18 Burst results for "Andrew Lo"

Tropical Storm Fay: What New York, New Jersey and Massachusetts Can Expect

WBZ Afternoon News

00:56 sec | 11 months ago

Tropical Storm Fay: What New York, New Jersey and Massachusetts Can Expect

"Continue to track the path of Tropical Storm Fay Tonight. The latest A storm of the Atlantic hurricane season made landfall earlier this afternoon right outside Atlantic City, New Jersey. Now it's making its way North. National Hurricane Center says it's about 90 miles south of New York City, making its way toward the north at about 14 miles an hour. Right now, the storm has maximum sustained winds of 50 miles an hour, and it's packing quite a bit of rain coming along with it now. Andrew Lo canto is the forecast with the National Weather Service, and he has more on the focus of the storm. There is expected to hit the hardest actually currently receiving it now, uh, basically mental, Annick ST Newjersey seven New York into Connecticut flashlight. Watch also up for part of western Massachusetts. Depending on where you are, you could see a bit of flooding. And if you're on the south facing Ocean Beach is a bit of high surf as well. So of course we'll keep you posted on the tropical storm throughout the night. Right here on W busy

New York City Atlantic City National Hurricane Center Andrew Lo New Jersey Annick St Newjersey National Weather Service Ocean Beach Massachusetts Connecticut
"andrew lo" Discussed on Linear Digressions

