for no model; with double exposure two of ten beat staying twice invested.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  17. 2021Industry project, UC BerkeleyExpectations

    Investment strategy with long history data and feature engineering

    Taige Hong

    With data back to 1871 a clear picture emerged: a lead over simply holding US shares rested on a single signal that existed only from 1937 to 2002.

    Chart of the result

    What the lead rested on: a signal that existed only from 1937 to 2002

    US stock market, monthly forecasts, before costs; data of Amit Goyal and Ivo Welch and FRED. Signal: the cross-sectional beta premium. Source: report of Taige Hong, UC Berkeley, 2021, pp. 6–7; own chart.
    • Seventeen models from regression to LSTM networks; a very simple random forest did best.
    • The lesson: check what a lead rests on before trusting it.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  18. 2021Industry project, UC BerkeleyExpectations

    Studying the relationship between customer reviews and stock returns

    Yifei Tong

    Whether Amazon reviews anticipate share prices was tested for Apple, Microsoft and Garmin: links showed up just as often in the other direction.

    Chart of the result

    How often a clear link showed up, by direction

    Weekly share returns, adjusted for market moves, against stars, mood and word counts of the reviews, with up to three weeks lead or lag; Apple, Microsoft and Garmin, 2018 to May 2021; clear means statistically significant at the 5 per cent level. Own count from tables 6.1.1 to 6.3.3. Source: report of Yifei Tong, UC Berkeley, 2021; own chart.
    • Reviews were collected by the author, about 10,000 per company.
    • Links showed up as often in both directions: a sign that reviews do not reliably run ahead of prices.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  19. 2021Industry project, UC BerkeleyExpectations

    Textual analysis on financial statements to forecast single stock returns with ESG verbiage

    Neil Gulati

    In the annual reports of 25 large US companies, words on social issues became more frequent; a link between the share of sustainability words and the share return could not be shown.

    Chart of the result

    Words on social issues in the reports of 25 large US companies

    Share of all words in annual reports (10-K) and proxy statements, by the ESG word list of Baier, Berninger and Kiesel, mainly human rights and public health; read off the chart. Source: report of Neil Gulati, UC Berkeley, 2021, p. 7; own chart.
    • Words on corporate governance stayed about as frequent, at around 1.4 per cent of all words.
    • None of the text measures was clearly related to the share return.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  20. 2022Master’s thesis, TU MunichExpectations

    Automated architecture and hyperparameter optimization for deep neural networks applied in forecasting the cross-section of stock returns

    Antonio Di Giovanni

    From a thousand attempts a search program designed neural networks on its own; the best came close to a published network designed by hand.

    Chart of the result

    How well the networks put shares in the right order (rank correlation)

    Rank correlation between forecast and actual return; 0 would mean a random order. US shares without the smallest, test 2000 to 2021, trained on past data only; a measure without trading, hence without costs. The best network is the best of the 127 trained to the end. Source: master’s thesis of Antonio Di Giovanni, TU Munich, 2022, table 4.2; own chart.
    • The search tried 1,000 designs and trained 127 of them to the end; that it comes close to a design by experienced researchers is the main finding.
    • Measured here is how well the networks ranked the shares, without trading and so without costs.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  21. 2022Master’s thesis, TU MunichExpectations

    Empirical asset pricing via machine learning – an after trading costs perspective

    Dominik Eichhorn

    Before costs, decision trees and neural networks clearly beat a simple three-indicator model; after trading costs the networks kept only a small edge and the trees none.

    Chart of the result

    Long-short return a month, before and after trading costs

    Average monthly return from buying the tenth of shares with the best forecast and selling short the tenth with the worst, weighted by market value; US shares without the smallest, 1980 to 2017, retrained every year on past data only. Costs: half the bid-ask spread. Networks: the ensemble of five fixed in advance, not the best of the 14 models. Source: master’s thesis of Dominik Eichhorn, TU Munich, 2022, table 2; own chart.
    • Trading costs ate most of the lead: the networks traded far more and paid more than twice as much as the simple model.
    • After costs the networks kept about 0.1 percentage points a month, a little over one per cent a year.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  22. 2022Industry project, UC BerkeleyEnvironment

    Ensemble sentiment dictionaries for financial applications

    Philip Spalding

    A sentiment dictionary learned from the filings of US companies beat the standard dictionary in 9 of 12 later tests; as usual with texts, words explain only a small part of price reactions.

    Chart of the result

    Test periods in which the learned dictionary explained more than the standard one (of 4)

    8-K filings of companies in the S&P 500; test periods 2013–14, 2015–16, 2017–18 and 2019–20, each learned on the five years before; measure: share explained (R²), compared with the dictionary of Loughran and McDonald. Own count from tables 2 to 4. Source: report of Philip Spalding, UC Berkeley, 2022; own chart.
    • The learned dictionary joins three models, each trained on one task: return on the next day, over 30 days, and swings.
    • As usual with texts, words explain only a small part of price reactions, here always less than one per cent.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  23. 2022Industry project, UC BerkeleyEnvironment

    Evaluating performance of volatility managed portfolios with machine learning and macroeconomic data

    Divya Singh

    Adjusting the equity share to forecast turbulence worked a little better with AI and economic data than with the classic method: slightly more return per unit of risk.

