Marcos López de Prado @lopezdeprado Twitter profile | Pikagi

Pikagi

Marcos López de Prado

@lopezdeprado

22,732

Followers

47

Following

220

Media

1,518

Statuses

Professor @Cornell @BerkeleyLab . Sovereign Wealth Fund. Machine Learning for Asset Managers ( @CambridgeUP ). Advances in Financial Machine Learning (Wiley).

New York, USA

https://t.co/Z4v60tIcah

Joined May 2018

Don't wanna be here? Send us removal request.

Pinned Tweet

@lopezdeprado

Marcos López de Prado

6 months

I would like to thank the readers of The Journal of Portfolio Management for the "Outstanding Paper" award that they gave to my article "Where are the Factors in Factor Investing?" The full monograph, on which the article is based, is available here:

Tweet media one

2

16

100

Last Seen Profiles

@IanItBoy

@lazysoybean

@AviSchiffmann

@captaincurt81

@SankofaRose

@bokeplokalmalam

@MorganChmi35231

@amlkhalil11

@VoltItalia

@winttokkie

@KDB_Tech2

@kaIjabri

@bokeplokalmalam

@stwmaniax

@stw_pdg

@zany_fysl

@vox_lex

@vseviov

@sabirawansnk

@Georgia_M35

@lanita_moon

@AuthorMarkRunte

@thePeerJ

@Jason_Leonard9

@Ton1134998

@GACASoftball

@CreationsPrisha

@_pungate

@DJ_Joh23

@4OaC3XfOJc4lm8Y

@themandata1

@BinorRaja

@PlusExplicit

@lvchle

@ankaradkmp

@Sensasi3some

@lopezdeprado

Marcos López de Prado

5 years

A common misconception is that the risk of overfitting increases with the number of parameters in the model. In reality, a single parameter suffices to fit most datasets: Implementation available at:

Tweet media one

29

470

1K

@lopezdeprado

Marcos López de Prado

4 years

In preparation for the Fall 2020 semester at Cornell University, I've updated all slides for "Advances in Financial Machine Learning" (ORIE 5256, School of Engineering). They are downloadable for free at:

Tweet media one

20

206

932

@lopezdeprado

Marcos López de Prado

11 months

Why do Factor Investing strategies fail? To find the answer, read "Causal Factor Investing" (Cambridge University Press, 2023). This monograph can be downloaded for free here:

Tweet media one

12

76

417

@lopezdeprado

Marcos López de Prado

5 years

The greatest danger to finance workers is not automation. Their greatest danger is other (likely younger) workers with coding skills. People who can process large amounts of unstructured data, perform high-throughput tasks, and work alongside algorithms.

Tweet media one

5

117

393

@lopezdeprado

Marcos López de Prado

5 years

Alex Lipton and I are excited to share the first closed-form solution to the problem of finding the optimal trading strategy for mean-reverting processes: It should be helpful to market makers in particular, and execution traders in general.

Tweet media one

7

85

388

@lopezdeprado

Marcos López de Prado

5 years

All 10 seminars of 'Advances in Financial Machine Learning' have been updated, for the Fall 2019 course at Cornell University (ORIE 5256). They can be downloaded for free at . More materials available at

Tweet media one

7

78

352

@lopezdeprado

Marcos López de Prado

5 years

Just received the printer proof for the cover of the Cambridge book. Forthcoming March 2020. It can be pre-ordered through Amazon: I hope you like it!

Tweet media one

25

48

352

@lopezdeprado

Marcos López de Prado

5 years

Financial systems are extremely complex, with an overwhelming number of variables and non-linear interaction effects. In this context, correlations are a blunt tool for modeling financial codependence. This seminar introduces more powerful measures:

Tweet media one

6

80

336

@lopezdeprado

Marcos López de Prado

4 years

What are the difference between Forecasting and Nowcasting? Is Nowcasting just short-term Forecasting? Answers below, in a nutshell. For additional details:

Tweet media one

Tweet media two

Tweet media three

Tweet media four

10

77

326

@lopezdeprado

Marcos López de Prado

6 years

I feel truly honored, grateful and humbled by JPM's award. I hope it serves to draw attention to the many applications of financial machine learning.

Tweet media one

21

36

326

@lopezdeprado

Marcos López de Prado

5 years

Backtesting is not a research tool. If a strategy does not perform well in a backtest, do not tweak it (overfit) until the backtest looks good. Instead, investigate how the research process misled you into backtesting a false strategy. Fix the research process, not the strat.

