Akhil Khunger, VP Quantitative Analytics, Barclays, has more than ten years of experience in the field. Akhil develops statistical models to forecast balance sheets, revenue and probability of default. He has experience with time series modeling, statistical modeling, machine learning and building implementation frameworks using Python and other programming languages. Akhil, who has a master’s in Financial Engineering from UC Berkeley and London School of Economics, has worked in CCAR and Bank of England stress testing and know the Basel III/IV framework, SR 11/7, SR 15-18, and SR 15-19. He also has experience managing junior model developers.
In this episode:
- The worst year for modeling?
- Diving into models
- Working with big datasets
- Getting to the real reasons behind trends and data
- The power of stress testing
- What human judgement can bring to the table
- The AI component in forecastingÂ
Full blog post and transcript
This is FP&A today. Welcome to fp NA Today, I’m your host, Glenn Hopper. On this week’s show, we’re stepping just outside the usual walls of corporate finance. I wanna bring you a perspective that I think every fp and a professional should hear. My guest is Akil Khunger, a quant and modeling specialist who spent his career building risk and forecasting models for global banks like Barclays, Citi, and HSBC. And here’s why I think Akil is an interesting guest to have on the show. We’re in this era where complex modeling is growing ever more accessible, and the lines between quant finance and corporate planning are kind of starting to blur. Tools are evolving, expectations are shifting. And if you’re an fp and a today, understanding how quants think about things like risk, uncertainty and modeling could be a real edge. So in this episode, we’re gonna dig into what fp and a teams can borrow from the quant world, how external factors like macro trends and tariffs can be modeled more rigorously and what the future might look like as AI and analytics reshape both fields. So let’s get into it. Akil, welcome to the show.
Akil Khunger:
Oh, thanks. Thanks, Glen. Like, uh, good to be here.
Glenn Hopper:
Yeah, I think your, your career journey is quite unique and I, rather than me trying to summarize it, can you kinda walk us through your path from engineering to financial mathematics and then financial engineering and all the way into your current role at Barclays?
Akil Khunger:
Oh yeah, sure. I think like as an undergrad, I was like, didn’t fully think what I was supposed to do. I get it into electrical engineering, which was quite exciting. And then I had quite a bit of interest in finance and I thought of something, doing something which is like merging my mathematical skills, quant skills and something with finance. And I decided to go for financial mathematics, which was quite like a technical course, uh, more into very mathematical details about all the things we might be using, but things behind that, a lot of the models which we do. So then I ventured into some trading, like I did work on a trading company for one a year or so, and then I thought I want to get more practical knowledge of like how to use things and not just have like more theoretical or like knowledge on like how to build things.
So that’s how I decided to come to us and also us as a bigger market and you see more opportunities. So I came to financial engineering and where I like met a lot of good inter P people and then also learned a lot of practical things on how to use things for modeling for variety of purposes. Like could be from risk to front office or trading to research. And then like ended up getting into risk modeling and through my statistical skills and uh, mathematical skills. And I learned more of coding skills on the job, which I also learned through a master’s. And I have got like, have seen a lot of things evolving over the years, like in different companies depending on like lot of regulatory requirements and lot of modeling requirements have been changing. And then new tools come into play like from initially even like change the programming languages from MATLAB to Python. Uh, and then now within that, like you’re still using Python or something, but you’re moving into AI stuff. Five, six years ago it just started as a bigger play and now it’s gonna be much more with AI coming into play as well.
Glenn Hopper:
And I’m guessing even back in undergrad as an electrical engineering major, I’m guessing coding was a big part of that. Were you doing CS as well in in undergrad?
Akil Khunger:
Yeah. Yeah. So there were like a few courses which we had to do out of coding, but uh, so we did a lot of c plus plus C and then that was like specific coding courses, which are computer science focused. But then as part of electrical engineering, I had to do a lot of MATLAB stuff for image processing and everything, which is quite interesting. And I think when I joined financial engineering or financial mathematics, I think MATLAB was a language at time, that time, which was used a lot and which has evolved now to Python. And our,
Glenn Hopper:
For our listeners who aren’t familiar with matlab, can you explain what that is?
Akil Khunger:
Yeah, so it is like, like a type of language similar to Python, like it’s that in a way that you don’t need to write. Like it’s not like c plus plus or C where you need to write everything from scratch, like define all the variables and thing. And it is like interpreted like a language, but it has more features in a way that uh, uh, like it is more secure than Python or R because it’s not open source. There is a company like Math Works, I think, or some like which control MATLAB and any changes are done in a very specific manner. Like, so there’s less risk of like losing control, which is in Python. But having said that, the problem is the cost because there’s somebody controlling it, MATLAB come with a cost and then advances are smaller because, uh, people are not able to free contribute to that language. Though still it was, I think because the cost would be the starting point, I think when people start migrating from that thinking that we can get a free language and then over time like, uh, you get comfortable with language, my lab is still pretty good language. I think when it actually goes into core electronics principles, like for image processing and everything, it’s still one of the very good languages.