Linear Digressions

16:22 min | 1 year ago

"andrew lo" Discussed on Linear Digressions

"Ns in the patterns of missing this in the data and the way that depending on what your hypothesis about why data is missing. That can have a big impact on the way maybe you analyze and impute the data or fill in the gaps or ignore it or drop records. There's lots of different things that you can do. In different applications you can find different answers depending on which one you pick so I would love if you can dig into a little bit more. What were some of the the patterns or lack thereof in the missing nece and the data. And how does that inform for the analysis strategy that you took in trying to fill it. That's a really important area that we spent a lot of time on and in fact I would argue that. The innovation in in our publication is much more on the side of dealing with missing data than it is machine learning and in fact the machine learning techniques that we use are pretty well known. And there's not a whole lot of innovation in that dimension. The problem missing data in Pharma. Examples is that for various different clinical trials. The data are so messy that even though the features that you'd like to use are clearly they're they're not always measured in every instance so a good example is the root of administration Australian of a drug. Clearly it's important to know whether or not you're going to be getting dragged into the patient via injection IV drip pill. An Aerosol era saw that kind of information is present in most cases but there are a number of drug indication pairs for which the route of administration is simply missing. And and it's important to be able to take that measure and use it rather than to eliminate all the data points for which route of administration is not hot measured. Because if we do that if we simply eliminated all of those missing data cases we would end up having incredibly sparse amount of data left and I think if I recall correctly that was. That was one thing that it looks like. You explored from more from a due diligence perspective than as your your primary strategy strategy. But just looking at how. How much does it degrade the models. If we were to just get rid of everything where we don't have complete records and I guess one thing that all just add as an aside aside and and please correct me if I'm wrong but part of the reason that this is so important is because many of the sort of off the shelf methods that you were using though crash if you if you send and in records that have missing data so unless you want to build your own custom algorithm that has who knows what properties that make it so that it for some reason doesn't crash when it tries to split the decision tree. Say on record that you don't have the you have a missing Dean for if you WANNA use those off-the-shelf methods than you do need to think about. How how are we going to make this. Look like a fully complete data set. I suppose exactly very often. You'll find that the more features you would like to use the more likely it is that you're going to run at this missing data problem. And so you've got to trade off use a small number features and then you'll have a larger data set or use a larger double features and then and figure out how to deal with missing data. We chose the latter strategy because of the two hundred features that we had access to if you simply used a complete fleet records approach in other words deleting all the observations where you had any missing features we would probably deleted about eighty percent of our data and for folks who are not as familiar with the protocols. Let's say around clinical trials. Why is all this data missing there. Just no incentive to collect in a unified way or what well. It wasn't until recently that the federal government instituted a policy that's at any kind of clinical trial has to be registered At clinical trials dot Gov and certainly before that it was kind of a wild wild west but even after the case there are variety of additional features that are not required to be reported in clinical trials dot Gov but nonetheless would be really useful for prediction and so What Informa- Does is collect all of that data as best that can using zing of writing of different means public sources and some cases private sources and they aggregate it and they try to clean it to the best of their abilities. Board it means is that a clinical trial. That's done say in Boston Massachusetts. Asa -Chusetts at one of our area hospitals will have fantastic data but a clinical trial. That's done somewhere. In a third world country may not be as sophisticated. The individuals involved involved in clinical trials may not have the experience to be able to collect all of that data. And so you're putting all of this information in the same platform that's why you've got very heterogeneous amounts counts of data in across these various different venues. And so when it. When the time came to actually to the imputation there are a number of different methodological approaches? Is that one can take to to doing this. imputation just like do we do imputation in the first place is a question. How do we do? It is one that has a garden of forking paths. There's many different ways you can go and they can change the outcomes so I was wondering if you could talk through the way that you architect the work so that you could do a a search that was the right amount of thorough thorough. But that also didn't overwhelm you with many different options and in particular I think one of the dangers in trying a bunch of different bunch of different models a bunch of different imputation techniques in that you can end up with a in a place where you're cherry picking results. You aren't fully confident that the thing that looks the best is actually the best. So from a from a scientific perspective and kind of from an organizational perspective. How did you think about the imputation. What did you end up deciding to to use for your final final model. So the starting point is Tundra stand. Exactly what methods are out there for doing imputation and by the way imputation is a word that statisticians tend to use more than data scientists interpretation. Extrapolation curve fitting are other terms that have been used before and each of these terms Actually represents a different statistical technique. In fact in some cases they're not even statistical curve fitting for example is a tried and true method of filling in missing data data. And so what we decided was to focus on a somewhat more statistical approach. I guess my background is ticks. So I tend to favor those methods. where I I understand from a probabilistic perspective what the methods are likely to do to my inferences? So we actually followed the literature of statistical imputation Asian. That's pioneered by Don Rubin and many of his collaborators and students the basic idea is to treat missing data as parameters to be estimated and there are the variety of different methods for estimation and we outlined those in our supplemental materials and we tried several of those methods and we actually did a very simple analysis this where we started with the subset of non missing data so in other words we actually went through the process of eliminating all data with any missing records records and that basically reduced data set to a very very small subset. We then took this data set and randomly decided to create missing nece by. Oh I throwing out different features for different observations in particular systematic way and then we apply the imputation algorithms to the so called gold standard data where we know what the missing observations are. And then we compare the machine learning forecast with the missing observations and imputation versus with the complete data. Set what we found and was that our methods actually worked pretty well pretty well in the sense that the inferences that we got from the imputation data set was pretty close to the gold standard. The data set now. The one caveat about this approach is that our way of introducing missing nece is using randomness and in the cases where the missing this is due to non random effects. That's GonNa Create bias that we can't capture or deal with so that is a limitation of our result and something that we highlight in in our supplemented materials and our main taxed and something that I think all data scientists Neil technique grapple with some manner for sure for sure. So you you you. You bite the bullet on the imputation and then once you have this completed data set Part of it organic part of it. Sort of synthetically completed can put into Some fairly standard tools coming out of psychic learn and are the ones that you mentioned in the paper and so when you use that machine learning approach it will give you a bunch of predictions it. You validated it. Both cross validated in the sense of looking at cases for which you already knew the answer. Standard intercross allegation and then also apply the model to some other drugs. That are still in the pipeline. So we'll know in five to ten years if you were right hopefully lessen some cases but it takes a while L. and one thing that you and I talked about a little bit before we sat down for this interview that I mentioned appreciating was the way that I felt that you were very careful the way that you wrote the paper to always stay on the predictive side in the language that you used versus the influential side so you mentioned a little bit what I mean without as you mentioned a little bit earlier that you kind of learned some interesting things. You've found some interesting patterns in the data but you were very careful. Oh not to draw. Conclusions like this feature is important and it's causing certain types of outcomes because machine learning methods and generally have to be careful about that so I wanted to give you a chance to to speak to that a little bit more What you think of. As some of the strengths and weaknesses of of machine learning approaches versus more statistical approaches. And the second question that I think is actually really interesting is because you have external partners who presumably probably at the very least curious about what you are finding in this study. How you if you struggled it. All to explain that dichotomy of prediction versus inference to them. And this is something we all struggle with if you have any tips for folks like me or the people you're listening about how you try to make sure that they are thinking about your results in a way that is scientifically sound. I don't know if I have any tips but I certainly can share your pain and then talk a little bit about the way that we've been dealing with it and thinking about it so first of all you're absolutely right. That distinction between prediction and explanation is absolutely critical particularly when you're did with biomedical applications because lives are at stake. And so you don't want to be making inferences when really all you're doing is identifying correlations so the way that we've been dealing with it is to try to determine first of all whether or not the predictions are robust by swapping out features by looking at how sensitive the the features are by. China identified the key features that lead to a particular prediction..

federal government Boston Informa Massachusetts Dean China Don Rubin
"andrew lo" Discussed on Linear Digressions