    Chart of the result

    Return per unit of risk when the share of equities follows forecast turbulence

    Sharpe ratio; US stock market, March 1980 to June 2022; share of equities at most 150 per cent, lower when higher swings are forecast; forecasts with 15 economic series (FRED); before costs. Source: report of Divya Singh, UC Berkeley, 2022, p. 19; own chart.
    • XGBoost, a method made of many small decision trees, did best; the gaps are small.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  24. 2022Industry project, UC BerkeleyExpectations

    Using machine learning to develop equity style rotation strategies

    Andrew Lazzeri

    Averaged over all models, switching between investment styles gave a little more return per unit of risk than simply holding in the US, and less in Germany, Japan and China.

    Chart of the result

    Switching between investment styles: change in return per unit of risk

    Information ratio of model-driven switching minus that of holding each style permanently; 0 means as good as holding. Average over six styles (value, size, momentum, profitability, investment, low risk) and about 30 model variants, long-short, after 20 basis points of trading costs; forecasts month by month with data known at the time; USA 1970 to 2021, other countries from about 1990. Academic style portfolios, not directly investable. Source: report of Andrew Lazzeri, UC Berkeley, 2022, p. 19; own chart.
    • Simple, heavily regularised models did best; elaborate ones such as neural networks often did worse than holding. That ranking is only known in hindsight.
    • Shown is the average of about 30 models; the best ones did clearly better, but which they are is only known afterwards.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  25. 2022Industry project, UC BerkeleyEnvironment

    Volatility managed factor portfolios: leveraging machine learning and macroeconomic data

    Peter Masforroll

    Investing less when turbulence was forecast softened slumps in four of six investment styles; most of the protection already came from the classic method, AI helped above all with momentum.

    Chart of the result

    Less invested when turbulence is forecast: how much shallower the deepest slump became

    Largest loss from a peak, unmanaged minus managed, in percentage points; minus means a deeper slump. Long-short style portfolios, average over eleven forecasting methods and five regions; USA 1980 to 2021, other regions from 2000, 2001 or 2010; rolling, with data known at the time; before costs. Source: report of Peter Masforroll, UC Berkeley, 2022, figure 11; own chart.
    • The classic method without economic data (GARCH) softened slumps as much or more: in the US it cut the largest loss by 14 percentage points, the AI models by 10 to 13.
    • The effect was strongest in the US and absent in China; the economic data came from the US.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  26. 2023–2025PhD thesis, LMU MunichStress tests

    DCC-GARCH copula models

    Dennis Mao

    Daily returns of DAX, NASDAQ and gold from 1990 to 2023 show calm and stormy phases and far more extreme days than a bell curve; a modular model builds both into simulations.

    Chart of the result

    How much more often extreme days occur than under a bell curve

    Excess kurtosis of daily returns, 1990 to 2023, 8,288 trading days; 0 corresponds to the bell curve (normal distribution), higher values mean more extreme days. Source: PhD thesis of Dennis Mao, LMU Munich, 2025, table 1; own chart.
    • Gold went its own way: with DAX and NASDAQ it moved almost independently (correlation around 0), while DAX and NASDAQ moved clearly together (0.51).
    • The model joins building blocks for calm and stormy phases, skewed outliers, changing co-movement and joint extremes, and simulates 5,000 possible paths.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  27. 2023Industry project, UC BerkeleyPrivate markets

    Cash flow models for alternative investments

    Juliette Ould-Aklouche

    A standard model of the cash flows of private equity and real estate funds was rebuilt in Python and calibrated on real, finished funds: a basis for planning the calls and payouts of new funds.

    Chart of the result

    What the model describes: a commitment of 100 to a fund, payments by year

    −80−60−40−20020406080100123456789101112 Year

    Paid inPaid outBalance so farValue of the fund

    Example with a life of 12 years, 10 per cent growth a year and calls of 30 per cent of the open commitment a year; example values, not data of the report. Model after Takahashi and Alexander (Yale, 2002); own chart. Try it with the sliders on the Life and financial planning page.
    • The project calibrated life and growth on finished private equity and real estate funds of the vintages 2004 and 2009.
    • The model is rebuilt in Python, so shocks and changing growth can be played through.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  28. 2023Industry project, UC BerkeleyAllocation

    Deep understanding of the Python package “riskfolio-lib” and empirical application

    Mélanie Fabiani

    A freely available Python library for portfolio optimisation was examined in depth and tried out on bonds, shares and gold.

    Chart of the result

    Four asset classes and the mix the library computes from them

    Monthly returns in euros, January 2010 to May 2023: German government bonds, MSCI All Country World, gold; the mix with the highest Sharpe ratio after Markowitz, computed on the same data. Source: report of Mélanie Fabiani, UC Berkeley, 2023, tables 1 and 2, figure 2; own chart.
    • Although shares earned most, almost half of the mix is bonds: they swung less and often moved differently from shares.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  29. 2023Industry project, UC BerkeleyGoals

    Dynamic portfolio allocation in goals-based wealth management over multiple periods

    Elisa Piaraly

    A model that adjusts the mix every year to wealth and time left was rebuilt in Python: in the example, wealth doubled within ten years in 84 of 100 cases.

    Chart of the result

    How often 100 grows to 200, by years of investing (example)

    Share of 10,000 simulated paths in which a starting wealth of 100 reaches the goal of 200, with the mix adjusted every year from 15 portfolios; example with the public data of the paper by Sanjiv Das and others (2019), read off the chart. Source: report of Elisa Piaraly, UC Berkeley, 2023, p. 9; own chart.
    • When wealth is behind plan, the model picks mixes with more return; when it is ahead, more cautious ones, decided anew every year.
    • The rebuild matched the original code in 96 per cent of the investment decisions.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  30. 2023Industry project, University of CologneMethods and teaching

    Machine learning for economists

    More than 20 students

  31. 2023Industry project, UC BerkeleyExpectations

    Machine learning models for stock selection

    Nathan Sheng

    In a historical simulation, newer AI methods picked shares with more return per unit of risk than a simple regression; how much more depended on their settings.