Glenn Hopper:
So you had this background in coding before you got into finance, so I feel like that brings a different mindset of the way you look at at finance. And also you, you, you have this engineer’s background and then you go into financial, math, mathematics. So you’re doing some, some pretty intense math there and that’s your first exposure to finance. While you were studying that, were you also taking your, your just core basic finance courses so you’d have the, a better understanding of the rules around finance and what, you know, what everybody was looking at and what was important?
Akil Khunger:
Yeah, yeah. So I think the courses were like mixed of mathematics and finance. Uh, to be honest, the mathematics courses were more tougher, took me more time. I, i I was not expecting I was really going to be easy for me, but it like was quite intensive, uh, to just, it was completely different thing. Uh, but then finance was actually, it’s more intuitive. I mean, it is like you need to understand stuff, but if you spend enough time like on financial analysis, like something on accounting or something on risk management, uh, some measures like something on option pricing, like not pricing, I would not say like in terms of quant, but like understanding what options are, uh, and as a basic things, understanding what, how the balance sheet is managed. Those things are like conceptual. And I think if it’s just that if you spend enough time on them, I mean, you have to spend that enough time on them to understand them properly. But there’s like, uh, in some of the quant stuff sometimes can get like, uh, uh, maybe still, I don’t understand 20% of the stuff we we have. Right. But because it’s, and there’s always new things coming up. Yeah.
Glenn Hopper:
And I wanna get, we’ll, we’ll in a few minutes we’ll get more into the difference between the financial mathematics and the financial engineering, but I think as we’re talking about the education here, it’s easy to see the skillset that you built. But for our listeners, I’m trying to help them visualize the role that you’re in today. So maybe if you could just tell us about what your role is at Barclays, what, what a day looks like for you and and what your main focus is today.
Akil Khunger:
Yeah, so I’m into quantitative analytics team. I work as a vice president there. And uh, so my main goal is to develop like stress testing models and financial planning models for different businesses, like focusing more on markets and investment banking. But I do focus some on cards portfolio and trying to get hedging costs of the card business as well. Our variety team covers a lot of other things like treasury, uh, cars and everything. So then like in terms of, uh, stress testing or business planning, like it’s not just for us business like we cover UK or like parlays is like, uh, UK bank, like, so I would say all the world’s model like global models for UK portfolio. And then also there’s a lot of push from Euro, Euro <inaudible>, like after Brexit that Euro regulat are doing things differently. So they want some different testing, different business planning and different scenarios.
They have their own scenarios. They want you to test things, develop different models for them. So those are like the things we cover on like long and as part of the whole process, we have to do a lot of coding apart from the model development where we have to implement the models and make them ready like for it, uh, usage. So it is quite like sometimes I do think like we are also doing a lot of more than I expected when I joined Barclays, like in terms of implementation work. But I think it’s good for learning that you learn a lot of new things. So you’re not just doing development, but you’re learning it core it implementation skills as well.
Glenn Hopper:
Yeah. I’m wondering from your perspective, corporate CFOs and and finance leaders right now are, are kind of losing their minds around all the uncertainty around geopolitical issues, tariffs, everything going on. I, I feel like in your position, yes, there’s this uncertainty, but I feel like maybe you’re digging into even more uncertainty than a typical, uh, corporate finance department is. So I guess with that, my question is, does this feel like a more stressful year to you or is it always something and there’s so many exogenous factors going on and everything that you’re always modeling different scenarios. So how does the environment feel this year for what you’re doing? I I, it seems like ev like everyone in fp and a, our models are all over the place. We keep doing new ones based on whatever’s going on today.
Akil Khunger:
Yeah, I think the total, this very challenging time right now. Uh uh, but having said that, I think because we tend to use like models that are built on like longer history and try to cover a lot of different regimes, so they tend to perform well. And as of now for this year, like we have not like largely seen too many differences. I mean, there are some things which keep happening and need to adjust your model, but I have not seen like some very big changes. And the one big reason for that is that the implementation stuff I was talking about that we have automated so much stuff like make it that our models can be run through different scenarios very quickly. So for example, like in April when the <inaudible> first time came out like, and he wanted to test out our model, different, different scenarios or something, we were able to do that pretty quickly in one week or so.