Linear Digressions

10:30 min | 1 year ago

"andrew lo" Discussed on Linear Digressions

"Hiring one this week's episode very excited about we are doing an interview with Professor Andrew Lowe at Mit Sloan School of business. Now you might be wondering what is the business this quote professor. Doing data science podcast. And I'm extremely excited to have him here. Because I think some of the items that he researches and studies Saudis are actually super important to data scientists who are trying to have real impact in the organizations that they work in. So thank you so much professor low for joining me this week. Thanks for having me So we'll be talking about a paper that you just wrote and published in Harvard Data Science Review About Modeling Drug Trial Outcomes and I think without any further ado. Let's get into it. You're listening to linear digressions so to start out as I mentioned. Your Business School professor and but as I was going back through some of the publications that you've done in the past some of your research interests it looks like you've had an evolution from things that look like more classical local business topics like finance those sorts of things into topics that focused more on that. Have some overlap with data science. Is that about right. That's that's right yeah. I started out very early on studying. Large amounts of data back when data. Science wasn't a really phrase. We call it empirical finance and in in particular looking at Millions of transactions of stock prices and trying to make sense of all of that data in some kind of a systematic way and then more recently focusing on applying apply machine learning techniques to studying consumer credit for mortgages credit cards and so on and It was really because of personal reasons. I got interested in healthcare number of friends and family dealing with cancer cancer. That got me to start working in this area. We covered the actual paper that you wrote in last the last episode of Linear digressions. And so for folks. Who Haven't I heard that one yet. I would strongly advise you to go back and give it a listen because there's a lot of layers this paper that that we spend a little bit of time unpacking. But I'd like you to start with. If you could give a maybe sixty to ninety second recap of the paper as as you thought about it as the papers author like what is the what were you trying to explore. What did you find well. First of all this paper was authored with two of my graduate. Students Qingwei Yacht and she him Wong and the three set out trying to understand whether or not it's possible all to predict clinical trial outcomes using various features of the drug as well as the particular clinical trial design and really the emphasis that we're taking taking is not so much on the underlying scientific outcomes but rather the clinical and in our perspective the financial outcomes. We really really come at it from the point of view of an investor. That's thinking about putting money in a particular biotech company or Pharma Company and we want to know as an investor. What's the likelihood that we're GONNA hit pay dirt and in this world that means making sure that your clinical trials have a good outcome and I think that's a really interesting thing to just double click on for folks who aren't as familiar with the clinical trial world. That as you say there's a there's a scientific aspect to it like. Does this medicine actually work. But that that's not exactly the same thing is will. This eventually turned into a successful product. So you could go another layer deeper with us right now and disentangle those a little bit more as someone wants. He's actually an expert at this. Sure I'll be happy to give it a try way that I started thinking about this from experience that I had talking to a very well all known biotech company. I was introduced that by Tech Company because my mother at the time was struggling with non small cell lung cancer and this company had a number of therapeutics one of which might I have helped her and so I was privileged to meet with the chief scientific officer and asked him what I thought was a relatively straightforward question and that is does your source of financing. have any kind of influence on your scientific agenda and I was very concerned about that agenda because the priority of the particular compound that could have helped my mother was is not very high on their list and the CFO and the CIO. The two of them together looked at each other and then turned to me. And that she she cited officer said influence. Our financing drives our scientific agenda. And you know as an economist. I get so you have to pay for stuff but as son of a dying patient. I was just absolutely outrageous and offended by that because what a stock market volatility interest rates and Fed Policy House. You have to do with whether you should treat cancer by angiogenesis inhibitors or immunotherapy. Nothing and yet. Those kind of considerations drive this biotech a company scientific agenda and I thought you know despite the fact that I'm not a biotech expert. They've got it backwards. The science should be driving the financing. Not the other way around and so I started looking into this and the more I looked into it more. I realized that the big issue in all of these circumstances is risk ask our investors going to get their money back after putting in in some cases hundreds of millions or billions of dollars over the course of ten or fifteen years on on these clinical trials and so while machine learning has been used to great effect in designing molecules doing pie throughput screening and other kinds of bio informatics. At the time. We didn't see anything going on on. Predicting Clinical trial outcomes for the purpose of investing. And that's the kernel of the paper. That's sort of our topic of discussion today. So to recap a few things that are making good framing for the next part of the conversation you worked with. I think a commercial sial one or two commercial vendors to take advantage of some data sets that they had gathered around. Clinical trial attributes outcomes on which was formed the the backbone of the analysis that you did and then as you kind of alluded to the outcome you're trying to understand was whether a drug will eventually reach approval from the FDA at at starting at two different but critical points in the pipeline so the phase two in the face three in the in the clinical trials process. That's that's right. What we're trying to understand is what the likelihood of approval is and we look at approval from the vantage point of face to drug as well as from face three-drug just to give have different investors in those different life cycles. A chance to get a sense of what the risks are and at the outset. What did you think would be the hardest thinks about building that model in practice as an experience data scientist. You probably know the answer already. It really has to do with the data. Yeah and I was. We were taken aback. Because you know we got the data from a commercial vendor Informa- They're a well known name. In the field of biomedicine they have a number of products called sideline biomed tracker far projects. These are things that people in the Bio Pharma industry us all day long. And so our presumption was if they were willing to give us an academic democ license. We would just be able to take the data and machine learning algorithms on it. We found out pretty quickly that there was a huge amount of effort. Involved in merging two who'd disparate data sets one about the particular drugs that are involved in clinical trials and diseases that they involve the drug indication pairs. We call it. But then second another database that has all the features of the clinical trial design for example. How many patients weren't a clinical trial? How long did it take to crew those patients. So we'll typically ended up having to spend a fair bit of time understanding what the underlying definitions were of the data that was collected and then merging to make sure that there were no mismatches mislabels and so on and that probably took around six to nine months of effort just to merge them and then once did it. The actual machine learning exercises rather quick Yeah that's Initally that's been my experience to you. Spend all the time cleaning your data. Hey at the end you you run it through second. Learn and Declare Success I suppose the one thing I did want to ask about before we leave the data a little bit too much is for folks who are working in industry They're usually working with data sets sites either that they buyer licensed from kind of data vendors. It sounds kind of similar to the use case that you have here or they'll be doing data science on internally collected data sets like doing analysis of data collected for AB testing or looking at customer retention or acquisition type data sets. So I'm I'm wondering waiting for you when you were interested in doing this. Academic approach using a commercial data site. Is that something that there is a a well trod path between and someone like you and someone that has data set that you're interested in or did you find it challenging at all to explain or get buying from sort of your partners in order to get that license for the data and to build the model that you're interested in. Oh was definitely the latter because what we were trying to do is something that apparently had never been done before before well especially from the financial perspective most financial investors haven't really thought about the problem of forecasting clinical well trial outcomes using these large data sets mainly it's because the scientific expertise is generally what they look for when trying to understand the prospects of a clinical trial so they'll get key opinion leaders that are experts in Alzheimer's or one kind of cancer or another and look at the design of the trial. Look at the molecules that are involved in clinical trial and basically make a judgment judgment as to whether or not this is a good or bad investment. The problem is that that's information. Confined to a very specific instance. And it doesn't necessarily generalize realize across a large number of trials and across a large number of different drug indication pairs the reason that we are so excited about getting access to this data. It was because we wanted to to make the best of all of the various different cases out there and figure out whether there's information that could be gotten and from our analysis we actually found some really interesting patterns. That in retrospect were pretty straightforward. But nobody in advance told us that we ought to be looking down those allies so we will get to. Yeah those are comes in a moment. I like to leave that as a little bit of a cliffhanger. People stay tuned into the episode. But I wanted to again to you mentioned briefly. How important and hard some of those data quality issues. Where when you first found it one of the things. We spent a little bit of time in the previous episode. Talking about was the missing..

