Andreas Steiner on LinkedIn: Perspective On Risk Management (2024)

You can now register for the next free ApaLibNET seminar in July.https://lnkd.in/dwuMdeHm

3

If you are not already a subscriber, you can still register for my free ApaLibNET webinar tomorrow.The webinar is dedicated to functionality related to temporal aspects of data: converting high frequency data to lower frequency data, turning quarterly returns for illiquid assets into monthly observations, working with data windows and exponential weighing schemes for data.Not rocket science or very exciting conceptually, but useful and basic functionality in applied data science.#datascience #quantiativefinance #statisticalanalysis #riskmanagament #investmentmanagement #quantiativeanalysis #ApaLibNET #Excelhttps://lnkd.in/dCSfqU_h

1

I will be in Geneva on June 11 and 12. Contact me if you would like to meet for a coffee, I am not a tight schedule. ☕🫖🥤🧋

2

I see a lot of practitioners struggling what to think about the whole data science (DS), machine learning (ML), artificial intelligence (AI) thing. Traditionally trained quantitative financial experts many times either go into full opposition or naively take over the aspirations and promises by these approaches.I would like to present the current state of my thinking in a one-day online seminar as outlines below. If you are interested, leave a comment or drop me a line. I need a (small) minimum number of participants to justify the time it takes to put together the materials. I think I found a way to get to the really interesting questions without getting distracted by the technical details too much.

9

3 Comments

What happens if we exchange Pearson correlation coefficients in portfolio construction with one of the commonly used distance measures (e.g. Euclidian, Chebyshev, Minowski)?This is possible because one can always convert correlation values into distance measures (and vice versa) with a simple mapping function.With the US stock/bond correlation, not much will change: all of the common distance measures are highly correlated with the Pearson correlation coefficient. This is not a surprise, as all measures are calculated from the same information set (i.e. the same time series data).But note that the dynamics are different: certain distance measures behave less erratic over time, which gives more stability over time when recalculating/rebalancing portfolio strategies over time. The disclaimer here is that this observation might be spurious, as the mapping function is defined up to a linear transformation only, and therefore the dynamics arbitrary up to a certain degree. This also means that correlations from distance measures can be looked at as shrinkage estimators for Pearson correlation coefficients.Just some loose thoughts. This is work in progress. I might add this discussion to my correlation seminar in July. For details, visit https://lnkd.in/drtVn8gW

21

1 Comment

As usual, market participants are "riding the bubble": aggregate equity exposure has reached an all-time high, according to Goldman Sachs.As if rebalancing has never existed. And after bubbles crash, market participants act surprised. Although it is not clear whether we are currently experiencing a bubble. But also this is typical: conceptually, the idea of a bubble is very much an ex-post (in hindsight) construct.

5

2 Comments

The #longterm realized #performance of US #stocks, US #bonds and #gold 1928 to 2023 based on annual data compiled by NYU Stern School of Business.I still do not see a good #investment case for gold.What dominates in the long-run analysis is the higher expected return of equity.Also interesting the convergence to the normal distribution: for equity and bonds, #nonnormalities and temporal dependence (1st and 2nd order #autocorrelation coefficients) are a minor issue.Gold, on the other hand, has plenty of #excesskurtosis (i.e. fat tails) and a strong momentum (large positive autocorrelation). I think these characteristics make gold an interesting #speculative asset. Seen as an "investment", gold is just very risky, especially then in relation to its bond-like return.It would be interesting to run these statistics on monthly data. As I do not have good access to market data, I cannot do that. But happily accept data "donations" (which will be treated anonymously) 😉

2

3 Comments

Has anybody done a comparative study about correlation-based, Euclidian, Manhatten Chebyshev or Minkowski distance matrices? I.e. whether there exist systematic differences out-of-sample when using historical distance matrices?Also, has anybody suggested using exponentially-weighted versions of Euclidian, Manhatten Chebyshev or Minkowski distances?I see some talk about distance concepts, but very little practical action.I might be presenting something in my forthcoming course about Correlations and Other Dependency Concepts in July, see https://lnkd.in/drtVn8gW for details.

5

Last reminder: tomorrow, I will hold the next free ApaLibNET webinar. Topic is clustering functionality (hierarchical and k-means), distance functions and graph theory functionality (Freeman centralization, Minimum Spanning Trees etc.)You can register on LinkedIn (MS Teams meeting link will become visible after registration) or on my website https://lnkd.in/engM8DV9 (bottom of page)https://lnkd.in/dE9Fzm9N