    Chart of the result

    Return per unit of risk of the share selection (Sharpe ratio), 1985 to 2021

    US shares; each month the best tenth bought and the worst tenth sold short, equally weighted; per method the best of several settings, the regression with one; before costs. Source: report of Nathan Sheng, UC Berkeley, 2023, table 1; own chart.
    • The neural network, the hybrid LSTM and extra trees were ahead of the regression in every setting; for the others it depended on the setting.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  32. 2023Internship, Frankfurt SchoolAllocation

    Multi-period portfolio optimisation

    Jan-Luca Frick

  33. 2024Master’s thesis, University of ColognePrivate markets

    Forecasting net values and cashflows of private assets with machine learning

    Janina Dierk

    Over twelve test years, AI models forecast the capital calls and values of private funds much more accurately than the usual industry model, distributions less clearly; combining both did not help.

    Chart of the result

    Private equity funds: typical forecast error in the test, lower is better

    Root mean squared error in percentage points of committed capital; 2,713 private equity funds (Preqin), forecasts February 2012 to March 2024, retrained every six months with data known at the time. The industry model (Takahashi-Alexander) works with the age of each fund alone. Source: master’s thesis of Janina Dierk, University of Cologne, 2024, tables 4 and 11; own chart.
    • Similar for venture capital, private debt and real estate: smaller errors everywhere for calls and values, smaller gains for distributions. A second method (gradient boosting) did practically as well.
    • Combining the industry model with AI did not help. The age of a fund explained most; market and economic data added little.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  34. 2024Industry project, UC BerkeleyStress tests

    Modelling the global economy, public and private asset classes for scenario analysis

    Angel Chen and Jan-Luca Frick

    A model built on economic data such as jobs, interest rates and trade followed US inflation in calm years; like most forecasts at the time, it showed the surge of 2021 and 2022 with a delay.

    Chart of the result

    US core inflation and the model, year on year

    Consumer prices without food and energy; model from employment, capacity use, trade, the government budget, interest rates and semiconductor output (FRED), each estimated on 20 years of data. Values read off the chart of the final presentation and rounded. Source: Angel Chen and Jan-Luca Frick, UC Berkeley, 2024; own chart.
    • The model followed the rise about a year later, like many forecasts at the time.
    • How strongly each driver mattered changed clearly over the years.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  35. Since 2025PhD thesis, University of Cologne, University of TübingenGoals ongoing

    Investor expectations: analysis, forecasting and strategy with artificial intelligence

    Thorben Lützen and Felix Klosterköther

  36. Since 2025Industry project, Goethe University Frankfurt, Kiel Institute, University of TübingenExpectations ongoing

    Real estate: analysis, forecasting and strategy with artificial intelligence

    Tim Brück, Jonas Zdrzalek, Steffen Zetzmann and Felix Klosterköther

    Flats spread across German cities swung less than any single city; the mix that was best in hindsight would have added about 27 percentage points of final value from 1996 to 2025 over market weights.

    Chart of the result

    Flats in 20 German cities, 1996 to 2025: price growth a year

    Nominal prices, gains in value only, without rents and costs. Swings a year: 2.7 per cent for both mixes, 2.9 to 5.8 per cent for the single cities. The best mix is computed in hindsight; nobody knew it in advance, so the lead one could actually have reached is likely smaller. Values of the cities read off figure 2. Source: GREIX portfolio analysis (opens in a new tab), Kiel Institute, April 2026, figure 2; own chart.
    • Every single city swung more than the mix: spreading across cities smoothed prices in every phase of the market.
    • Which cities made up the best mix changed with the phase: since 2022 prices in Munich, Stuttgart and Hamburg fell, while Duisburg, Dortmund and Chemnitz still rose.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  37. Since 2025PhD thesis, LMU Munich, University of Tübingen, University of ColognePrivate markets ongoing

    Private markets: analysis, forecasting and strategy with artificial intelligence

    Konstantin Scheurer, Felix Klosterköther and Tom Zimmermann (opens in a new tab)

  38. 2026Industry project, UC BerkeleyGoals

    Goal-based investor profiling with mental accounts

    Stefan Wennemar

    Instead of an abstract risk number, investors name for each goal a loss they want to exceed only rarely; US survey data yielded four types of wealthy investors.

    Chart of the result

    Four investor types among wealthy US households

    Shares of US households with a net worth of at least 5 million dollars and of their net worth; Survey of Consumer Finances 2022, weighted. Source: final presentation of Stefan Wennemar, UC Berkeley, 2026, slide 26; own chart.
    • Business owners are just over a third of wealthy households but hold 42 per cent of their wealth.
    • Goals are stated as a loss limit with a frequency, for example a loss of more than 5 per cent in at most one year in 20.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  39. 2026Industry project, UC BerkeleyEnvironment

    Methods for extracting information from financial texts

    Jonathan Hartadi

    Six methods of text analysis compared: simple word counting tells a US company’s sector from its annual report almost as well as modern language models.

    Chart of the result

    Telling a company’s sector from its annual report

    Assigning 472 S&P 500 companies to 11 GICS sectors from the business description in their latest 10-K; F1 score per report on test data (1 would be perfect). Word counts: mean over five test rounds, each report tested once; language models: one run on one test round (95 reports). Source: project of Jonathan Hartadi, UC Berkeley, 2026, notebooks 09 and 09b; own chart.
    • All six methods lie close together, between 0.80 and 0.88; the language models lead logistic regression by 0.01 to 0.02.
    • In a second application, the tone of 490 annual reports from 2025, measured by a language model, was related to the share-price reaction in the following days.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  40. 2026Industry project, UC BerkeleySelection

    Mutual fund selection using alpha prediction

    Bailey Lin

    With the authors’ data a published study on picking funds could be reproduced; with newly prepared data up to 2025 the lead did not hold.