And not just, I mean our model, I would say like across <inaudible>, we were able to test out whole models in a single go to see what the impact would be for capital RWA income balance sheet, everything. I think what prepared us well for this thing is the COVID period because something similar happened to COVID and I think it’s, I mean I was not in Barclays that time, but I think similar thing we did HSBC at that time at probably Park River that time, like regulators forced us to try different things and then people read okay, that you have to be smart enough to update your models or make changes on the fly quickly in a govern well governed way or, and you should be able to run different scenarios quickly so that you’re not scrambling for a lot of governance works and at the last minute. So I think that time it was much more difficult, uh, and also like the situation was even much more difficult and right now, I would say at that type.
Glenn Hopper:
Yeah. And I do, I wanna get a more into scenario analysis later, but when I was looking at your background, I was trying to visualize the, the difference and understand the, the intricacies between the two. And I think because of the typical, our typical listener and that the path we take, it’s um, can you help us visualize, since you studied financial mathematics and financial engineering, can you walk us through the differences and, and then maybe tie ’em together and talk about how you’re applying both of ’em today?
Akil Khunger:
Yeah, sure. So financial mathematics is more like, uh, like a research oriented subject, like kind of any models you are building using right now. Like, or like you come up with a machine learning new network model or something and then apply to finance and then you come up with some, like first <inaudible>, you use some black and schools model or something. So that is more like you understanding the whole, uh, math behind it and try to come up with strategies by changing those founding principles, like coming up with some new models from scratch. And you’re, so basically you are spending a lot of time on research. Financial engineering has 25% of that, uh, that you do need to understand the basics well. But then the more effort is on like trying to use those skills in a smart way. What are the best way to use those tools?
Like more emphasis on finance, as you were saying that going through finance courses. I think if I would split the finance courses, like maybe I had like 20% finance courses in finance mathematics, but 60% finance courses in financial engineering. So you get to see where you want to apply those principles and understand more about the real finance world. There you can actually use those. And I think that’s why I think both of them get used, uh, when we are financial engineering is very easy to see. Like as you are go through a lot of discussing with different businesses, we talk, we discuss about model businesses, we have to understand the product and have to understand the simpler knowledge about finance principles, like to even showcase at what we’re saying, like logically makes sense. Then I think those skills are like, if you’re trying something new, if you want to drive it gives us more con, gives me more confidence. Okay. Like I know some mold maps behind that. Uh, I I can, I can defend like that even to someone who’s very technical,
Glenn Hopper:
Just picturing the complexity of these models. And even I think today FBNA models and, and we have helped with, with software and, and certainly different broader skill sets that are letting us do a lot more complex modeling than, uh, than we have historically. But with all that, I’m wondering, did you ever consider just, uh, traditional corporate finance or fp and a or were you from the beginning you knew you wanted to go the quant route and have your focus there?
Akil Khunger:
Yeah, I think I would say in the beginning I was more focused on quant because I really wanted to use my skillset. And then I learned, I think in recent times I have sometimes do think that maybe I should try something different. Like, because I now know more about like, look working like about the businesses and understanding like products well. So at times I do think that maybe I would do that like in some years, I don’t know, like I’ve not really looked at it very seriously, but that comes to mind like you want to challenge yourself, do things differently, see what challenges like other people may face. And I think maybe use your current skills in a way that improve the other process if you go other side.
Glenn Hopper:
Okay. Let’s dive into the models now because I think that’s what I, I think this is gonna be fascinating to, to walk through because I think on fp and a side, we know pretty much how we’re building our models and could everything from everything happens in Excel to using planning software and all that to start it. But with the stress testing and the scenario models you’re doing, I’d love to hear maybe your approach to modeling and we, and we can talk, uh, about kind of the fundamental differences in how quant teams and fp and a teams approach forecasting. So walk me through, on the quant side, you have a a, a new model you’re gonna build for whichever business, what you’re looking at. And maybe even, I’m guessing that one big difference is because of the nature of what you’re doing, are you waiting historical performance and all the external factors? ’cause you don’t know what the annual plan is gonna be for the business necessarily. Right? Or you don’t know what their goals are or all the internal stuff. So I don’t know, I’m, I’m trying to answer the question for you now <laugh>, I think, but maybe, maybe I’ll stop talking and and turn it over to you and hear, um, hear your approach to modeling.
