I highly approve of the amount of lemonade in this video. I do believe however that people that let their lemonade consumption depend on weather conditions are not living life to its fullest potential.
This is the best explanation I could find for an MA(1) process and you generally have the best explanations of econometrics concepts. Thank you for posting all these videos! :)
Hi, thanks for your question.Sigma doesn't effect the conditions under which the process is stationary or not, so in that sense it is not an important parameter. However, sigma is important in the sense that it controls the variance of e; so in the above example a larger sigma means that the variance of the shocks is higher. I hope that helps! Thanks, Ben
My interpretation of an MA process is that, for some variables, the error term may contain a sort of omitted explanatory variable (change in weather here, e.g.), and so by regressing on them we may partly be regressing on some feature that is vital in explaining the makeup of our dependent variable. Correct me if I'm wrong in my thinking
Wow! I dont think Ive ever actually seen proper examples of MA processes in the books. They just talk about the process in general in a theoretical way. But this way you'll learn what to expect from a series before you even look at its correlograms
Great video, but some question about the explanation about why the change of Lemonade sales drops at time t+1. At time t-1, the change of sales is 0, at time t, the change of sales is 1, at time t+1, the change of sales is -0.5. So the sales will increase by 0.5 and keep still. A better explanation could be: when the temperature first increases at time t, people buy more lemonade; when the temperature stops increasing at time t, people buy less lemonade than time t but still more than time before t; this is because people tend to buy much more lemonade when temperature increases for the first time(let's call this shock), and when the temperature becomes stable, they tend to buy less(shock ends).
This video is helpful, but I still have two questions like how they get MA(0) for x(t) = u +ε(t). Another question is that are those error terms calculated from random walks or autoregressive model?
What i get confused is that the change in temp must have its own time series ar model, than from that you calculate the error term, then use that value for this MA(1) model, so basically your using one time series model and feeding those error terms into this MA model?
i don't understand how can we calculate et and et-1?? I mean, we can observe Xt (yt), it can be oil price, gdp etc, but what is e? It says on wiki it is white noise or schock. ok, but how it is usefull to say that something is equal to white noise or shock?
My understanding is shocks are not measurable..but they are having an effect on the system...in this case in Stock market as an example...if we hear that the FED reserve is stopping Quantitative easing or another postive or negative annocement, then it will create a shock to the time series of the market, if you want to call it that...that shock is the annoucement...and then we see a residual effect as time goes by t-1 or t-2 lags from today depending on the model..recall that whote noise is give mean =0 and with variance sigma squared...this is be deifnition...
So basically your using two variables, temperature and change in lemonade, corresponding to each time index, instead of using the own errors of past lemonade errors on itself?
I still can't see the use of AR MA and their combination, why would I make a model with only lags of endogenous variables or error term, please explain to me the usefulness of them. thx
So in the oil example how will the oil prices drop back to its original level? The delta oil price just keep increasing when you get hurricanes and then become zero after 2 days? I guess you have to properly quantify the error variable for that..
I can't understand how come the shock that has occured in the current period t has a residual effect in the previous period t-1. It makes more sense to me if they are two different shocks occuring in different periods with the one in t-1 having a smaller weight since more time has passed. What am I missing?
Thanks for the great tutorial. But I got a doubt with the first example. The model you have mentioned -0.5Et-1 but when you explain, you changed it to Et+1. How is it?
How is epsilon_t represent the change in temperature, while epsilon_{t-1} represent the change in lemonade? Same greek letter represent two variables??
I dont like the way you disguise the x_t variable as a change in demand, wouldn't it be more appropriate to denote x_t as the error hat term in the next period to ease understanding
I just tested 2 MAs (at the same time) on GBPCHF. Together they actually prevent me from getting into bad trades as long as I’m in addition trading with the *ASH STRATEGY by aleksandrov* (you can google it) while there are just a few where it gets me into bad trades but I would say it's rare, and I doubt there's a baseline or two that work together that prevent you from taking a bad trade 100% of the time. I know you said to find a unique baseline, what if we used 2 instead? But not entering when they cross, but entering when price crosses both. Maybe it's just for that pair, but I will tell you, out of 20-30 trades in the past year, only 3 were bad. Even less if you wait until price crosses and closes the two baselines. You might be missing some trades, but you're gaining more pips overall because you're not in losing trades.I will deff keep looking for more, and keep backtesting and forward testing, it's actually wicked fun specially when you see results. I wouldn't feel 100% right making money if I didn't earn it. :P
I highly approve of the amount of lemonade in this video.
I do believe however that people that let their lemonade consumption depend on weather conditions are not living life to its fullest potential.
+BertrandArne Haha. I agree, there's no way weather gets in the way of my lemonade habits. Best, Ben
This is the best explanation I could find for an MA(1) process and you generally have the best explanations of econometrics concepts. Thank you for posting all these videos! :)
This is the best example of MA process I have seen across books, videos, blogs, classroom lecture notes & other internet sources. great work!
Hi, thanks for your question.Sigma doesn't effect the conditions under which the process is stationary or not, so in that sense it is not an important parameter. However, sigma is important in the sense that it controls the variance of e; so in the above example a larger sigma means that the variance of the shocks is higher. I hope that helps! Thanks, Ben
Thank you sir ben! You're my true professor in econometrics. Never learned a thing from school. :)
I will name my first son after you i swear it
+A Person Ben Person? Good name :)
Five years later, he could sire mine
Thanks Prof. Lambert for the illustration.
Yes. That's correct for the variance of a MA(1) process
Excellent intuitive explanation of the MA modell..