Tweet media one

5

70

316

@lopezdeprado

Marcos López de Prado

4 years

Abu Dhabi Investment Authority (ADIA), UAE's biggest sovereign wealth fund, has appointed a Cornell University professor, Marcos Lopez de Prado, as global head of quantitative research & development, the fund said on Tuesday.

Tweet card media

News and Insights

32

30

305

@lopezdeprado

Marcos López de Prado

5 years

Many problems in finance require the clustering of variables. Despite its usefulness, clustering is almost never taught in Econometrics courses. In this seminar we review two general clustering approaches: partitional and hierarchical.

Tweet media one

7

63

271

@lopezdeprado

Marcos López de Prado

5 years

Seven investment problems where Machine Learning beats traditional methods (analytical solutions, econometrics, etc): 1. Hedging 2. Asset allocation 3. Outlier detection 4. Bet sizing 5. Feature importance 6. Default probability 7. Execution, HFT Can you suggest other examples?

20

61

269

@lopezdeprado

Marcos López de Prado

4 years

Many quant funds have performed poorly during the #COVID19 selloff, because they confound research with backtesting. In the scientific method, the purpose of testing is to refute a hypothesis, not to help formulate it. Backtest overfitting is ubiquitous:

Tweet media one

18

62

265

@lopezdeprado

Marcos López de Prado

5 years

Most asset managers build portfolios using historical (backward-looking) correlation matrices. To correct this problem, the TIC algorithm computes forward-looking correlation matrices, implied by knowledge graphs:

Tweet media one

6

59

262

@lopezdeprado

Marcos López de Prado

4 years

AI-Powered Hedge Funds Vastly Outperformed, Research Shows. Hedge funds using artificial intelligence returned almost triple the global industry average, Cerulli found.

4

59

254

@lopezdeprado

Marcos López de Prado

5 years

There are three fundamental ways of testing the validity of an investment algorithm against historical evidence: a) the walk-forward method; b) the resampling method; and c) the Monte Carlo method. To learn more about their pros and cons:

Tweet media one

9

49

252

@lopezdeprado

Marcos López de Prado

4 years

Many quantitative firms have suffered substantial losses as a result of the COVID-19 selloff. In this note, Prof. Lipton and I highlight three lessons that quantitative researchers could learn from this episode.

Tweet media one

13

57

244

@lopezdeprado

Marcos López de Prado

4 years

Decades ago, machine learning (ML) algos were black-boxes. In recent years, researchers have developed approaches that make ML as transparent as classical regression. This seminar explains how Shapley values enable ML-driven induction and abduction:

Tweet media one

6

60

241

@lopezdeprado

Marcos López de Prado

5 years

Most discoveries in finance are false due to p-hacking. This seminar explains how machine learning methods can overcome the flaws of p-values, and facilitate the discovery of new theories:

Tweet media one

4

62

238

@lopezdeprado

Marcos López de Prado

3 years

The False Strategy Theorem states that, with enough number of backtests, any Sharpe ratio level is achievable, even if the underlying investment strategy is unprofitable. Find the new paper published at The American Mathematical Monthly:

Tweet media one

6

53

234

@lopezdeprado

Marcos López de Prado

4 years

Cornell Engineering students will soon be enrolling in the Fall 2020 courses. Here are some of the topics that they will learn at ORIE 5256 (Advances in Financial Machine Learning).

Tweet media one

8

29

232

@lopezdeprado

Marcos López de Prado

4 years

Nice article. A key takeaway is that @CliffordAsness ' position is supported by a public track record. @nntaleb 's position is based on "selective leaked information." That does not mean that Taleb is wrong, but given AQR's transparency, Taleb has the burden of proof. Worth reading

Tweet media one

@opinion

Bloomberg Opinion

4 years

The best way to guard against so-called black swan events is a matter of perspective

1

34

101

8

32

227

@lopezdeprado

Marcos López de Prado

4 years

Is a Sharpe ratio of 2 better than a Sharpe ratio of 1? Not necessarily. Sharpe ratios are not comparable, unless we control for 3 variables: T, skew, kurt. The probabilistic Sharpe ratio (PSR) provides a coherent way to compare investment performance:

Tweet media one

7

41

230

@lopezdeprado

Marcos López de Prado

4 years

Please join me on June 10 @ 11am ET for an interview with Prof. Frank Fabozzi, editor of The Journal of Portfolio Management. We will discuss how machine learning is revolutionizing finance.

