It is not possible, but you can construct a new one. It takes effort, but you learn a lot through the process.

We drop one variable in a category of dummy variables for many reasons. In our case, the category is the yearly dummy variables. The first reason is to avoid multicollinearity, which otherwise would be present. The sum of dummy variables inside a category is 1, thus, if V2016 and V2017 are both 0s, then, the model considers these data points as for the absent variable V2015; note that in this case, we avoid redundancy. Another reason is the reference point; since 2015 is omitted, the coefficients of the other years become in reference to it, and not with each other's. Try the example while omitting 2017 instead of 2015 and notice the difference. Note particularly what happens to the coefficient of 2016, because it is now in reference to 2017. Last, please have a look at the dummy variable trap.

You are most welcome.

Thank you!

Hello, you can use the trading company template for that purpose.

You are most welcome.

Happy to hear so!

You are most welcome!

Hello, you are most welcome. The index in this case is for a given period. You can still estimate what you are after by considering the average change.

You are most welcome.

Hi Oliver, you are most welcome. Yes, they would.

or something similar, with lots of thanks

You are most welcome. I don't have such, but you just need to tweak this template a little to get what is required. Have a look at the previous video, A Feasibility Study - Step by Step. Together with this template, you should easily accomplish what you are after.

You are most welcome. Glad to hear this.

Hi, glad to hear that. Please let me know how I can help.

Hi Pablo, you are most welcome. I would assume that defining the "weight" of the GLM model as the square root of the fitted values would do it. However, you can also use the second method for this step as illustrated at 4:15.

@@AliNasserEddine Thanks for the answer! When you say in 00:43 that "in the second stage the squared residuals from the first step are regressed on the holding periods plus a constant", does that "constant" mean the following "0" in R code? Or how can I incorporate that correctly? Like this: index2 <- lm(formula = index$residuals^2 ~ 0 + holding_periods, data = holding_periods) . I would really appreciate your answer, that would help me a lot in my tesis. Thank you Ali!

You are most welcome. No, we leave it to be estimated by the regression model itself. In the first step, we set it to 0 by adding 0 to the model like you illustrated. Also in the last step, we do the same. But in the second, we leave it out of the equation so the intercept is automatically estimated. That is, ...formula = index$residuals^2 ~ holding_periods... should do it.

Glad to hear so. You may refer to the tutorial on using the Case-Shiller methodology for repeat-sales regression; they use some type of weighting; it is based on observation level. Here is the link: kzbin.info/www/bejne/mobaiK2Oi8eFfqs On the other hand, I believe what you are after isn't straight-forward and hard to be accomplished through regression. For instance, there might be variables with negative coefficients because they negatively contribute to prices. More importantly, the variables in a regression model aren't constant, and the coefficients you get through regression are your best estimates. If you are to use weights based on the variable's contribution to price, then you are implicitly assuming that the variables are constant! Having said so, there are definitely many ways to accomplish ideas. If you explain precisely what you are after, I may be able to help you better.

Thanks for reading the comment. So now i want to evaluate the base price of my property for renting(short term) . Ex: I want to measure how much weight does amenities contribute to price, location contribute to price, number of room contribute to price,etc. For me im just a begginer about data, my way to do is test regression all of that elements to find coefficiency, after that i find weight by how those element correlate with the price. Its so meaningful for me if you can so me the procedure. Have a nice day!

You are most welcome. I think the coefficients can be sufficient in this case; please note that many call them the impact on price, which somehow equal to "weights". If we can isolate all variables and determine their costs, then we can easily determine their weights through the division by their total. In your case, the most important is the data. You need data on short-term rent for other properties in the country you are studying in order to evaluate the effect of variables on rental price. For example, let us say there is another apartment in another city that is similar to yours in terms of size, number of rooms, etc. If the short-term rent for this apartment is 100, and for yours is 120, we can say that your location adds 20 to price compared to the other location which would be the reference in this case. The point is that there should be a reference point; hence, the impact of a variable on price would be in comparison to and not absolute. However, you want to know the base price and the effect of variables from there. This represents a little issue, because what do you consider as base? Just the apartment unpainted and without windows and tiles? Or a finished building without land? One possible workaround is to do the following: First, you calculate the current sale price for all apartments in your building. Then, you check the value of a similar land; the real base for any building. This should give you the price contribution of building. Then, you check the price of a very basic unpainted non-ready apartment near to you, and compare it to yours. This should allow to evaluate the impact of additions in your apartment to price, and so on. However, such process is inefficient. Moreover, it won't be really accurate as we might assume. In this sense, it is better to use reference points than absolute values. Such reference points are incorporated in regression models when we drop values. Data is key for good estimates, and the process is illustrated in this tutorial. I hope the above helps a bit, but please let me know if further clarifications are required.