    Chart of the result

    Extra return a year of the funds picked (alpha)

    Each year the tenth of US equity funds with the best forecast was held, equally weighted; alpha against five factors, momentum and liquidity, monthly value times twelve. Replication of the study by DeMiguel, Gil-Bazo, Nogales and Santos (2023), once with their published data, once with data prepared anew. Source: final presentation of Bailey Lin, UC Berkeley, 2026; own chart.
    • How sensitive such results are to the preparation of the data is itself a finding: the new data held clearly more fund-years.
    • With the new data the differences between the methods are not statistically certain.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  41. 2026Industry project, UC BerkeleyPrivate markets

    Subscription lines and forecast bias in the Takahashi-Alexander model

    Hrafnhildur Líf Jónsdóttir

    When private equity funds bridge with credit lines, a forecasting model underestimates investors’ later payments in the simulation; knowing how common such lines are removes most of the gap.

    Chart of the result

    Cash shortfall in the simulation, with and without knowledge of credit lines

    Largest quarterly cash shortfall per fund in per cent of the commitment, the value 95 per cent of the simulated buyout funds do not exceed; forecast from the second fund year. Source: final presentation of Hrafnhildur Líf Jónsdóttir, UC Berkeley, 2026, slide 19; own chart.
    • Unaware of the credit line, the model reads too slow a pace from the low early calls and underestimates later payments.
    • For the correction, the share of funds with credit lines in the market is enough when the loans are repaid with investors’ payments.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  42. Winter 2026/27Lecture, Duale Hochschule Schleswig-HolsteinMethods and teaching ongoing

    Capital Markets

How the projects feed into my work with families: Life and financial planning.

Experience

  1. Since 2022
    Head of Investment Strategy & Research, member of the Investment CommitteeFINVIA, Frankfurt
    • Works out, with his team, the strategic and tactical asset allocation across all asset classes, liquid and illiquid, starting from client profiling, an analysis of the existing portfolios and life and financial planning across generations.
    • Forecasts returns, risks and correlations, optimises portfolios over several periods, runs scenario analyses and stress tests, and models cash flows and NAVs of private markets.
    • Takes care, with his team, of portfolio construction, manager selection and monitoring of the liquid investments.
    • Leads research collaborations with students and doctoral researchers from international universities.
    Inside an office with dark tables and plants; through tall windows the lit towers of Frankfurt’s banking district at dusk.
    Frankfurt, 2025: the view from the FINVIA office over the banking district.
  2. 2018–2022
    Executive Partner, part of the founding teamHQ Asset Management, Düsseldorf
    • Designed a cloud-based investment platform (R, Python, SQL, Azure, Docker, GitHub) for asset allocation, risk management, equity selection, forecasting, optimisation, backtesting, explainable AI and AutoML.
    • Brought artificial intelligence and alternative data into the investment process, led research and portfolio projects across the firm and developed ESG-integrated solutions with MSCI Barra.
    • Supervised master’s students on data-driven projects, supported sales and presented research at conferences and universities.
    The Medienhafen in Düsseldorf at dusk: office buildings on the harbour basin, reflected in the water; behind them bridges over the Rhine.
    The Medienhafen in Düsseldorf, 2019. This is where HQ Asset Management had its office.
  3. 2008–2018
    Vice PresidentSal. Oppenheim, Cologne
    • Co-led the asset allocation strategy for €7.1 bn of institutional and €1.6 bn of private assets; set up the strategic allocation process and the annual outlook.
    • Developed quantitative models for tactical and strategic allocation, including risk simulation, factor allocation and macro indicators.
    • Managed five mutual and institutional funds (€0.5 bn) and contributed to mandate proposals in the investment committee.

Papers

The relation between stock market risk and return

With Stefan Mittnik (opens in a new tab). Chapter 2 of the PhD thesis. Presented 2016–2020, among others at the 7th NextGen AI Conference.

Over decades, more fluctuation brings more return. Over weeks and months this is unclear; research comes to contradictory results. This paper looks for the answer in the newspaper. Through the New York Times' programming interface, all 94,427 articles since 1851 that contain the word “recession” were found and counted day by day; The Economist had counted the word only quarterly in 1992. From 1981 dictionaries also measured how gloomy or confident the articles sound.

How often the New York Times writes “recession”

010203019301950197019902010

Articles with “recession” per day, average over 20 daysRecession in the US (NBER)

Since the 1950s the number has risen clearly in every recession, mostly peaking towards its end or shortly after, and has then fallen again. The highest value came in 2009 with just over 28 articles a day. There were spikes without a recession too, for instance after the crash of October 1987 and in 2016.

New York Times, articles per day that contain “recession”, 1926 to 2017, moving average over 20 days. Source: PhD thesis of Benjamin Moritz, chapter 2 (with Stefan Mittnik), figure 2.3; redrawn from it. Recessions: NBER via FRED (opens in a new tab).

Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

For investors the peak in the count comes too late. The tone of the articles, by contrast, turns confident just when prices are lowest and fluctuate most, as in 1982, 1991, 2001 and October 2008. What is risky, then, is not strong fluctuation but rising fluctuation: from early 2007 to its peak on 28 October 2008 the market lost 17 percent a year, from then to the end of 2010 it gained 22 percent a year.

