Hello:Need help here guys. I am very good with matlab..I would like to do some research using agent-based modeling (but alas!! although I have read a little about it..and do not know Jack about how this is done...I can figure out myself but first...1) Does matlab have any agent-based capability? I look around and it seems it does but I couldn't get anything specific..such as what is needed to make it work (if any specila feature is needed, etc so I can make the University get it)2) What is the industry standard software for this? Would any paper I do get done (if) will be down-looked by not using the standard package?3) Are there any freebie that is ALSO any good? (Notice that I am not asking for any information that can be googgle out, ok? If you have persona hand-ons experince THAT IS THEKIND OF ADVICE I NEED.Thanks, guys, Hope someone help me here.M

maelohere at ccfea we have an msc on agent based finance/economicsthe guys that teach it use matlab for the part on optimizationfor the agent based part they use JAS and SWARM (Java based)you can find more info at our website www.essex.ac.uk/ccfea there are some tutorials there for the software used etcwhat are you planning to do using agents? if you want i could give u the details of some people that do their research on agents here (staff + phds)kyriakos

REZ:Well, it sounds pretty stupid idea but....we have some "krazy" idea that the behavior of managers is not quite aligned as it should be with stockholders' regarding the risk tasking of projects..that is, we think that since managers stakes at losing their jobs (perks, etc.) in the company is higher than the stake that many stockholders have (most of then proxies voters in any case and some do not even count as managers control goes).So, if managers are more scared to loose their jobs and have, (basic assumption here) much better information than outsider stockholders about the firms' projects risk...they would discard (more often than not) projects that would benefit stockholders but are too risky... from managers view.So, standard theory goes, the solution to align interest is: the manager becomes a stocholder (or similar situation, ergo, give them options)...but then, if the managers biggest stake is in the firm,,,to give them more stakes on the same firm would only give them more reasons to avoid projects that would benefit stockholders (themselves included) but would not be of much more benefit to managers as they still face loosing their bigger interest (employment).Some questions here to be answered (hopefully??):1) Is stardard theory correct? Meaning, giving managers options really align them with stockholder's interests?2) If No. 1 answer is yes (should be)..how much "stockholder interest" is neeed to overcome the possibility of loosing their job? (How many options, of what kind maybe???) or when does exactly the manager becomes a stocholder? One option, ten..maybe one thousand?? On what factors does this "Alignment" depends (company size??? )3) Would it be the case that instead of "aligning" the managers interest, when they become stockholders as well, their interest goes the other way around? Do they become even more risk adverse?? After all, now they DO have even more to loose, don't you think? 4) If No. 3 answer is yes (who the heck knows?) then, to provide options from others firms (that is, to diversify the managers' portfolio if they do a good job) could be a better solution? Maybe, give them puts of a rival firm??I was thinking (well, I am not sure I was, when this idea came up and I decide to join the group of rebels)...in any case, I think that a model MUST be quantitative here...a simulation that shows results similar to what we should observe..that is, we give our "managers" options of our firm ten if the do their job (stock price goes up) when presented (INPUT) with different projects measured by their level of risk (least say managers risk: low, middle, high and stockholder risk: low, middle, high..and of course their interactions (LL,LM, LH, ML,MM,MH, HL,HM,HH)....and then compare the simulation results...Managers are expected to select projects with low risk from the stockholders view, not their view . Other factor might be include as well (maybe) .Then, the bomb...we switch...We give the managers option FROM A different firm than ours...and we se what happen then... Well, I haven't figured it out yet in detail...we just discussed the idea last Thursday, OK? We also where pretty drunk... So I am just toying with the idea...but something in that line I think should be the correct procedure..and I DO feel that agent-based modeling is the way to test the thing.Well, you can laugh now but...help is greatly appreciate it!!M

Try netlogohttp://ccl.northwestern.edu/netlogo/Its free, easy to learn and seems like it could incorporate the kind of rules you've listed.Good luck!