Tweet media one

17

36

222

@lopezdeprado

Marcos López de Prado

4 years

Very few market makers experienced losses during the #COVID19 selloff. They learned their lesson after the flash crash of 2010, and turned to nowcasting for managing real-time risks. COVID-19 should be the quant's Sputnik moment. Forecasting is the past. Nowcasting is the future

Tweet media one

9

46

217

@lopezdeprado

Marcos López de Prado

5 years

Traditional portfolio construction techniques (e.g., Black-Litterman) tend to produce unstable solutions, with high rebalancing costs. This paper introduces a machine learning algorithm for the robust estimation of the efficient frontier.

Tweet media one

Tweet media two

5

40

219

@lopezdeprado

Marcos López de Prado

4 years

After only 12 days, the book has been downloaded by more than 16,500 readers. A big thanks to everyone for your interest, support, and amazing feedback! Please get your FREE copy here (until May 6, 2020):

@CambUP_Econ

Cambridge Economics

4 years

New Cambridge Element Machine Learning for Asset Managers, by Marcos M. López de Prado, out now! Read for free for the next 4 weeks at #cambridgeelements

Tweet media one

14

63

191

15

33

214

@lopezdeprado

Marcos López de Prado

5 years

Eight ways to interpret the output of an ML algorithm: 1) PDP 2) ICE 3) ALE 4) Friedman's H-stat 5) MDI, MDA 6) Global surrogate 7) LIME 8) Shapley values @ChristophMolnar makes a convincing case that ML algorithms are not black boxes, contrary to popular perception.

5

60

208

@lopezdeprado

Marcos López de Prado

5 years

Financial ML tip: When data is scarce, prefer classification methods over regression methods. By discretizing the target array (y), we train the algo on a larger number of examples per target level. If you also need to predict the size of the effect, add a meta-labeling layer.

15

43

203

@lopezdeprado

Marcos López de Prado

5 years

If you are familiar with econometrics and wish to transition to machine learning, this paper is your road map: For each analytical step of the econometric process, there is an equivalent step in machine learning analysis.

Tweet media one

8

54

198

@lopezdeprado

Marcos López de Prado

5 years

Finance will change dramatically over the next five years. Hundreds of thousands of financial jobs are expected to be automated by 2025. It is important that finance employees are retrained to work alongside algorithms.

Tweet media one

4

67

190

@lopezdeprado

Marcos López de Prado

3 years

Is it possible to train machine learning models using tiny datasets? Surprisingly, the answer appears to be yes: This finding could have profound implications for financial machine learning, where datasets tend to be small and stationarity local.

Tweet card media

A radical new technique lets AI learn with practically no data

“Less than one”-shot learning can teach a model to identify more objects than the number of examples it is trained on.

www.technologyreview.com

9

39

193

@lopezdeprado

Marcos López de Prado

3 years

Key difference between (classical) statistics and machine learning: how they approach the bias-variance tradeoff dilemma. Generally speaking, if you know the data-generating process for certain, use statistics. If not (e.g., finance), use machine learning.

Tweet media one

2

57

187

@lopezdeprado

Marcos López de Prado

2 years

Are you an "association quant" or a "causal quant"? The answer depends on the type of empirical evidence that you produce to support your findings. If you wish to become a causal quant: 1) Read here: 2) Register here:

Tweet media one

5

25

189

@lopezdeprado

Marcos López de Prado

4 years

Francis Galton discovered the concept of Correlation in 1889. Ten years later, Karl Pearson wrote a paper warning that Correlation is not Causation. While most statistics books incorporate this warning in the first chapter, subsequent chapters tend to ignore it. Theories matter.

Tweet media one

6

40

184

@lopezdeprado

Marcos López de Prado

4 years

RenTec's Medallion fund is up 24% YTD. In this interview, Prof. Simons explains how Medallion uses machine learning:

Tweet media one

Tweet media two

6

35

180

@lopezdeprado

Marcos López de Prado

4 years

Do not forecast what you can nowcast. First, develop a strategy assuming that you had perfect foresight of a predictor (e.g., unemployment). Second, use machine learning to nowcast the predictor. Third, replace the perfect foresight with the nowcast.