Full list of papers on SSRN (opens in a new tab) and Google Scholar (opens in a new tab).

Talks

33 talks since 2013: at central banks and regulators, universities, asset managers and conferences. 9 of them, on the joint paper, were given by the co-author Tom Zimmermann; they are marked.

Central banks and regulators 5

  1. 2015
    Federal Reserve BoardWashington, D.C. · Tree-based conditional portfolio sortsgiven by co-author Tom Zimmermann
  2. 2015
    Bank of EnglandLondon · Tree-based conditional portfolio sortsgiven by co-author Tom Zimmermann
  3. 2015
    Bank for International Settlements (BIS)Basel · Tree-based conditional portfolio sortsgiven by co-author Tom Zimmermann
  4. 2015
    Norges BankOslo · Tree-based conditional portfolio sortsgiven by co-author Tom Zimmermann
  5. 2016
    Commodity Futures Trading Commission (CFTC)Washington, D.C. · Tree-based conditional portfolio sortsgiven by co-author Tom Zimmermann

Universities 12

  1. 2013
    Harvard University, Financial Economics SeminarTree-based conditional portfolio sortsgiven by co-author Tom Zimmermann
  2. 2014
    Harvard University, Financial Economics SeminarTree-based conditional portfolio sortsgiven by co-author Tom Zimmermann
  3. 2014
    University of Zurich, Computational Economics and Finance SeminarTree-based conditional portfolio sorts
  4. 2014
    LMU Munich, Financial Econometrics SeminarTree-based conditional portfolio sorts
  5. 2016
    LMU Munich, Financial Econometrics SeminarThe relation between stock market risk and return
  6. 2017
    LMU Munich, Financial Econometrics SeminarThe relation between stock market risk and return
  7. 2018
    LMU Munich, Financial Econometrics SeminarThe relation between stock market risk and return
  8. 2018
    LMU Munich, Colloquium of the Department of StatisticsTree-based conditional portfolio sorts
  9. 2021
    TU Munich, Advanced Seminar in Finance and AccountingBuilding a modern asset manager with technology and AI
  10. 2021
    University of Cambridge, Optimal Portfolio Research SeminarBuilding a modern asset manager with technology and AI
  11. 2025
    University Alliance Ruhr, workshop “Narrative Economics: Science meets Practice”Narrative economics: narratives as early indicators (keynote)
  12. 2026
    Darmstadt Business School, MBA programmeDarmstadt · AI in investment management: what managers can learn (guest lecture)

Asset managers and banks 4

  1. 2014
    Sal. OppenheimCologne · Tree-based conditional portfolio sorts
  2. 2014
    Wellington ManagementLondon · Tree-based conditional portfolio sortsgiven by co-author Tom Zimmermann
  3. 2015
    BlackRockCalifornia · Tree-based conditional portfolio sortsgiven by co-author Tom Zimmermann
  4. 2017
    DWS InvestmentsFrankfurt · Tree-based conditional portfolio sorts

Conferences 12

  1. 2015
    11th World Congress of the Econometric SocietyMontreal · Tree-based conditional portfolio sorts
    An empty conference room: rows of chairs, a panel table with a laptop, a lectern and a screen.
    The room in the Palais des congrès in Montreal, 2015, shortly before Benjamin Moritz's talk at the 11th World Congress of the Econometric Society.
  2. 2015
    22nd Annual Meeting of the German Finance AssociationLeipzig · Tree-based conditional portfolio sorts
  3. 2015
    9th International Conference on Computational and Financial EconometricsLondon · Tree-based conditional portfolio sorts
  4. 2015
    6th CEQURA ConferenceMunich · Tree-based conditional portfolio sorts
  5. 2015
    2nd SOFINE-CEQURA Spring Junior Research WorkshopBavaria · Tree-based conditional portfolio sorts
  6. 2016
    15th Colloquium on Financial MarketsCologne · Tree-based conditional portfolio sorts
  7. 2019
    10th CEQURA ConferenceMunich · The relation between stock market risk and return
  8. 2019
    AI and Data Science in TradingLondon · Building a modern asset manager with technology and AI
  9. 2019
    The Trading Show EuropeLondon · Building a modern asset manager with technology and AI
  10. 2020
    The Trading Show EuropeLondon · Building a modern asset manager with technology and AI
  11. 2020
    7th NextGen AI Conference (textual analysis)Frankfurt · The relation between stock market risk and return
  12. 2025
    Preqin webinar “The Rise of Private Capital in DACH”online · Private markets in Germany, Austria and Switzerland: allocation, liquidity, data (panel)

Teaching

  1. Winter 2026/27
    Lecturer, Capital MarketsDuale Hochschule Schleswig-Holstein, Business Administration, 5th semester, 5 ECTS, together with Dr. Florian Heitger.
    • Five parts: market, environment and measurement; interest rates and credit; companies and shares; further markets such as real estate, private equity, currencies and derivatives; portfolio, with portfolio theory, the CAPM and the way from goal to portfolio.
    • Own material: about 1,250 slides with exercises and solutions, a calculation drill and a practice exam.
    • Business game “Investment committee”: groups of four invest the assets of a made-up foundation, present twice and at the end test their portfolio in three unknown scenarios.

    In the business game the students sit on the investment committee themselves. Each year the foundation pays out 3% for its purposes, pays 0.4% in costs and wants to make up for inflation of 2%. That gives the target return of 5.4%, and with it the question of how much risk it takes.