professor Professor Andrew Lowe Mit Sloan School of business Business School Pharma Company Informa Qingwei Yacht chief scientific officer Bio Pharma officer Fed Policy House FDA Wong scientist Alzheimer
"andrew lo" Discussed on FT Alphachat

FT Alphachat

02:02 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Needs desires and predilections once we do that i think they'll be an incredible renaissance of indexation that will take us to the next generation the future john bogel 's and vanguards and i suppose the final question algorithm is another one of those he woods has a bad name among many of the people who will be listening to this how much can we reduce human intervention and how much can we leave to algorithms we're understanding the potentialities and the the areas of our own brains veteran better over time how far can that go well one of the key misunderstandings of artificial intelligence is that we're creating algorithms to replace human behavior and intervention and i think the better way to think about it is that we're developing algorithms to leverage human behavior and intervention what we want to be able to do is to allow humans to use their judgments in ways that are going to be most effective while at the same time automating the things that can be easily automated and we're at the stage now where finance has a lot of automation but we haven't yet figured out how to design strategies so that they can actually help manage portfolios to the tune that individuals need to have given their own individual circumstances where close but we're not there yet and if we can do that if we can develop algorithms that can actually allow us to manage our portfolios as if we were the ones calling the shots but in a rational fashion that would reduce a lot of the issues that investors face today were close but we're not there yet i think in another ten or fifteen years will actually be a lot closer and at that point we may achieve what john maynard keynes set out as a goal for all communists that is going to economists should be like going to the dentist exactly more pleasant but you have the same basic tr.

john bogel john maynard keynes woods fifteen years
"andrew lo" Discussed on FT Alphachat