Akil Khunger:
Yeah, I think you put it quite well that we are like starting with a lot of historical data and I think we want a lot of data like to see different regimes or different periods and if it is not possible internal data, we try to look at external data, some industry data at times and try to model using that and then try to apply some linkages through that is where I think business intuition comes in. Like we, we could work closely with the businesses like well finalizing like how to make that linkages. So while developing the model, like we can never develop like a future proof model, right? Like if strategy changes or something. So that is where one thing will fundamentally differ from FPNS side and our, because we are not like, as you say, we don’t know the annual plan. I mean sometimes we do know, but then also it’s difficult to incorporate that.
So what sometimes we do try to like things like which are more stable, like balance sheet or something. We try to model them in a way that your current value has an impact on the future value. Uh, and I think that should be a way for, so that any recent strategy changes can be reflected in your mo in your forecast anyways. Like, so if you’re picking from a starting point and you made some changes in the business, if you’re starting from a new point, you’re already taking that look out and sometimes you need to look at recent averages as a starting point or something. But still picking back on the historical data for the model design, uh, it is just at the starting point, you may want to use it sometime. It is not possible like the way the data is and you may not be able to do that.
I think what FP and a team are more focused on are very actuals and plan and they can optimistic at time, uh, that it’s also like they have been told, okay, to stick to this plan. Like treasury will not give you funding if you don’t stick to the plan. So they can’t, they don’t want to say, okay, that I will, I will show my lesser revenue or less balance sheet that uh, uh, I will not get that funding. So there’s sometimes forced into making some decisions, like based on the plans. And that is where I think, uh, there is like more and more need of like, uh, using kind of mix of both world, I think which was done like quite interestingly at Citi. Like they were very focused upon using, definitely using like our models for business planning directly in a very formalized way. And any changes would have to go through like an overlay or change process, uh, from the FPA teams. Uh, mean it is done everywhere. It <inaudible> also, but it’s not that formalized process. Uh, when you, especially when you’re looking for business planning because
Glenn Hopper:
Fp and a today is brought to you by data rails, the world’s number one fp and a solution data rails is the artificial intelligence powered financial planning and analysis platform built for Excel users. That’s right, you can stay in Excel, but instead of facing hell for every budget month end close or forecast, you can enjoy a paradise of data consolidation, advanced visualization reporting and AI capabilities, plus game changing insights, giving you instant answers and your story created in seconds. Find out why more than a thousand finance teams use data rails to uncover their company’s real story. Don’t replace Excel, embrace Excel, learn more@datarails.com. I’m, I’m wondering about the tools that you use. Are you starting in Excel just because you can visualize easier in Excel or are you just immediately, everything’s in a data frame, you’re in MATLAB or Python or whatever for doing this? What are the tools you’re using when you’re building out models?
Akil Khunger:
I think it depends like, uh, business to business and sometimes what data we are looking at. Uh, I think sometimes you do need to start with Python or like data frame because the data is so big that you need to do some analysis and uh, insert it into like a smaller version and then maybe yes, you will open that in Excel, uh, or even like plot some charts in Python to see where you are right before making any modeling decision. Sometimes you may want to create Excel, create some data and then send it to your FPNA partners like, because so that they can visualize. Uh, I think that’s another thing we do try to have all our results like in like we do want to put some like numbers in Excel so that everybody can open it. Like it’s not that who don’t have knowledge about coding or something, they can’t open it.
Glenn Hopper:
Yeah, because you have such big data sets, are you doing things like looking for correlations between different categories, whether it’s in the internal data or the larger macroeconomic? Are you kind of trying to find signal in or leading indicators and anything? Are you, how much time are you spending trying to break down and and and figure out what’s driving trends? If there’s, if there’s something you hadn’t looked at or is it, is it pretty standard, you know, these kind of correlations are gonna match in this industry? What what’s your approach to that?
Akil Khunger:
I think yeah, whenever we are starting some business from scratch, like some new requirement coming, then you do need to spend a lot of time on finding which even like, which not just which correlation, but like whether if it make logical sense, right? You may see a very good correlation with some variable, but that by just be like by chance. So those things like that is where we need to talk with people a lot or like directly business a lot like to understand, I think a lot of time even goes into data cleaning. I think before any of this can be done, like to make the data reuse usable in the right format. Like, uh, removing outliers or dealing with any missing data or any strategy changes. Like if something happened in 2019 for example, like you will not know and you will say, okay, you’ll try to fit them variable. You say, okay, it’s very good relation, but that’s just that you decided to wind up some business or try to grow some business. Like, uh, so those are things you need to get information from product control teams or FPA teams and incorporate them into your modeling because I think blindly if you put some correlation, you will get something, but that may not make sense. Yeah,
Glenn Hopper:
Yeah. It’s funny, you don’t want to p hack your way through the data and come up with a, you know, assume causation because you know the spray is correlations or, or whatever. So yeah, totally get that. To your point on COVID or, or any of the other global financial crisis or any of the other stuff that’s happened, it’s, you can’t, you can’t model the black swan events that are gonna come up. So since you do work with fp and a teams, you know, in fp and a, we are typical, I’m thinking now of not when we’re redoing quarterly forecasts or whatever, but I’m thinking back on the annual plan. I mean, we have our historicals and our trends that come from that, and we’ll have high level driver based models that are, okay, we’re gonna hire this number of salespeople that should correspond three months later or whatever it is to this number of, um, sales and, and they’ve got all those models.