    What the foundation has to earn each year

    = 5.4% target return

    • 3.0% payout
    • 0.4% costs
    • 2.0% inflation

    Rules of the committee

    • €10 m of assets
    • cash at least €340,000, one year of payout and costs
    • private markets at most 20%
    • each single share at most 5%
    • no loans, no crypto assets, no short selling

    Course of the game

    1. 1start
    2. 2presentation 1
    3. 3portfolio frozen
    4. 4presentation 2 and test in three unknown scenarios
    Made-up foundation, rules as on a real investment committee; not graded.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  2. Summer 2012
    Lecturer, Financial InnovationsKoblenz University of Applied Sciences, Remagen campus, Department of Mathematics and Technology. Master's programme Mathematics in Finance and Life Science, 2 hours a week.
    • Three parts: foundations of macroeconomics and growth (inflation, business cycles, interest rates, from the Malthus to the Solow model), the history of financial innovations with a focus on investing, and financial econometrics (regression, portfolio theory, CAPM, the Fama-French model, GARCH).
    • Programming in MATLAB: efficient frontier and resampling, Monte Carlo simulation, simulated share prices, and an exercise with the 25 portfolios by size and value from Kenneth French's data library.
    • Exam as a seminar paper in MATLAB, in groups of two or three, for example on risk-based optimisation (minimum risk, risk parity, maximum Sharpe ratio), mean-CVaR or multi-factor models.

    A guiding idea of the lecture: on financial markets history is quickly forgotten. The economist John Kenneth Galbraith thought that in hardly any field history counts for as little as in finance, and the same mistakes come back again and again. A timeline of financial innovations therefore ran through the lecture, from the first bonds in Venice to portfolios of risk premia.

    850 years of financial innovations

    1200140016001800190019502000
    1. 1172Bonds in Venice After the break with Byzantium, Venice raises an interest-bearing forced loan from its citizens; in 1262 the loans become one debt at 5 per cent whose shares are traded.Investments
    2. 1602Shares for everyone Anyone can subscribe to the Dutch East India Company, and its shares are traded continuously on an exchange; tradable shares had existed in Toulouse since the 14th century, but in a small circle.Investments
    3. 1611Exchange in Amsterdam Amsterdam builds an exchange where the company's shares are traded; soon forwards and options are agreed there as well. Exchanges for bills were older, in Bruges and Antwerp.Markets and institutions
    4. 1637Tulip mania Prices of tulip bulbs in Holland soar and collapse in February 1637. By a strict measure it was hardly a bubble: trade was small, and the famous story comes largely from moralising writings of the time.Crises
    5. 1694Bank of England 1,268 subscribers lend the state £1.2 million at 8 per cent and become the bank's shareholders. Only the Swedish Riksbank (1668) is older.Markets and institutions
    6. 1774The first investment fund After the crisis of 1772/73 Abraham van Ketwich launches “Eendragt Maakt Magt” in Amsterdam: diversification for investors with little money.Investments
    7. 1780Protection against inflation Massachusetts pays its soldiers with notes whose value is tied to the prices of corn, beef, wool and leather.Investments
    8. 1896The first share index Charles Dow publishes the average of twelve industrial shares: the Dow Jones.Ideas
    9. 1900Prices as a random walk Louis Bachelier describes share prices as a random walk and lays the ground for valuing options.Ideas
    10. 1929Crash and Great Depression The crash on Wall Street is followed by the deepest economic slump of the 20th century, and by the regulation of banks.Crises
    11. 1934Security analysis Benjamin Graham and David Dodd value companies from their balance sheets and earnings.Ideas
    12. 1944Bretton Woods Fixed exchange rates to the dollar, the dollar convertible into gold, plus the International Monetary Fund and the World Bank. The gold window closes in 1971, fixed rates are given up in 1973.Markets and institutions
    13. 1949The first hedge fund Alfred Winslow Jones buys shares on credit, hedges part of them with short sales and keeps 20 per cent of the profits; it counts as the first hedge fund.Investments
    14. 1952Portfolio theory Harry Markowitz shows how spreading money lowers risk: the beginning of modern finance.Ideas
    15. 1964CAPM William Sharpe explains returns by the risk of the market, the beta.Ideas
    16. 1970Efficient markets Eugene Fama: prices contain everything that is known, so beating the market for long is hard.Ideas
    17. 1971Index funds Wells Fargo invests a pension fund's money in all shares of the New York Stock Exchange instead of picking some; from 1973 funds track the S&P 500, and in 1976 Vanguard opens one to private investors.Investments
    18. 1973A price for options Fischer Black, Myron Scholes and Robert Merton find a formula for the value of options; in the same spring the Chicago Board Options Exchange opens. Nobel Prize 1997 for Scholes and Merton, Black had died in 1995.Ideas
    19. 1979Fear of losses Daniel Kahneman and Amos Tversky: losses hurt about twice as much as gains please, a cornerstone of behavioural finance.Ideas
    20. 1987Black Monday The Dow Jones falls by 22.6% in a single day.Crises
    21. 1993Three factors Eugene Fama and Kenneth French: besides the market, size and valuation explain returns.Ideas
    22. 1994Insurance against default J.P. Morgan passes the default risk of a credit line to Exxon to the European Bank for Reconstruction and Development for an annual fee; by the common account the first credit default swap.Investments
    23. 1996Risk parity Bridgewater launches “All Weather”: each asset class is to contribute the same to the risk.Ideas
    24. 2000Technology bubble The Nasdaq falls from 5,048 points in March 2000 to 1,114 in October 2002; in the same spring Robert Shiller's book on overvaluation, “Irrational Exuberance”, appears.Crises
    25. 2008Financial crisis On 15 September 2008 Lehman Brothers files for creditor protection; the financial crisis deepens.Crises
    26. 2010Portfolios of risk premia Diversifying across risk premia instead of asset classes (Bender, Briand, Nielsen and Stefek).Ideas
    Selection from the lecture of 2012, told anew. Dots: tap for the text; the axis is drawn wider from 1890, because most happened in the last 130 years.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

    The whole story, from clay tablets to AI: 5,000 years of economy and finance.

