 It brings me great pleasure to introduce our final speaker. He's from Aarhus BSS, Aarhus University. Not showing any favoritism whatsoever Mr. Karsten Wozinski is here to talk to us today about breaking the walls of mortality Forecasting. Big round of applause for Karsten. Thank you. I'm ready. Yes, so the last century has seen huge and Unprecedented improvements in life expectancy all over the developed world Here among others Denmark, which has seen it and improved from around 50 years to 80 years so this improvement obviously has huge impacts on a wide range of In societies, yeah, so the life expectancy is basically just an intuitive way to Describe all the age-specific death rates so In general, it's very important to have accurate forecasts of the modality and life expectancy for both governments and pension funds as As this affects how we decide the pension age or how much should be saved and so on However, there's a general problem in this field that we have two competing methodologies on how to predict the future mortality or life expectancy and this obviously in leads to an to a Undesired uncertainty so What we have done is that we create a unifying framework in mortality forecasting where we this is made possible by using some techniques that is commonly used inside in economics or econometrics so historically historically the two Different approaches is to either use demographic theories about how the mortality Develops with respect to age That is one approach. The other approach is to use statistical factor modeling general Which do not use any theoretical background So what we have done is that we reformulate the demographic theories using these econometric tools To into a statistical factor model framework as well So more than create a unifying framework. We also are able to improve upon resisting Upon existing models and that is the main The main thing here is that we are able to identify relations between the age specific death rates between each age That are stable over time So when we identify things that are stable over time, we can use this in predicting the future mortality or life expectancy and we find that this creates and a significant improvement on Existing methods and this is especially the case for longer horizons Forecast which is also what the governments and pension funds are most concerned with in terms of reserving and changing the life expectancy the pension age Yes Thank you Carson. Yes The jury questions may need a microphone for that Thank you just Just to go back. What is the big problem? with the inaccuracy in in in the forecasting life expenses so Early on we didn't really forecast the life expectancy improvement. So we just took what was Given inside pension funds. So that has resulted in a number of bankruptcies among pension funds and We we decide the government as well. They decide how much they should Use in in various medicine on medicine and all kind of stuff But we also have to decide the retirement age and of course that has a huge impact whether we live longer or not So this is also you also seen the increases in retirement age around Europe But they have come like with a delay because people didn't Didn't anticipate it Well enough so it's very important to have accurate forecasts to to like a word Bankruptcies defaults Yes and I Imagine there are other models for forecasting like you can forecast the weather and things like that But what's the what's specific about your model of forecasting? So in general there has so you know the demographic theoretical approach they didn't have any they they didn't they were They were like desired very much because you had theories about how the mortality should behave with respect to age But it didn't predict. Well, it was very bad at predicting. But what we have done by reformulated into a statistical factor model. We are able to Produce a model that actually predicts better than existing model and we have and you find this relationship that is stable over time It's called a co-integrating relation very statistical term that is stable over time and when we get something that is stable over time Then we can use this stability in forecasting. So this would also be the case in like 10 years 20 years when we forecast But and this this has not been like incorporated in usual models That's a main advantage So Pete Hine once said that it's difficult to do forecasts particularly when they concern the future, right? Yeah, so I know So, how will you know if you're how will you know that your model is better than the existing models before it's too late and another? so we have so we have used a lot of the Procedures that you can use to test the forecast predict the predictive ability of a model which is like Some complicated statistical mechanics, but it's really basically uses how good I read predicting data So we have data actually for a long time from 80s eight from the 18th century on mortality and in a lot of countries and then we can use this to recursively Estimate and forecast and then see how good we perform and we significantly outperform existing model in general and the problem is with this Yeah, but you also see that in the latest periods It's better, but we also the idea by finding a stable relation. So this is the relation is stable over time and We are quite certain that that would be the case as well in the future. So we actually have some information where as the existing models do not have any A requirement of stable relation that should hold so that is just basically It's called a random work, but it's just some unstable relation. They forecast and it's naturally More uncertain if you forecast something unstable than if you forecast things something stable. So that's there Can you elaborate a bit on how your research can? Benefit the government or the public authorities so the government they decide So the welfare commission they decide they they were published a report a couple of years ago about how the I've expected she would increase throughout the 100 years ahead. So you use this and Then from this they decided The retirement age how it should increase and all that stuff, but but we have already seen that that model Has done terribly wrong In the last 10 years because it was an unstable relation they use so We can use this to like better set the retirement age and increase it in retirement age But you could also use it when you want to see how many people Will live for a certain age because you know that at some point they will The mid they expenditure to medical the medical expenditures and stuff like that will increase Enormously at some age Final question from the jury. Yes, this stable factor that you found that sort of makes your model unique. Yeah Where does it come from? Is it can you put some words on what it is? So usually they just so we have a age specific this rate from zero to hundred and ten and then you just use a general statistic reversal and and then Find some trend in this data and then use predict this but what we do we? we divide the The ages up to infant mortality and then there's this called accident hump, which is around the late teens where Especially men die from accident hump women die from from maternal death when they give birth and then they have as But this this has been falling especially the last 50 years and then you see you have an increasing Mortality with respect to age after that expect that's due to senescence. So what we see so we create we Go it's very technical. So we go in and put these factors and then that an effector effector that affects all ages as well and then when we do This we have we can extract some trends for each of these Models instead of just a general trend and then between these We're able to find a Relation between these trends that is always Stable and that's the difference. So we all we don't we won't we know that the changes in those goes at a specific rate Which is kind of technical and I can't go further into details, but Okay, great. We have time for a quick question from the audience No, if you were born today What's the life expectancy for someone just out of curiosity? That was the last Known data from we don't have data for two thousand right in the age of retirement is Today is in Denmark is 67 but you but this is okay It's often confusing. This is just the period life expectancy that's based on all the mortality rates based today. So this Doesn't incorporate future improvements in life expectancy, right? So because they are uncertain Right, so there's a lot of variables that you can't really account for so you so you could also Create the true some would say life expectancy But that would be uncertain because you will have to predict the future mortality rates and death rates are great Thank You Carsten big round of applause for Carson everyone Okay, ladies and gentlemen one minute to score Carsten