QuoteOriginally posted by: MaeloHello:Need help here guys. I am very good with matlab..I would like to do some research using agent-based modeling (but alas!! although I have read a little about it..and do not know Jack about how this is done...I can figure out myself but first...1) Does matlab have any agent-based capability? I look around and it seems it does but I couldn't get anything specific..such as what is needed to make it work (if any specila feature is needed, etc so I can make the University get it)2) What is the industry standard software for this? Would any paper I do get done (if) will be down-looked by not using the standard package?3) Are there any freebie that is ALSO any good? (Notice that I am not asking for any information that can be googgle out, ok? If you have persona hand-ons experince THAT IS THEKIND OF ADVICE I NEED.Thanks, guys, Hope someone help me here.MHi m8Just got kyriakos' email. I'll look at your longer post latter, but as far as matlab is concerned. The simple anwser is yes! in fact it is an all inclusive kit, so you are able to program just about anything; neural networks, genetic algorithms, learning systems, and so on. Its only draw back as I discovered is its not too intuitive. Which is why I prefer using Java. However, depending on your level of experience with MatLab, you may find this not to be the case. The basic problem is that MatLab is more a Mathematical/statistical/science/engineering orientated enviroment/Language, its not really good at handling OOPs programing though it can with some sweat and blood.However I've found that for proper agent modelling you need to be able to handle objects (your agents are effectively objects consisting of or using other objects) and make them interact with other objects a lot more flexibility than I feel is possible with MatLab. Which is why a more object orientated programming language (e.g. C++, Java) would be your best bet. Especially seeing that from the length of your second post, I get the impression you would want your agents to do a lot, which in turn means you may want to later add more objects, experiments and so on to your benchmark model. I just feel trying this in MatLab will be more of a headace in the long run.If you google ACE/agent modelling in Java/C++ you should find links to the various libraries and papers. JAS ( jaslibrary.sourceforge.net ) is what we use but you may also want to look at RePast ( repast.sourceforge.net ) purely because there's more documentation out there on the net, seeing as you are new to agent modelling. There's also java-sarm and I did come accross a set of libraries specifically for C++ programmers but can't remember what its called. However, they are all based on SWARM ( www.swarm.org ), which is in Objective-C.Re: netlogo you can try this out first but again because its a complete program in of itself, you may not get the flexibility you may later want. However its a very good place to begin.The rest is all reading and developing your modelhope this helps

Just read through your second post and like I said, you really DO NOT want to be using MatLab for your project. MatLab claims to be object orientated, but it just gets too messy. At CCFEA we do use MatLab in the first term of the CF902 module but the algorithms coded in that section of the course are pritty easy to code in MatLab. However when it come to the bigger programs which your project seems like it definitely will be, its just much more efficient to use a real OOP programming language, hence Java. Re the project itself:I wish I could thinkup ideas like that when drunk, LOL!Seriously though it is a good idea and agent modelling is probably the best way to go. Sadly can't point you in the best direction for papers on your topic area though once you finish going through what you can access from the CCFEA site you may what to get hold of this book:Economic Simulations in Swarm: Agent-Based Modelling and Object Oriented ProgrammingSeries: Advances in Computational Economics , Vol. 14 Luna, Francesco; Stefansson, Benedikt (Eds.) 2000, 328 p., HardcoverISBN: 978-0-7923-8665-0also take a look atwww.duncanrobertson.com/research/Emergence.pdf orisce.edu/ISCE_Group_Site/web-content/ISCE_Events/Naples_2002/Naples_2002_Papers/Robertson.pdfthe papers are not on your idea but may have references to more useful behavioural modelling and strategy optimisation stuff.

obewaj:I will check it out. keep in touch, ok? I am pretty sure I will ask a lot more question later.On the "good ideas" while drunk topic...I wish I could any idea at all WITHOUT being drunk!!!

No worries m8 will be on the lookout!

Well I think your question was answered. But just for some 411's on agent based modelling as with any paradigm it can be simulated in principle with any high level language worth its salt. So I guess its upto the creativity of the programmer on implementation issues. However for robustness of your simulation esp with visualisation, radom generators, adaptive data stuctures e.tc you're better off with the specialised API's . Here's an example of what has been done at CCFEA , its based on VB 6.0 which is not fully object oriented more of modularised procedural programming. BTW VB 6.0 was mainly used for RAD (Rapid Application Development) of business logic and related information systems, it just show that agent based modelling can be done in practically any language, there are consideration of course depending on tradeoffs and or application domain