Tweet media one

7

35

170

@lopezdeprado

Marcos López de Prado

5 years

“Top 25 Best Machine Learning Books You Should Read”

6

51

173

@lopezdeprado

Marcos López de Prado

5 years

By popular demand, I have expanded the explanation of Meta-Labeling in Session 3 of Advances in Financial Machine Learning (Cornell University, ORIE 5256): In particular, I explain how Meta-Labeling can transform a weak predictor into a strong predictor.

Tweet media one

3

33

167

@lopezdeprado

Marcos López de Prado

5 years

Machine learning (ML) is often misused in finance: . However, the current crisis in financial research was not caused by ML. It was caused by the misuse of classical statistics: . If applied correctly, ML can help solve this crisis.

6

44

169

@lopezdeprado

Marcos López de Prado

5 years

This short paper reviews 10 applications of financial machine learning: The presentation can be found here: Could you please propose additional applications?

Tweet media one

5

24

168

@lopezdeprado

Marcos López de Prado

5 years

Compare an algo that forecasted a change of 1 for a realized change of 3, with another algo that forecasted a change of -1 for a realized change of 1. Both algos made an error of 2, but only the second lost money. In finance, predicting the sign is often more important than size.

8

27

162

@lopezdeprado

Marcos López de Prado

5 years

A comprehensive introduction to machine learning:

Tweet media one

3

38

164

@lopezdeprado

Marcos López de Prado

1 year

In February 2018 "Advances in Financial Machine Learning" came out. Five years later, it is still a very popular volume among practitioners and academics, with translations in Chinese, Russian, Japanese, Korean, and Polish: A big thanks to all my readers!

Tweet card media

Advances in Financial Machine Learning

Advances in Financial Machine Learning

8

8

157

@lopezdeprado

Marcos López de Prado

5 years

Presentation slides for "Machine Learning Asset Allocation": Full paper available at

Tweet media one

1

36

161

@lopezdeprado

Marcos López de Prado

4 years

This seminar explains the pitfalls of standard portfolio optimization, and how machine learning can overcome those pitfalls:

Tweet media one

4

39

157

@lopezdeprado

Marcos López de Prado

5 years

The financial economics literature confounds backtesting with theory. Even if backtests produced accurate and representative measurements of a strategy's performance, they wouldn't tell us why the strategy made money. No matter how many years long, a backtest proves nothing.

Tweet media one

10

29

157

@lopezdeprado

Marcos López de Prado

3 years

A brief video introduction to the HRP algorithm for asset allocation: For the original white paper, see:

Tweet media one

3

33

154

@lopezdeprado

Marcos López de Prado

5 years

Machine learning algorithms can deliver robust estimates in cross-sectional studies. In this Monte Carlo experiment, RANSAC achieves an accuracy of over 95% in the presence of 25% outliers. In contrast, OLS' accuracy drops to 60%. See the video at

Tweet media one

6

31

153

@lopezdeprado

Marcos López de Prado

4 years

Aaron Brown argues that Universa's claimed 4,144% return in March 2020 is bogus, because the base used was the monthly premium paid. The true return was a mere 12.8%, vs. an average monthly loss of .22%. Not the first time we see Black Swan claims busted:

@justinaknope

Justina Lee

4 years

Aaron Brown fires another measured, smart salvo in the tail risk funds debate Those Astronomical Returns Aren't What They Seem

2

12

61

18

29

152

@lopezdeprado

Marcos López de Prado

5 years

A popular belief is that Machine Learning is more prone to overfitting than Econometrics. In reality, Econometrics is more likely to overfit due to its: (a) reliance on train-set error estimates, and (b) assumption that only one trial has taken place.

Tweet media one

5

43

150

@lopezdeprado

Marcos López de Prado

4 years

I respectfully asked @nntaleb to publish his track record, in support of his claim that his Black Swan strategy is profitable. Just as he did with @CliffordAsness , his response was to engage in personal attacks, and incite the attacks of his followers. Disappointing.