Education

  1. 2012–2018
    PhD in Statistics (Dr. rer. nat.), LMU MunichSupervisor: Prof. Stefan Mittnik. PhD thesis: Applications of textual analysis and machine learning in asset pricing. Completed alongside full-time work.Thesis (PDF) (opens in a new tab) Defence slides (PDF)
    The Siegestor on Ludwigstraße in Munich on a winter evening; on the right the corner building of the LMU Department of Statistics.
    Ludwigstraße in Munich with the Siegestor, February 2013. In the corner building on the right, Prof. Stefan Mittnik held his chair at the LMU Department of Statistics, where Benjamin Moritz did his PhD.

    Stefan Mittnik took his PhD in 1987 at Washington University in St. Louis. Among his teachers there, he names Hyman Minsky, the economist of financial and banking crises, and the physicist Edwin Jaynes with his maximum-entropy methods as strong influences. The key experience came in October 1987: a few weeks into his first post as assistant professor at Stony Brook, he saw the Dow Jones fall by more than 20 per cent in a day, a move that the models of the time, built on Brownian motion, did not allow for. With Svetlozar Rachev he then developed models of the capital market with fat tails, following Benoît Mandelbrot, who had brought stable distributions into economics (“Stable Paretian Models in Finance”, 2000). Later, with Willi Semmler, he studied how stress in the banking system can make the economy unstable; in 2014 he co-founded Scalable Capital. A chapter of the thesis was written with him. Two questions from that time run through this website: what happens at the edges of the distribution rather than in the average, and why long calm can prepare the next crisis. Source: Walter Krämer, interview with Stefan Mittnik, AStA Wirtschafts- und Sozialstatistisches Archiv 17 (2023), pp. 101–107 (opens in a new tab)

  2. 2004–2008
    Diploma in Business Mathematics, Koblenz University of Applied SciencesDiploma thesis: Vergleich zweier Portfoliokonstruktionsstrategien im aktiven quantitativen Portfoliomanagement: Long-Only vs. Short-Extension (130/30), 2008.

    The diploma thesis compared two ways of building an equity portfolio against an index, in a simulation over five years: long-only, which may only buy shares, and short extension, which may also bet on falling prices (go short), by short selling or with futures on single shares (single stock futures), typically 130 per cent long and 30 per cent short. The conclusion was a cautious yes: the short extension carries the fund manager's view into the portfolio more fully, but costs more and brings new risks; which leverage is best depends on the market.

    Long-only or short extension: how much of a view reaches the portfolio

    longshort wantedweight in the index

    1. A
    2. B
    3. C
    4. D
    5. E
    6. F
    7. G
    8. H
    View carried out
    Short share

    Illustration with eight invented shares, A the largest in the index. Small shares limit long-only most, because their weight is small. In the thesis the gap was measured with the transfer coefficient.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

Summer schools and courses

  1. 2021
    Writing clearly (“Verständlich Schreiben kompakt”)National Institute for Science Communication (NaWik), Karlsruhe, online
  2. 2016
    Advances in Macroeconomic Forecasting: Nowcasting, Turning Points, and Extreme EventsGabriel Pérez-Quirós, CEMFI, Madrid
  3. 2014
    Financial Econometrics and Finance MarketsEnrique Sentana (CEMFI), University of Crete, Rethymno
  4. 2011
    Intermediate MacroeconomicsLondon School of Economics
  5. 2010
    Financial Econometrics and Forecasting: Modern and Advanced TechniquesFrancis X. Diebold, Swiss Finance Institute, Geneva

Internships, school and student jobs

Even at school and during his studies the work revolved around the themes of today: reading balance sheets, describing risks, running stress tests, steering private holdings and building portfolios.

  1. 2007–2008
    Intern, Active Quant ProductsSal. Oppenheim, Cologne. September 2007 to March 2008, wealth management, active quantitative equity selection
    • Helped develop and introduce two new ways of building portfolios: short extension (130/30) and minimum variance.

    Minimum variance means: the portfolio should move as little as possible, without any forecast of returns. That works because ups and downs cancel out when shares do not move in step. At Sal. Oppenheim he helped develop and introduce such a strategy; there it mixed many shares, here two are enough to see the idea.

    Mixing two shares: spreading lowers the risk

    oppositein step
    Mix with the least risk
    Its risk
    Its return

    Invented example: calm share 4% return and 12% risk a year, lively share 7% and 22%; no short positions. The curve shows all mixes from 100% calm to 100% lively.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  2. 2006–2007
    Placement semester, Credit Risk Portfolio AnalysisDresdner Bank, Frankfurt. September 2006 to January 2007, Risk Governance & Controlling
    • Helped develop macroeconomic stress tests (multi-factor models, SUR estimation) and scenario analyses for the bank’s credit portfolio.

    How does a crisis hit a bank's loans? First a scenario is translated into numbers for the economy: growth, interest rates, the oil price, the stock markets. Statistical models then show how much the defaults of companies rise in each sector and region. Finally this is carried over to the credit portfolio: how much loss is to be expected, and how much capital has to be ready?