And the most important sheet becomes that, you know, you’ve got your assumptions and, and then your driver’s tab where you’re just putting all this in and sometimes during the, while you’re making the annual plan, you know what the target is for revenue, the target is for EBITDA or whatever. So you can sort of tweak the, uh, tweak the plan to say, well, this is what we’re gonna do. And I mean, and it, it does make sense from a strategic perspective. If you want to hit this revenue number, you have to bring the salespeople on three months earlier, let’s see what that does to the model or, or whatever. Are there some techniques that maybe you’re using in quant that a typical fp and a department is that, and, and we’ll, we’ll talk more about technology and the availability of these tools later, but are there some techniques that you guys do that maybe internal fp and a teams could, could consider and and use in, in their modeling?
Akil Khunger:
Yeah, I think like, like especially like not just pivoting two very recent data, right? Like just to look at the last two or three months or last two or three quarters. Like I see like finance team, like they have a very tendency to stick to the most peaceful points and like try to model everything based on that, which is fine, I understand. But there could be things like seasonality or something, which can be very important in corporate as well. Not just like micro drivers understanding like why something happened in the past is also important. Like not just assume it was like just a strategic change. So sometimes, like you may say, okay, there was strategy change to grow the business, but why did they do at that specific time only maybe the economy was such, maybe like the whole situation was such that it actually made sense to grow that business so you could actually get a good macro driver, like which is needed.
Like, and it could be both ways, as I said, like there’s a other way, as I mentioned the previous answer, that you have to treat that with caution. It could be something serious correlation, but it may be realistic. So you need to discuss with the management and see like business managers or like, uh, uh, like is that, like what was the reason that why you prioritize this business at that specific time? Is there like anything economy driven or something? So those are the things you need to probably consider and try to use like, uh, trends or something, right? Simple can be done in Excel, like try to see trends in something. So like visualize visualization is always helpful. I’m pretty sure they’re already doing that. Uh, but maybe they’re not doing that for long enough in history.
Glenn Hopper:
Gotcha. Yeah. If things are headed down or things are headed up, but they’re predictable, it’s a lot easier to model. And when we don’t know where things are headed, everyone I talked to in the industry, we’re running more scenario analysis thi this year, halfway through the year beyond, halfway through the year at this point than we ever have. And it’s still so much uncertainty. Um, so I think we’re getting exposed to scenario analysis and stress testing more than we ever have. And for us, that usually means going back to the driver’s page and changing the numbers and saying, well, what happens if, you know, tariffs go up and it impact, we, you know, are paying 15% more or 40% more or whatever the number is for a product. But from your world, stress testing I know is very important. So could you walk through what stress testing is, uh, when in your models and maybe there’s something that fp and a teams could, could take from that.
Akil Khunger:
Yeah, it’s about like to simple terms, I think like how, how your business will react to extreme movements and things like spreads or volatility, market volatility or unemployment rate. And then there are a lot of other drivers, like specific drivers, but there’s also the point to it that, uh, how maybe you think you react to this bus, these factors in a certain way, but reacting to like a 0.5% unemployment rate may be very different now than if it was at 6%, 7% rate. So you have to understand those levels, like not just in a one way, okay, like if it goes by 0.5% up 1% w will do this. Maybe the impact will be much different if it happened from other level, right? Like if it interest rates drop from 6% to 4% today, it could be very different from dropping from 3% to 1%. So that is the thing need to be very careful like in across all variable. I just give interest rate is a very good example here. Like, uh, so these are some of the things which you need to be very careful when you’re looking at drivers. Like not just look at one dimension. Like you could look at multiple dimensions there.
Glenn Hopper:
Yeah, and I’m sure your models, I, I get very complex and the, I think about some of the models I built early in, in my career I was so proud of and there were just so many stacked assumptions and formulas going through. And then when you’re trying to pitch that to management and they ask a question, how did you get this number? And you have this long route that you have to get through it, obviously you want to keep them as simple as possible, but sometimes when you start putting in more factors in it’s complex business and you’ve got all these different skews and they’re all affected by by different things, the models just inherently are gonna get more complex. And I think the interesting thing is because of Excel, it can be even harder to kind of trace back the roots to it versus you can show the formula in Python or or MATLAB or whatever where you can say, this is what we’re doing, this is the regression we’re doing or whatever.