FT Alphachat

02:02 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"On the less risky more profitable parts of their business which is late stage drug development and marketing and drug delivery what we really need to deal with is early stage drug development that's gotten a lot riskier because we've actually gotten quite a bit smarter about developing these drugs and so all of the amazing designer drugs that are coming out that's creating an increased risk at the early stages that we can actually help reduce using financial engineering okay and that's a way of dealing with an adaptive market failure a problem with the with monk it's president my final question you've hinted already that passive funds may now be becoming too central to market so the next crisis is going to involve passive funds one way or another where do you see the ecosystem moving next tell me who the herbivores will be in who the carnivals will be in another ten or twenty years time i think that index funds are going to have to change the way they construct portfolios and match various kinds of portfolios to investors gone are the days when you can put all of your money in one fun and forget about it and believe that when you retire you'll have enough money to be able to do so in the style to which you would like to become accustom i think what we have to do now is to think much more carefully about different kinds of passive investing different betas if you will and be able to put together portfolios in a much more sophisticated way so the world of indexation is still very active and it's going to grow over time but the kind of indexes that come about are going to have to become much more personalized in the same way that we have precision medicine and personalized therapies i think we need to have precision indexes and personalized portfolios and we have the infrastructure today we have the ability to do that what we don't yet have is the algorithms the software the ability to model human behavior and detailer a particular portfolio strategy to your specific.

president twenty years
"andrew lo" Discussed on FT Alphachat

FT Alphachat

02:02 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Is that you get the power of diversification so that you would create a funds that's invests in maybe as many as a hundred difference attempts at finding different ways of curing or ameliorating cancer and if two or three or four of them work the fact that fifty or sixty of them turn out not to lead anywhere need matter they will still be a a return it's gets harnessing diversifications make it possible to finance cancer research is that is that a fair some asian of your idea is heading for for that for for cancer gearing it is and in fact it's an idea that is not new to the industry if you take a look at pharmaceutical companies they are portfolios of multiple projects the problem is that the portfolios may not be big enough in order to deal with the challenges that face us today we may need to have a large mega fund to be able to contain all of the various different projects in order to reduce the risk enough that investors are able to get a decent rate of return and patients are able to get the drugs that they need so being able to scale up the industry using these mega funds is an example of how financial engineering can actually help in ways that existing structures really don't allow is there a sense in which there's a perverse incentive in that cecil epa a number of the bigger existing large pharma companies are trying to cut back on our in the or find ways not to spend so much on on our deep because it's such a risky oppor operation that so much of a chance that you're going to invest billions and have nothing to show for it in financial terms that's right in fact i think that pharmaceuticals have been unfairly maligned because they are doing exactly what shareholders we investors are telling them to do we're telling them to make more money reduce the risk and be able to drive up their share prices and so by cutting rnd they're actually being able to do that because they're focusing.

cecil epa
"andrew lo" Discussed on FT Alphachat

FT Alphachat

02:16 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Okay this tumult topics i'm hoping to to cover the book is runs to almost six hundred pages and it's it's i don't think you should is with reading it's it is thoroughly enjoyable which is quite an achievement to more points i'd like to cover one is whether there is a way financed deservedly has a bad reputation at the moment people are angry with financiers is there a way that finance if we understand better the way that markets work can become more of a force for good for positively what absolutely i think that's very important point and one that i tried to make at the very end of the book to give some sense of what the future finances as well as the finance of the future we haven't been able to deal with these issues effectively because we don't recognize the fact that human nature does drive virtually every industry and that means that we're going to get some very positive things from industry and some very negative things because of human nature we all have within us the seeds of great things as well as terrible things and so once you recognize that fact the dual nature of human behavior we can then begin to think about how to channel the forces to be able to make better use of these tools for example the idea of financial engineering has become quite a negative connotation over the course of the last couple of decades but the fact is that financial engineering's responsible for some of the most important innovations in modern society and in some recent work that my colleagues and i have been doing we've been proposing the use financial engineering for dealing with some of society's biggest challenges like cancer are fusion energy or climate change using the power of financial markets we can actually get tremendous resources deployed to deal with these seidel challenges and i don't believe we can deal with those challenges in any other way so finance can be a force of good as well as a force of greed now it's just briefly discussed the curing cancer example still physi something we all of us care about s and you've written about it and been encouraging this in a long way as as i understand it the idea here is.

"andrew lo" Discussed on FT Alphachat

FT Alphachat

02:12 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Doing the things that we know we're going to want to do and that we should not do during these periods we need to have more of a recognition of the origin of that kind of behavior and design policies to be able to deal with it in a more effective way now does that mean choi i agreed the us constitution must be the most successful example of a system of checks and balances that has endured remarkably longtime over two centuries now the key thing about the us constitution it seems to me as a foreigner coming here is that it's very much principles based and written in very great generalizations so that can have flexibility over time if we take example of dulled frank for example well obviously the most relevant example for the for the us financial system at the moment there are principles motivating it's but it's extremely tightly prescriptive does that imply that that's the wrong way to go that's it doesn't allow for the kind of adaptation and eva lucien that you've been around a call set of principles that yesterday jesting well there's no doubt that that's part of it but i think that there's a bigger difference that we have to keep in mind and the difference is that the us constitution's brilliant and genius is not so much that it was principal space but rather that underlying those principles is a deep recognition of and skepticism of human behavior right we recognize that left to our own devices we will end up doing things that could be counterproductive even to our own health and welfare and so one way to do that is to create a system of checks and balances what we're we're constantly using the wisdom of crowds to help us make better decisions and doddfrank doesn't yet have that feature it has a list of prescriptions and some of them are very important and very productive but it doesn't recognize that the ultimate origins of financial crises is human behavior and we need to deal with that route issue before we can ever deal the ffective ly with these crises.