    From a scenario to the credit portfolio

    1. Economy

      • Growth↓
      • Interest rates↓
      • Oil price↓
      • Stock markets↓
    2. Defaults by sector

      • Cars
      • Engineering
      • Chemicals
      • Metals
      • Transport
      • Retail
      • Hospitality
      • Construction
      • Real estate

      hit first

    3. Credit portfolio

      Expected losstodayin the scenario
      Capital neededtodayin the scenario

    America buys less, so German exporters sell less. Interest rates and the oil price fall, and so do share prices. Sectors that deliver a lot to America are hit first, such as cars and engineering.

    Simplified drawing of the method with example scenarios. Arrows show the usual direction, a double arrow a strong move; no figures. The bars only roughly show which scenario hits harder.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  3. 2005–2006
    Student employee, fund controlling for venture capitalKfW-tbg, Bonn. November 2005 to June 2006, holdings in young technology companies on behalf of the federal government, then about 750 companies and 50 venture capital funds
    • Monitored the funds and direct holdings the tbg had invested in (portfolio management), mainly by analysing fundamental data.
    • Fund controlling: linked data sheets, annual and quarterly reports; developed the risk management system further with new databases.

    The tbg invested directly in young technology companies and through venture capital funds. Portfolio management meant keeping watch over these holdings, above all with fundamental data: revenue, costs, cash. For young companies that still make losses one question counts most: how long will the money last?

    Watching a portfolio: how long does the money last?

    1. Software20 months
    2. Medical devices14 months
    3. Semiconductors9 months
    4. Biotech16 months
    5. Sensors11 months
    6. Lasers22 months
    7. Solar7 months
    8. Robotics13 months
    A year or more5 of 8
    6 to 12 months3 of 8
    Under 6 months0 of 8

    Five companies have money for at least a year, three for six to twelve months: they will soon need a new financing round.

    Invented portfolio of eight young technology companies. Months of money left = cash divided by the monthly outflow (burn rate). Funds were watched as well; they are not shown here.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  4. 2005
    Student assistant, Industrial Risk ServiceAXA, Cologne. September 2005, head office
    • Cleaned the data of a claims analysis and showed it as distribution charts.
    • Fitted distributions to the observed claim sizes and frequencies (lognormal, Weibull, exponential and extreme value distributions) and presented the results.

    The Industrial Risk Service dealt with the losses of corporate clients: how often does something happen, and how expensive does it get? The typical picture is lopsided: most claims are small, but the few large ones make up a large part of the costs. To plan for them too, one fits distributions to the observed claims. Which one is chosen decides above all how much one expects from the large claims.

    Many small claims, a few expensive ones: which distribution fits?

    small: up to 5,000 eurosmedium: up to 50,000 euroslarge: above 50,000 euros

    Number of claims

    observed
    fitted

    Costs

    observed
    fitted

    The exponential distribution gets the average right but knows almost no large claims: it expects 0.3 instead of 16 claims above 50,000 euros per 1,000, and their share of the costs falls from 25% to 0.3%. With it, an insurer would underestimate the large claims.

    Invented example: 1,000 claims, half of them below about 2,000 euros, on average about 6,200 euros. Here they were drawn from a lognormal distribution; with real claims nobody knows that beforehand, so several distributions are compared. For the very largest claims, extreme value distributions are used as well.

    Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

  5. 2001
    Temporary assistant, credit monitoringVR-Bank Rhein-Sieg, Troisdorf. July and August 2001, summer job while at the business grammar school in Siegburg, credit department
    • Kept the record of the documents that borrowers must submit under section 18 of the German Banking Act and requested missing ones from customers.
    • Analysed smaller balance sheets under guidance.

    His father looked after corporate clients at this bank for decades. Working in his bank that summer, Benjamin Moritz learned in practice how a bank gets to know a company before it lends: through its balance sheet. The balance-sheet analyses were the most exciting part of the job, and the look at a company's figures has stayed with him, from credit monitoring to the balance-sheet ratios in his AI models.

The common thread: what began at school and university

  1. 2001, VR-BankAnalysing balance sheets
    TodayBalance-sheet ratios in AI models for selecting shares
  2. 2005, AXADescribing claims statistically
    TodayForecasts of returns, risks and correlations
  3. 2005–2006, KfW-tbgMonitoring funds and direct holdings
    TodayCash-flow and NAV models for private markets
  4. 2006–2007, Dresdner BankStress tests for a credit portfolio
    TodayScenario analyses and stress tests
  5. 2007–2008, Sal. Oppenheim130/30 and minimum variance
    TodayPortfolio construction and optimisation
Grey: the positions at school and during his studies; teal: the focus of his work today.

Chart: Dr Benjamin Moritz, benjaminmoritz.de/en/cv/

Awards

  1. 2020
    PLEXUS Advancement Award for Artificial Intelligence in FinanceFor the PhD thesis “Applications of textual analysis and machine learning in asset pricing”; presented in March 2021

    PLEXUS Investments gave the award for the first time in 2020, for scientific work on artificial intelligence in finance. The thesis links textual analysis, such as counting newspaper articles on recession, with machine learning, such as decision trees for picking shares.

  2. 2015
    Best Paper Award, annual meeting of the German Finance AssociationFor Tree-based conditional portfolio sorts, with Tom Zimmermann

    The German Finance Association (DGF) is the scientific association of finance in the German-speaking world; its annual meeting 2015 took place in Leipzig. The paper uses decision trees to sort shares by several characteristics at once.

Working together

Students and researchers who would like to write a thesis, join an industry project or work on a joint paper can write to me with a short outline of the idea and the timeline.