Tweet media one

@lopezdeprado

Marcos López de Prado

4 years

@JamesMarsh79 @nntaleb Funny, @nntaleb has fewer citations in SSRN: SSRN rankings: This is a distraction. The real issue is, where is the empirical evidence to support that Black Swan strategies work? Taleb, I wish you every success. @CliffordAsness

2

0

20

18

12

152

@lopezdeprado

Marcos López de Prado

5 years

I’m excited to join the faculty of Cornell University, as professor of practice at the School of Engineering:

Tweet media one

7

5

147

@lopezdeprado

Marcos López de Prado

5 years

I look forward to my testimony to Congress this Friday, where I'll explain how AI is transforming finance, and the impact it will have on America's jobs and competitive advantage. Follow the testimony live here:

Tweet media one

9

25

141

@lopezdeprado

Marcos López de Prado

6 years

The first issue of the Journal of Financial Data Science is out. You can download the articles here:

Tweet media one

1

41

139

@lopezdeprado

Marcos López de Prado

4 years

Study: Machine learning helps develop better market theories | Cornell Chronicle

Tweet card media

Study: Machine learning can predict market behavior | Cornell Chronicle

Machine learning can assess the effectiveness of mathematical tools used to predict the movements of financial markets, according to new Cornell research.

news.cornell.edu

4

45

139

@lopezdeprado

Marcos López de Prado

5 years

Why do most Econometric investments fail? Are they not supposedly founded in science?: This seminar explains the 7 reasons Econometric investments disappoint, and what could be done to correct this situation. Can you suggest other reasons and solutions?

Tweet media one

13

38

138

@lopezdeprado

Marcos López de Prado

5 years

Ashutosh Singh and @JacquesQuant have published a nice notebook on Meta-Labeling: I recommend it to anyone interested in boosting the Sharpe ratio of their investment strategies. Great job, and thanks for sharing!

Tweet media one

3

37

137

@lopezdeprado

Marcos López de Prado

6 years

A sincere thank you to my readers! One year after the publication of "Advances in Financial Machine Learning," the book still ranks among Amazon's top 5 bestsellers in Investments, Finance and Machine Learning. My pledge to you is to continue writing about problems you send me.

Tweet media one

15

12

136

@lopezdeprado

Marcos López de Prado

5 years

Podcast on Machine Learning and the future of finance. By via #soundcloud

Tweet card media

A M PE3 Marcos E1

Listen to A M PE3 Marcos E1 by Alumni-Mentor.com #np on #SoundCloud

1

23

137

@lopezdeprado

Marcos López de Prado

5 years

To a larger extent than other mathematical disciplines, Statistics is a product of its time. If Francis Galton, Karl Pearson, Ronald Fisher and Jerzy Neyman had had access to computers, they would have created an entirely different field: Machine Learning.

Tweet media one

3

34

133

@lopezdeprado

Marcos López de Prado

3 years

Prior to their launch, the strategies underlying factor investing ETFs delivered average annual excess returns of approx. 5% (i.e., in backtests). After their launch, average performance is 0%. For more information on Finance's replication crisis, see

Tweet media one

6

31

135

@lopezdeprado

Marcos López de Prado

5 years

An interesting article on how the Deflated Sharpe Ratio, and similar statistics, could help prevent backtest overfitting in crowdsourced AI investments:

Tweet card media

Overfit for Purpose: Why Crowdsourced AI May Not Work for Hedge Funds

The idea of crowdsourcing AI, or machine learning, which is what engineers prefer to call it, has been successfully explored by companies such as Kaggle and CrowdAnalytix. They generally use a...

3

33

136

@lopezdeprado

Marcos López de Prado

3 years

If you are interested in blockchain technologies, there is one book that you need to read: Professors Lipton and Treccani demonstrate their mastery of the subject, and provide tangible examples of how blockchain can replace many legacy technologies.

This link will take you to a page that’s not on LinkedIn

5

23

131

@lopezdeprado

Marcos López de Prado

3 years

This plot shows the Deflated Sharpe Ratio (DSR) on strategies with Sharpe Ratios of 1.5, 1, and .5 computed on 10 year backtests. The conclusion is that, in most cases, as few as 3 trials are enough to produce an investment strategy that is likely false.

Tweet media one

8

24

135

@lopezdeprado

Marcos López de Prado

4 years

In this video, the good people of @QuantConnect demonstrate how to build a Nowcasting strategy based on one of my seminars. I won't comment on the specifics of this particular strategy, but overall the video should be of interest: #MachineLearning

Tweet media one

3

16

132

@lopezdeprado

Marcos López de Prado

4 years

In this webinar, I discuss three quant lessons that we can extract from COVID-19:

Tweet media one

2

21

133

@lopezdeprado

Marcos López de Prado

2 years

When making a scientific claim, not all types of empirical evidence presented in support of that claim are equally strong. Some types of evidence are more susceptibility to being spurious than other types. For more details, see section 6.5 here:

Tweet media one

4

33

134

@lopezdeprado

Marcos López de Prado

5 years

There is an unresolved contradiction at the heart of financial economics. First, we are told that market are extremely efficient. Second, we are told that simple regressions suffice to extract billions of dollars in annual profits. Most econometric results are spurious by design.