us frank choi eva lucien principal two centuries
"andrew lo" Discussed on FT Alphachat

FT Alphachat

02:10 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Right and similarly when things are going badly after a financial crisis we tend to impose lots and lots of regulations but that's probably the wrong time to be doing it and so this kind of feast or famine the cycle of regulatory stringency of versus being more relaxed that's also part of human nature and unless we recognize that feature of our behavior where never going to be able to break free from those cycles the hope is that with the adoptive markets ipod uscis as the driving force behind some of these regulatory innovations we might be able to begin to develop countercyclical regulatory policies to deal with these issues there's a very famous book about behavioral psychology cooled nudge i mean how do you affect that kind of change do you have you need to be countercyclical written as this as a big in big letters coming up on a regulators computer screen every ten minutes or something to remind them or how do you nudge things in the correct direction well so i i'm actually a big fan of that book right i find it fascinating but i want to remind the listeners that nudge is both the verb and a noun right and although i think the author of that book was using it as a verb to be able to nudge individuals to making the right decision very often as a noun we tend to dismiss people who are nudges simply because they really don't provide us with the kind of advice that we wanna hear i think that nudging is not enough i think we have to redesign the very structure of regulation to develop these kinds of systematic counterbalancing affects in the same way that here in the united states we have bicameral legislation we have three branches of government and we see that that's actually quite valuable to have this almost adversarial aspect of these types of organizations that can counterbalance some of our worst tendencies at the most trying times that's what regulation is really supposed to be about it's supposed to prevent us from.

united states ten minutes
"andrew lo" Discussed on FT Alphachat

FT Alphachat

01:58 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"That's right and imagine if we actually had access to these network diagrams in advance we could have managed these kinds of exposures much more deftly in order to be able to reduce the impact or maybe even a void some of these kinds of global crises okay so that's gives us some idea of how this can muddle the crisis and how we might have might have been able to full see some of it what therefore does that imply for regulation we've had sesame if you believe the verdict that markets have made so far there seems to be quite a genuine belief that the american here in the us have done quite a good job of reregulating after the crisis or at least if you look at what people are prepared to pay for us securities the market seems quite comfortable there are other less convinced about what's happened in europe what's does this suggest we should be doing about regulation and should we be comfortable with states of reregulation as we have it at the moment while i think the adaptive markets autho sis suggests that we wanna take a different approach to regulation and that is to recognize the human nature is a critical part not only of the regulated but also of the regulators it's a very complex dynamic that actual has a relatively simple illustration in the swinging of a pendulum with regard to tight versus loose regulations i think we all recognize that there are periods where regulation becomes relatively lax typically during market a boom 's and a bull markets when everything's going well regulators may be a little bit less likely to rein in these various different financial institutions and activities of course that's exactly when they ought to be raining in those activities but it's very difficult because human nature tells us that if everything's going well why bother trying to take the punch bowl away.

europe
"andrew lo" Discussed on FT Alphachat

FT Alphachat

01:59 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Of that okay so to try to make sure people can visualize this understand israel your mitee colleague bob merton has done a lot of work on this concept as well say you take one company like a big investment bank then look at how many different interrelationships or relationships it has with other companies other funds other entities in the mole at lines you have the mall oath paik the chart looks the more connected it is and the more worried you should be about the health of that company what you should more concerned you should be about the importance of that companies that get to the concept exactly if you think about the relationship between say particular sovereign debt greek debt for example and the various different institutions that hold greek that as part of their asset holdings imagine if greek debt defaults that sends shockwaves to all of the various different financial institutions that happen to invest in that asset so one way to capture that in some work that i've been doing with bob merton other coauthors we construct a network diagram which looks like a spiderweb where connections between certain kinds of securities and investors will give you a sense of just what happens when a particular security fails that spiderweb can easily turn into something that looks more like a ball of yarn a much more densely specified graph when you've got deeper connections and crisis conditions and what happened during the financial crisis as well as during the european debt crisis is that these connections allowed relatively small shocks to propagate much more quickly and developed a much much larger financial crises arguably in the case of europe now that quite a lot of the begin stations of managed to disengage from greece to a large extent that's why people are less less worried than they were a greek disaster at this point would be less disastrous for the rest of europe than it would have been in twenty ten.