But I think it’s tough. You want to have as detailed a model as possible in fact and put in as many variables and, and factors in so that you can, uh, stress test a do scenario analysis plan for different situations. You’re trying to sort of build out this perfect model, but you also have to balance that complexity with usability and explainability. And I’m sure you have this if you’re pitching a model in your, in your forecast to, to a business, they’re not gonna just say, okay, great, you waved your magic wand <laugh>, uh, wand and, and you know, shook up the uh, uh, the crystal ball or whatever. And, and here’s what it comes out. So how do you balance that level of complexity with the explainability and usability and is there something that fp and a pros could take from that as well?
Akil Khunger:
Yeah, I mean that is something like I’ve learned over the years. Like if you go to someone like business and just try to see like a whole mathematical equation, they will just pretend to listen, but they’ll not be listening to you. So I think the PowerPoint became a key here. I think the way you present results, and I think think in terms of relationship with the drivers, it’s should be clear relationship. You don’t need to go ride the equations, but put something like, this goes up, this goes up, or this goes down something simpler if you have some triggers, okay, this is the trigger point and it just goes above that. This happened, this happened. Like sometimes you don’t have to go into detail how you got there. I think you should definitely know that. And if somebody wants to know that and like some people are more tech, you can, if you try to showcase how you get there, it can become more difficult defend in terms of people who don’t understand technical term.
So focus on what you have and like why do you think is correct? Like not because how you got there and have like all the supporting slides in the appendix or somewhere like how you got there if it is needed. And then show different security analysts, like if it changes like this will happen, like show some benchmarking like maybe with their results, right? Their, their business planning model fp a model. Like, okay, this is the difference we are seeing and this could be because of the s sector we are, we are not considering any new businesses. We don’t consider this from the plan and this might need like, so those kind of things we need to list very carefully. Another thing is like not try to cover too much stuff in one goal, like, uh, uh, try to focus on small things like even if it’s a 15 minutes meeting, have more 15 minutes meeting rather than like have more like one hour meetings like, uh, that can either, like you can lose the plot, like, and they can under, they can have better understanding of small things, like especially when you talk about technical terms.
Glenn Hopper:
Yeah, that’s great advice. We talked a little bit before the show and this is, I know you in your typical modeling rely a lot on historical, the more you can find the seasonality and the trends and be able to have that as your baseline. But in corporations, so many times we’re modeling something whether we’re going into a new market or we’re launching a new product or it’s a, a brand new innovation that there’s not really anything historical to base it on or, um, or even, I guess this would apply obviously to any startup that’s coming into a space. Um, but it can be very difficult to model that when you don’t have the, um, that historical data. And I’m wondering like if, if we’re considering launching a, a new market, say maybe that’s a a good example to use, what kinds of like risk and return modeling technique should we use, uh, when we’re trying to figure out if that makes sense from the beginning when we don’t have that historical data for the market?
Akil Khunger:
Yeah, I think one thing which I have done that in the past is like for example, that product is new in us. Like does that product exist in your portfolio in other region? Try to get the data. I know it is not going to be exact same, like you have some difference, but it’ll give you an idea like how the business generally react, right? Like through the drivers at that business, at that region. That is like one way to go about it. Uh, other ways to look at any industry references, like you can find like, uh, any industry data or something or something similar like, uh, even not that product but some different product, uh, or that product might be similar to some other product in your business. You can use that as a reference like, uh, until a time you get enough data to model that separately.
So if two businesses are expected to move similarly, then you can use others as a pivot to model that if, if either the industry or the similar business in other region is not available. And then having said that, like sometimes you really don’t know anything and then you just need to properly follow the path of like fp and a like say, okay, agree to business like what you say, at least for six months, that’s what we are gonna do and we are gonna gather data for six months and then we will come up with a new model. Like that’s sometimes you have to accept and do that.
Glenn Hopper:
And machine learning is so amazing given enough data how it can, how accurate it can forecast. But we know, I mean, time series analysis is very difficult and there, that’s why there’s, uh, that’s why we’re not all just letting robots, um, pick stocks for us and caching in <laugh> on our, our great stock picking, uh, time series tool. There’s no bigger proponent in the world than, uh, for data-driven decision making, uh, than I am. But human judgment still plays a role. So I’m wondering how do you, well there’s kind of two parts to it. One, where do you think, again, you can build the best model in the world and that’s a great map, but you don’t wanna confuse the map with the terrain, uh, all the time because things happen. The black swan events or just some assumptions, some base assumptions were wrong. So one, where do you think the role of human judgment comes in when you are modeling and you are making determinations about which direction to point the model? And then the second part of that is how do you then present all the findings from the model to people who are not technical and they don’t want to hear about the whatever tools, seasonal projection tools that you used, and they don’t want to know how the sausage was made, they just wanna know that they can trust it.