8

28

128

@lopezdeprado

Marcos López de Prado

4 years

Many financial events are as unpredictable as earthquakes. By collecting unstructured observations from 1000s of sources, Nowcasting provides early warning systems. This can help investors hedge before panic ensues, and exchanges halt trading before circuit-breakers are needed.

Tweet media one

8

25

128

@lopezdeprado

Marcos López de Prado

5 years

Prior to commissioning the strategy, evaluate its probability of a false positive: After its deployment, monitor the following: * probability of alpha decay * probability of strategy drift * probability of a regime switch

Tweet media one

2

32

130

@lopezdeprado

Marcos López de Prado

5 years

Chen and Pearl [2013] found that six of the most influential Econometrics textbooks make fundamental mistakes, like confounding correlation with causation, or prediction with causation: . This confusion is widespread among articles in econometric journals.

Tweet media one

5

38

126

@lopezdeprado

Marcos López de Prado

4 years

Please join me for a CFA webinar on Machine Learning Asset Allocation.We will review the two main sources of instability in portfolio optimization, and how machine learning achieves better outcomes out-of-sample compared to traditional methods.

Tweet media one

5

24

125

@lopezdeprado

Marcos López de Prado

4 years

Only a couple of days left... Please make sure that you get your legal FREE copy of ML4AM from Cambridge's website. After Wednesday, they'll remove the link, and copies won't be free. Enjoy!

@CambUP_Econ

Cambridge Economics

4 years

New Cambridge Element Machine Learning for Asset Managers, by Marcos M. López de Prado, out now! Read for free for the next 4 weeks at #cambridgeelements

Tweet media one

14

63

191

4

33

126

@lopezdeprado

Marcos López de Prado

5 years

"Crowding is most likely an important factor in the deterioration of strategy performance [...] We identify significant signs of crowding in well known equity signals, such as Fama-French factors and especially Momentum."

2

28

126

@lopezdeprado

Marcos López de Prado

4 years

Join me on April 17th for a free webinar on quant lessons from COVID-19, nowcasting, and the estimation of theory-implied correlation matrices.

Tweet card media

Webinar | The Importance of Nowcasting: Three Quant Lessons from COVID-19 | April 17th 11:30am EDT

In this webinar, Marcos Lopez de Prado will highlight three lessons that quantitative researchers can learn from this episode based on his latest SSRN paper, which was developed in collaboration with...

www.thinknum.com

2

33

123

@lopezdeprado

Marcos López de Prado

5 years

The impact of AI on financial jobs:

Tweet media one

5

26

122

@lopezdeprado

Marcos López de Prado

4 years

The Bloomberg Quant Seminar series organized by Bruno Dupire is always a great opportunity to learn about new developments. On May 28th, I joined Alex Lipton, Peter Carr, Harvey Stein, Ioanna Boier, to cite a few. You can watch the event here:

Tweet media one

3

13

122

@lopezdeprado

Marcos López de Prado

4 years

Congratulations to all authors included in JFDS' new issue: JFDS has become the primary reference in financial data science because of our contributors' work. Thank you for expanding the frontiers of data-driven investing.

Tweet media one

1

10

122

@lopezdeprado

Marcos López de Prado

5 years

CLASSICAL STATISTICS (left): Correct specification+predictive variables are not enough. If you miss a single interaction effect, the forecasting power out-of-sample may be zero. MACHINE LEARNING (right): Just provide predictive variables. To learn more:

Tweet media one

Tweet media two

4

17

121

@lopezdeprado

Marcos López de Prado

4 years

Despite of this year's extreme gains, the overall performance of "Black Swan" hedge funds continues to disappoint. An alternative is to develop regime-specific strategies, by using nowcasting to switch between different algorithms:

Tweet media one

8

22

120

@lopezdeprado

Marcos López de Prado

4 years

It is time for @nntaleb to stop making sensationalist claims about Black Swan strategies, and publish his full track record. Taleb's arguments are far from strictly academic. As a co-owner of Universa, he has a conflict of interest. He benefits financially from these claims.