israel bob merton greece europe
"andrew lo" Discussed on FT Alphachat

FT Alphachat

02:03 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Voting behavior can actually affect markets in a very dramatic fashion i completely agree with you on that one i can remember people talking to me in a seven the head of the ahead of the crisis saying that what you've got to worry about is that this is like the seren getty and everybody knows that if you if you won't be the wildebeest doesn't get chomped by the federal the lion you need to be in the middle of the hood and that's actually what's most dangerous you also look very intriguingly at network feary it's connectedness which is another a another idea that a lot of only discovered for the first time once we discovered it mattered in a seven and eight could you explain how connectedness matters and how that's affected the adaptive markets by puff assess that's a very important concept and the concept really emerges as a contrast to what is currently being done now in areas like risk management for financial investments okay so typically when we think about risk management and investor behavior we use a very statistical approach that that returns of a particular investment aren't going to be dictated by the statistical distribution but in fact what happens in financial markets is that investors are going to be reacting to each other's behavior and so the more tightly connected a market is the more investors are tied to each other's fates the more likely it is that small perturbations in market values can spread like a cascades or a virus to the entire market so measuring connectedness measuring how one institution or individual is tied to the fate of another institution or individual can give us insights into how easily these kinds of shocks can propagate and how they can be easily magnified in very very quickly if we have the wrong event happening at the wrong time and ultimately the financial crisis is the perfect illustration.

seren getty
"andrew lo" Discussed on FT Alphachat

FT Alphachat

02:21 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Okay and it's also why economists multiples ten twit quite as well as physicists models student i suppose exactly now let's try to do some of this ecological research in real time or at least gets an example of it baps the biggest phenomenon of the moments in the financial ecosystem is the rise of beta of beak passive fund management and the decline of the traditional model of active management long mutual funds with about one hundred stocks in them how would you go about mulling let's with the adaptive model how how would you predict this obviously very dramatic shift over the last ten years towards passive well that's a great example last year i published an article titled what is an index and that article was exactly geared towards trying to understand the shift in passive investing as we have more and more assets in that space so we have to begin first with the recognition that john bogle has really transformed the financial industry founder of anga the founder of vanguard has really transformed the industry part of the reason he did so was not because of fishing markets but because of something that mr will calls the h h stands for the call cost matters hypothesis and i find this an incredibly compelling idea the fact is that using passive investment vehicles like vanguard mutual funds investors can save a significant amount of money in terms of the fees that are usually charged and that matters over time so what's happened over the last several decades since vanguard started providing their products and services is that a large number of investors have flocked to these index funds and that's a good thing in general but what it does do is to create systematic risk another words if it turns out that a fund loses money that means that you and i both experienced losses if we're invested in the same fund the fact that so many people now are indexing means that we are now all tied together to the same outcomes and when we start getting those outcomes in a synchronous fashion we will start hurting in that very same way and her.

john bogle founder ten years
"andrew lo" Discussed on FT Alphachat

FT Alphachat

01:44 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Precision as we do now and efficient markets the only issue is that this is early days for the adoptive markets pompous and so we don't yet have the corpus of research that we have for officiant markets which has been around for for many decades in particular what the adoptive markets ipod this tells us to do is to actually collect different kinds of data from what we're doing now and analyze them differently in particular we have to think about financial markets more as an ecosystem rather than a mechanistic kind of system and what that means is that we have to start collecting information about the ecosystem the same way that an ecologist or volition airy biologist would we have to ask what the key species are in the ecosystem what their biomass czar how they compete how they survive and adapt all of the various different aspects of the flora and fauna of financial markets have to be measured quantified and analyzed we don't do that right now we have very different view financial markets that's really driven by this physicists perspective but in fact we don't have a physical model we really have a biological one that you have anything against all your friends in the physics department's mit but that may not be the relevant science exactly physics is a much more simple approach to modeling the world than biology because the underlying phenomenon are that much simpler you know richard feynman than great physicists said at best one year at caltech graduation in the midst of a stock market crash feinman said to his students imagine how much more complicated physics would be if electrons had feelings and i think that really captures the difference between physics and biology.

richard feynman feinman caltech one year
"andrew lo" Discussed on FT Alphachat

FT Alphachat

02:02 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Going and reacting almost in a biological perspective not a physical or mathematical one so when you have these kind of crises typically what happens is that investors react very strongly by pulling assets from risky securities and putting them into safe assets flight to quality or as i call it freaking out when investors freak out they reduced the expected return on risky assets and increase the expected return on safe assets because they're selling the risky assets causing their prices to go down and buying the safe assets causing the prices to go up when you look at the data during those periods what you see is that investors are not getting rewarded for taking risk they're getting punished for taking risk so the whole idea of a risk reward tradeoff which is central to modern finance is turned on its head during these periods of crises and i think that's the part that really requires a different narrative a different explanation and i believe that if we had the market policies at our disposal we could have treated those periods very differently from a policy perspective now one of the appeals of deficient markets i puff asus i say it's toll tin business school and so on is that's it does allow you to come up with very precise onces in numerical terms to very precise questions the you can be as high would you diversify away all the risk of such and such a bond and they will actually be an answer to that question is moving to the adaptive markets i puff asus accepting that you cannot have quite the degree of precision that's be em h the efficient hypotheses office or is it giving us an equivalent powerful but more complicated model while i would argue that it is a more complicated model but eventually we actually can have the same level.