Akil Khunger:
Yeah, I think that’s like, I guess everyone, uh, judgment is very critical. Otherwise nobody will, uh, offer us will have any jobs like robots only working, uh, there. So machine learning or just gives you an idea, like it gives you like some good candidate to choose from initially and then they may not be like, maybe the top five are not the best one. You never know. So, and then I think the first judgment is on your side, like based on the ideas you know about the business and your own knowledge and your own thinking, okay, maybe even the first model looks statistically the best model or machine learning is saying this is the best model. But then no, you think, okay, this relationship maybe data is showing it doesn’t make sense, like, uh, or if relationship is right, like if you see the forecast scenario says, okay, in this scenario it is showing like something this like this will happen, but this is not gonna happen, right?
Like it could say the interest rates going up and your revenue’s going to really shoot up, but you think it can increase, but you’re not going to like double, like if interest rate double right, then new factors will come into play. So those are the things you need to be careful. And then that’s where also like then you talk regularly to your business partners and get their feedback okay, what they think. And in terms of that, I think as I mentioned earlier, like don’t need to mention about everything. Like say, okay, these are the candidates, show them like the relationship, show them the forecast scenario analysis, some <inaudible> analysis, show them like changes in the outputs and let them see okay, what looks more reasonable. But yeah, having said that, like sometimes we do need to, they may say, okay, they may just want to pick something. And if you strongly see that, okay, mathematically it’ll not make sense at all, then sometimes you have to really stand your ground and say, okay, uh, like there are some other publications because all the strict integrity standards like sort of validation teams in the company, you can’t just pick anything like, uh, it has to be a mixture of both. So I think, yeah, human judgment plays a rule a lot like, uh, in everything
Glenn Hopper:
And on the human judgment domain expertise is so important. And we talked about democratization of data for years and how great that was, that there’s so much data out there and it’s so much more accessible by more people and that, and that’s great, but just having the data wasn’t enough to turn it into value. The, the barrier to entry for being able to use that data was a background in BI or data science and understanding machine learning and knowing how to manipulate the data and what to do. But even understanding all that, you also had to write Python. You had to be able to write SQL queries, you had to be able to speak the computer’s language to talk to it. And now, and, and there’ve been for years development of a lot of, uh, low-code and no-code tools that give access to people who otherwise wouldn’t have it.
And now with generative AI and vibe coding, people can not have really much domain expertise in, in data and in engineering at all. But they could ask, um, chat GPT to write code that would do Monte Carlo simulations, do uh, black Shoals what, you know, whatever they’re building and they could not really know what they’re, they’re doing. I feel like with vibe coding, you could go <laugh> and just create an app in, in 20 minutes that is, is doing something for you, but then you turn it over to people and when they ask where the numbers came from, and this is certainly not in for compliance and audit and all that, you can’t just say, oh, the magic black box gave it to me. That, and that’s, you know, people can get in trouble from that. It just feels to me like the future of internal fp and a they are gonna be trying to model to the level that you guys do now. And I’m wondering, I guess from your perspective, having that deep domain expertise, if we were gonna have access to these tools, you’ve gotta understand how they work. So I’m wondering what would need to change in fp and a hiring and upskilling so that if we’re using these tools, we’re using them responsibly and we can explain it. What, what, what do fp and a teams, if they are gonna have access to more complex forecasting, what do they need to know and understand and be able to use and be able to talk about?
Akil Khunger:
Yeah, I think they need to like not think too much about coding part of it because that chain AI and everything is gonna take care. Or you can mention, okay, I want to look at this and it’ll create that for them. Or even if they have to write a code, they’ll be to provide a code as a new copy paste and make minor changes. So you may need some coding, but not that much. But I think what becomes more important is understanding the math behind it or like coming back to principles of statistics and modeling, like, and also finance principles, not just math principle, like whether it makes sense in finance setting, whether like, suppose you get like just a very bill example, like assets as negative, like whether it even makes sense, right? Something like that. So like those kind of skills like become more and more important, like how to use AI more efficiently.