Tweet media one

4

17

115

@lopezdeprado

Marcos López de Prado

5 years

Some thoughts on the future role of financial economists, financial data scientists, ML tournaments, and the democratization of alpha: @numerai @richardcraib @DardenMBA @MebFaber @Risk_Journals @PM_Research_

Tweet card media

Marcos Lopez de Prado on the Democratization of Alpha

Marcos Lopez de Prado, co-founder and CIO of True Positive Technologies, discusses changes in quantitative investing processes, expected changes moving forwa...

www.youtube.com

1

20

116

@lopezdeprado

Marcos López de Prado

5 years

The "finance experts" vs. "data scientists" controversy is pointless. Rather than picking one, learn to use both. Tournaments provide an elegant solution, by splitting the investment problem into two specialized tasks, one for finance experts and the other for data scientists.

4

13

116

@lopezdeprado

Marcos López de Prado

3 years

Hence a major reason econometric investments often fail:

@ChristophMolnar

Christoph Molnar

@ChristophMolnar

3 years

We assume linearity

Tweet media one

6

75

470

7

18

114

@lopezdeprado

Marcos López de Prado

4 years

Please join me on May 28th at 17:30 ET for a FREE seminar at the Bloomberg Quant (BBQ) series. I'll discuss how ML-based portfolio construction methods accomplish a substantial reduction of the mean squared error, compared to traditional methods:

Tweet media one

1

23

111

@lopezdeprado

Marcos López de Prado

4 years

Two recent studies on Hierarchical Risk Parity (HRP): 1) 2)

Tweet media one

Tweet media two

6

27

110

@lopezdeprado

Marcos López de Prado

5 years

Compute the distance between 2 correlation matrices. def corrDist(corr0,corr1): num=np.trace((corr0,corr1)) den=np.linalg.norm(corr0,ord='fro') den*=np.linalg.norm(corr1,ord='fro') cmd=1-num/den return cmd

Tweet media one

Tweet media two

5

17

110

@lopezdeprado

Marcos López de Prado

5 years

What's the difference between backtest overfitting (BO) and backtest hyperfitting (BH)? BO: Each researcher runs millions of backtests, and report only the best ones to the company. BH: The company picks the best backtests among the already overfit backtests.

3

25

110

@lopezdeprado

Marcos López de Prado

4 years

Machine learning has an important role to play in finance. It solves the bias-variance dilemma without relying on unrealistic assumptions, even on small datasets. This presentation explains how:

Tweet media one

5

22

111

@lopezdeprado

Marcos López de Prado

2 years

Thank you to the over 1,500 people who registered to ADIA Lab's inaugural seminar, titled "Can Factor Investing Become Scientific?" To learn more, visit: * Slides: * Manuscript: * Paper:

Tweet media one

3

20

109

@lopezdeprado

Marcos López de Prado

4 years

Asset Management: Tools and Issues (World Scientific, 2020) provides a gentle introduction to the world of quantitative asset management: More publications on the topic can be found here (many of them for free):

Tweet card media

Asset Management: Tools And Issues

Long gone are the times when investors could make decisions based on intuition. Modern asset management draws on a wide-range of fields beyond financial theory: economics, financial accounting,...

0

24

106

@lopezdeprado

Marcos López de Prado

5 years

This is problematic, because many popular econometric methods regularize fitted functions by counting the number of regressors (eg AIC, aR2). This paper shows that those classical approaches are misguided. Model complexity is not necessarily a function of the number of regressors

7

13

108

@lopezdeprado

Marcos López de Prado

6 years

The Journal of Financial Data Science will publish its inaugural issue in the next few days. Authors include Harry Markowitz, Cam Harvey, Anthony Ledford, Ananth Madhavan, Sanjiv Das, Gordon Ritter, Petter Kolm, and many others. I hope you like it!

Tweet media one

1

31

105

@lopezdeprado

Marcos López de Prado

1 year

For a theory to be scientific, it must declare the falsifiable causal mechanism responsible for the observed phenomenon. With that in mind, this paper ranks the empirical evidence in finance, based on their scientific support:

Tweet media one

3

24

106

@lopezdeprado

Marcos López de Prado

4 years

A closed-form solution for optimal mean-reverting strategies

2

15

104

@lopezdeprado

Marcos López de Prado

3 years

"HRP delivers highly diversified allocations with low volatility, low portfolio turnover and competitive performance metrics." For the original study, see:

Tweet media one

1

22

107