asus
"andrew lo" Discussed on FT Alphachat

FT Alphachat

02:03 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Information the second is that the fully reflect part that's the part that we have to keep in mind when we think about how humans actually make decisions and incorporate those kinds of pieces of information into prices it's human behavior that really causes us these kinds of issues that psychologists have been documenting once we understand how human behavior works we see that sometimes prices don't reflect all available information or if they do they reflect more than just information they also reflect emotion and that makes the theories much more complex okay let's try looking at some worked examples from a mutually shared history you've written a lot about sports is now almost forgotten instant ninety seven ninety eight the asian crisis and then lt see and then very scarily the quanta quake of two thousand and seven when a bunch of very interesting quantitative hedge funds suddenly lost inordinate amounts of money in what appears to be a clear blue sky and then obviously we have the crisis of two thousand and eight how did each of those incidents help you develop this theory and is there any way this theory might have helped us at least mitigates what's happened in those those very scary events well in fact it was exactly those events that really formulated my thinking about the adaptive markets hypothesis and got me to start down this path try to understand how markets really work so the basic idea behind crisis is that investors are reacting and they're not reacting rationally necessarily they're reacting emotionally in fact if you look at the hedge fund industry i call that the galapagos islands of the financial industry because you can see of lucien happening before your very eyes you can see species coming.

lucien
"andrew lo" Discussed on FT Alphachat

FT Alphachat

02:02 min | 3 years ago

"andrew lo" Discussed on FT Alphachat

"Come to know it is it's to do with the theory of the human brain is profound as that or are we talking more about looking at the kind of flaws in decision making that the behavioral financials led by common and i have discovered over the last few decades well i think it is as profound as the origins of the human brain and how it works and how it differs from other species so the idea behind the adaptive martin's hypothesis is that the efficient markets hypothesis is not wrong it's just incomplete it's only part of the picture and while the behavioral economists and psychologists have documented departures from rationality what we really need is a theory that encompasses those kinds of behaviors and so by understanding how humans actually behave through their brains and through evolution over time we can actually begin to develop this broader theory and that's what the adaptive markets ipod asus is okay so you borrowing lot from dalwin ian or lots of insight from darwinian theories of evolution and eva lucien re biology exactly now in the case of the efficient markets i puff assistive diversion of rationality there is is such that all the information about a securities already at all times reflected in share price maybe with some instantaneous period foil adaptability how does that change once you bring in your different notion of rationality while to begin with pharma's genius in formulating the efficient markets ipod asus and samuel talking about gene farmer of of the investing chicago yeah that's right jeanne foams genius in formulating the efficient markets hypothesis along with paul samuelson is that prices fully reflect all available information now that statement contains two parts the first is that prices fully reflect the failing.

martin samuel gene farmer paul samuelson dalwin ian chicago
"andrew lo" Discussed on WSB-AM

WSB-AM

02:35 min | 3 years ago

"andrew lo" Discussed on WSB-AM

"While may even at the fateh habitable miles and i'm going to be here begg of joining us on our mccain show with gone into rapid by in the next segment but we got ta give roger the first pre rapid fire minute hello roger you've got the very rapid fire men of what's on your man or woman carrying thanks for taking my call yes sir i my wife and i were at the beef from radio require my roommate yes sir i just wanted to give you a little great uh we got there we were there from about schick go seven birdies so we were nodded and the crew foot all hard break at guide you but it was very crowded we talked to the manager little bit after having heard not only on our morning show return on your show also where your day and and we will see either dan no ordered andrew lo and behold bill berry surprise down with great and that was talking about you know what was happening and he made the comment you launch was four times the normal crowd while for those who may not know hold on a second the restaurant beef old brady's the manager the owner of the restaurant this added that he was not going to show thursday night football on his tv says like he normally does because of uh some players disrespecting the flag and the national anthem and so roger you and your wife went to that restaurant and that it ends fat and there was crowded when you were there but it also gap packed out last night after you left yes exactly while we were there one agree ah europe resembling then from the district rick or out there took the microphone made a few comments about real or what was going on the owner floundered in in turn made some comments about israel and introduce a family rift with their support that yesterday evening method loss in afghanistan well that's great but thanks a lot roger for the update you see folks your bosses do matter you can vote with your pocket books you can vote with your money and you can vote which voices youth vote with beat you're listening to the herman cane show.

roger schick brady football rick israel mccain dan andrew lo afghanistan