So I think, uh, fp and a people, I think they, I I, I’m assuming, like I know I’ve seen that people already have good finance knowledge, but they need to brush up more on like have to get good statistic knowledge, statistic specific. I would not say like pure math statistics knowledge is like very important. And I think that might, you see that is more and AI come into play, like people who have good technical background, they may end up getting more into these roles, but yeah, doesn’t mean that finance skill will not be important. They would be like equally important if not more than what they’re right now. Uh, what does requirement might reduce is terms of reporting or something. The tedious task can be automated, uh, if you come up with an automation structure and use AI more effectively.
Glenn Hopper:
Yeah, and I think you’re so right on statistics because that’s the gateway drug into machine learning and into the, uh, <laugh> and into everything because it’s, and there’s, there’s different levels of statistics. There’s the business statistics courses you take in an MBA program or, or undergrad, but then if you go deeper into it, there’s really, you start to see a lot more value in, oh wow, I really could apply this, uh, statistical modeling approach to, to my fp and a. And when you understand that, and you don’t have to go into our studio and or Python or whatever and do all that, then, but if you understand it fundamentally, then you can do it very quickly and you can explain the model and you can explain how you did your model selection and the, the testing and the sampling that you did and everything. So, yeah.
Yeah, I’m, I’m right there with you on that. For our listeners, maybe their early career and they, they’ve started out in fp and a and they, they listened to you and they think, man, I really wanna, I want to go be a quant if someone, you know, you came into it through engineering in that way, but if someone, say they have a, a undergrad or a master’s degree or whatever in, in finance, and they’re thinking, I really wanna make this shift. What do they need to start preparing for and thinking for as they line up their education in, in which direction should they go educationally, um, to move into that direction?
Akil Khunger:
Yeah, uh, I think definitely have some statistics course in the background, like, uh, when they’re doing their bachelor’s or something, or even like some other degrees they might be doing on a site or even running on their own. And then some coding. I think coding, again, you don’t need to do the course, but need to spend enough time on that. Uh, I mean, I, I do think coding might become less important over time, spin these hours, but for now it is still there. Like, uh, maybe it is, maybe it’ll not be there in the actual world, but in the hiring, like people may still grill you on coding, like, uh, for some more time, uh, a few more years, I’m pretty sure about that. So that’s why you do need to like, work on these two things. Like, uh, maybe not like too much like, but yeah, some of it is required. Yeah.
Glenn Hopper:
Yeah. Yeah. This has been good. I’ve, I’ve loved this episode going in diving into a, a little bit different world than where we normally spend our time as we’re winding down, we have two questions we ask every guest. And the first one is, what is something that not many people know about you? Something we couldn’t find from your LinkedIn profile or looking you up on Google?
Akil Khunger:
Uh, as well, when I actually went one to Ukraine, like before the war, like, uh
Glenn Hopper:
Oh, wow.
Akil Khunger:
Yeah, so it was quite beautiful. And I did, like, in my undergrad, I did like a social internship, like working with some steel factories and like trying to talk about pollution or some stuff and meeting with young children and like trying to teach them English, like, uh, status. Like I did one of my summer like in undergrad. Quite exciting experience.
Glenn Hopper:
Wow. I’ll avoid <laugh> going into all the political and global ramifications of it, but I, I wish for peace in the region and I know there’s been a lot of destruction, so. Alright. Everybody’s favorite question and I, I think this is gonna be interesting from you ’cause I, I, I don’t know how much time you actually spend in Excel, but we love here at fp and a today and, and at Data Rails we love Excel. So we ask everyone, what is your favorite Excel function and why?
Akil Khunger:
Oh, I think I use a lot of things in Excel. Like even thing, like, we do a lot of thing in coding, but we still use Excel a lot. And I think still for another favorite function everybody must have used, we look up, I think it’s like, must like, uh, it makes some things very easy and like to do quick analysis, dirty analysis. Sometimes lot of times you have to do a very quick analysis and setting everything in the code and you don’t want to spend that much time and you go and excel in these kind of tools, like help you a lot.
Glenn Hopper:
So I, when I was a guest on this show before I took over as host, uh, my answer was also V look up. And since then I just get railed on it. Everybody’s like, X lookup, what are you talking about? Or index, match or what,
Akil Khunger:
Yeah,
Glenn Hopper:
<laugh>, uh, whatever. But, uh, yeah, I, I still, I I, I have to admit, and again, I’m not, I’m not deep in Excel as as I used to be, but I’m, I’m still gonna use VLOOKUP a lot. I don’t know. It’s hard to, it’s hard to shift what you were raised on <laugh>. Well, Akil, this has been fantastic. I really appreciate you, uh, coming on the show.
Akil Khunger:
Yeah, thank you. Thank you. A great talk.