My interpretation of an MA process is that, for some variables, the error term may contain a sort of omitted explanatory variable (change in weather here, e.g.), and so by regressing on them we may partly be regressing on some feature that is vital in explaining the makeup of our dependent variable. Correct me if I'm wrong in my thinking
Great explanation; very helpful! Thank you Ben!
... I love you.
Great videos, thanks for all the effort! However, I don't understand why the error epsilon is determined by temperature change?
Thanks a lot. You save lifes, I guess ;)
Thanks Ben, greetings from Portugal
perfectly explained - thanks again for yet another helpful video :)
Wow! I dont think Ive ever actually seen proper examples of MA processes in the books. They just talk about the process in general in a theoretical way. But this way you'll learn what to expect from a series before you even look at its correlograms
Great video, but some question about the explanation about why the change of Lemonade sales drops at time t+1. At time t-1, the change of sales is 0, at time t, the change of sales is 1, at time t+1, the change of sales is -0.5. So the sales will increase by 0.5 and keep still. A better explanation could be: when the temperature first increases at time t, people buy more lemonade; when the temperature stops increasing at time t, people buy less lemonade than time t but still more than time before t; this is because people tend to buy much more lemonade when temperature increases for the first time(let's call this shock), and when the temperature becomes stable, they tend to buy less(shock ends).
This video is helpful, but I still have two questions like how they get MA(0) for x(t) = u +ε(t). Another question is that are those error terms calculated from random walks or autoregressive model?
What i get confused is that the change in temp must have its own time series ar model, than from that you calculate the error term, then use that value for this MA(1) model, so basically your using one time series model and feeding those error terms into this MA model?
Great video! thankyou
i don't understand how can we calculate et and et-1?? I mean, we can observe Xt (yt), it can be oil price, gdp etc, but what is e? It says on wiki it is white noise or schock. ok, but how it is usefull to say that something is equal to white noise or shock?
My understanding is shocks are not measurable..but they are having an effect on the system...in this case in Stock market as an example...if we hear that the FED reserve is stopping Quantitative easing or another postive or negative annocement, then it will create a shock to the time series of the market, if you want to call it that...that shock is the annoucement...and then we see a residual effect as time goes by t-1 or t-2 lags from today depending on the model..recall that whote noise is give mean =0 and with variance sigma squared...this is be deifnition...
Nicely explained Ben
Good examples for sure
So basically your using two variables, temperature and change in lemonade, corresponding to each time index, instead of using the own errors of past lemonade errors on itself?
I still can't see the use of AR MA and their combination, why would I make a model with only lags of endogenous variables or error term, please explain to me the usefulness of them. thx
Sir, how we can calculate moving average from previous data.
question, how does white noise where e sub t are iid N(0,.25) affect such models? is sigma significant?
So in the oil example how will the oil prices drop back to its original level? The delta oil price just keep increasing when you get hurricanes and then become zero after 2 days?
I guess you have to properly quantify the error variable for that..
Thanks Ben :)
Very nice video, thanks!
is Cov( Xt , Xt-1) = 0 in the cas of MA (1) then ??
Is it realistic to use OLS to calibrate an MA(q) process? Or the way to go is always MLE?
still very hard topic but amazing video
@Ben Thank you for sharing! What is the significance of using MA model in the context of ARMA(1,1) model?
You're my hero
Thank you so much sir!!!!!!!!!!!
I can't understand how come the shock that has occured in the current period t has a residual effect in the previous period t-1. It makes more sense to me if they are two different shocks occuring in different periods with the one in t-1 having a smaller weight since more time has passed. What am I missing?
Stanimir Sotirov The shock on period t doesnt have an effect on period t-1, it has as effect on periods t and t+1
Thanks for the great tutorial. But I got a doubt with the first example. The model you have mentioned -0.5Et-1 but when you explain, you changed it to Et+1. How is it?
whether it is from Et to Et+1 or from Et-1 to Et, both represent moving forward one time unit (the t). so both are equivalent, as far as i can see.
could you please help me with Mean Reverting process. thanks.
How is epsilon_t represent the change in temperature, while epsilon_{t-1} represent the change in lemonade? Same greek letter represent two variables??
i looked up: var(y sub t )= sigma squared + sigma squared * theta squared, am i right
Thank you very much! I like your accent. Good to understand even for germans. Are you from the south of england? greetings
Even for Germans, that is a compliment. London area is where he lives.
I dont like the way you disguise the x_t variable as a change in demand, wouldn't it be more appropriate to denote x_t as the error hat term in the next period to ease understanding
good video by the way
Whui duint yu stai in england und shuit up?
?
I just tested 2 MAs (at the same time) on GBPCHF. Together they actually prevent me from getting into bad trades as long as I’m in addition trading with the *ASH STRATEGY by aleksandrov* (you can google it) while there are just a few where it gets me into bad trades but I would say it's rare, and I doubt there's a baseline or two that work together that prevent you from taking a bad trade 100% of the time. I know you said to find a unique baseline, what if we used 2 instead? But not entering when they cross, but entering when price crosses both. Maybe it's just for that pair, but I will tell you, out of 20-30 trades in the past year, only 3 were bad. Even less if you wait until price crosses and closes the two baselines. You might be missing some trades, but you're gaining more pips overall because you're not in losing trades.I will deff keep looking for more, and keep backtesting and forward testing, it's actually wicked fun specially when you see results. I wouldn't feel 100% right making money if I didn't earn it. :P
Not clear explanation of MA-coefficients.
Explanations are confusing