@@AliNasserEddine thank you so much for those words, you understand my point and i have exactly problem about gathering data same what you said. If you have spare time can I contact with you to collaborate about this project?

You are most welcome. I will have better schedule in few weeks time, and I am open for mutual benefit and happy to collaborate, but it depends on the type of collaboration. Please feel free to contact me in any case. My email is [email protected].

You are most welcome.

You are most welcome. Generally speaking, the more data you have the better it is as long as there is no overfitting, but it really depends on the case. Sometimes, 5 data points are representable for a million. Also, the nature of the matter plays a role. For example, it is easier to get data for rental properties in New York than it is for getting data about heart surgeries due to frequency. Also, it is easier to get data on rental properties in Sweden than in a third-world country due to the level of corruption, professionalism and technology. There is no given answer to your question, but many argue that 10 data points for each variable should be sufficient; personally, I judge the matter and form my opinion accordingly. Please note that you can still run the regression with fewer data points, and the reliability of the results depends on the nature of your data. You may also want to check the one-in-ten rule.

By using averages. This method is for creating an index. Your goal can be reached ~by using averages. If you provide me with further details such as the variables you are taking into consideration, I can help you better.

@@AliNasserEddine thank you sir,...how can I contact you for further details

You are most welcome. Please write it here so everyone can benefit. If the case i.e. your project is private, please be general.

@@AliNasserEddine okay,...for example,I have data collected from the usage of resources in a cloud computing infrastructure especially IaaS and I want to predict the price and deliver a bill to users. is it possible to use hedonic regression to do that prediction or estimation of the bill?

No, the Hedonic regression framework is designed to develop an index that measures the market movements for heterogenous assets. In your case, they mostly charge per GB or in similar manners, and the prediction you are after is tied to the users/clients and not the suppliers. You can estimate the usage for a client by clustering; that is, you evaluate the situation of the client and place his position with similar customers, then, you take their average, and that would be the prediction of his usage. For example, an incoming customer enquires about the cost, you check the size of his company, number of employees, etc. Then, you check among your customers those that are similar, and those customers would be the cluster. The average of their usage would be the prediction for the new client.

You are most welcome. That is correct. Please consider the following: we buy two similar houses in the same location for $100K each. Then, we go for another one and buy it again for $100K. If someone asks us how much does a similar house cost you there, we firmly say $100K. After one month, we buy a fourth with similar specifications in a closeby location for $110K. Afterward, if someone asks us about the cost in the greater area, we answer about $100K; 'about' and not forsure. Now, imagine too many houses with different specifications and locations, too many different values; whatever we say, we won't be as accurate.

@@AliNasserEddine thank you sir

Sure, but could you please advise where you exactly got this number from? In general, the numbers just follow the calculations. Is't possible you missed the net working capital (NWC)? Because we have to add it at the end.

God (swt) bless you too.

Hi Amr, you are most welcome. I shall do it when time allows but that would take a while, and there is another recommended tutorial that I said I would do. That one, then this one. However, the same concept applies to Python; it is just another programming language: different syntax, different interface.

Hi Amr, just to clear the air, the tutorial will be about Python and regression in general because the Hedonic index is just a form of linear regression. However, I might shed certain light on Repeated-Sales and Hedonic indices in the same lesson. Again, this will take until I have enough time for the purpose; maybe months, maybe years. By then, if someone else covered the same, I won't prepare the video but post an update-comment instead.

@@AliNasserEddine Hi Ali, Thanks a lot for your reply and your help. I will be waiting for your video about python. You have made the hedonic method very simple in this video.

You are most welcome.

You are most welcome. Yes, you can, but you have to switch the direction and change imports to exports. Or better, you keep the imports and add another sheet for exports. In this case, you have to reflect the addition in the calculations of revenues and costs.

Happy to hear so. Thanks for the suggestion; will do a review on the method when it becomes possible. That would take time though; will reply to this comment again when done.

@@AliNasserEddine Highly appreciated that! Reason behind as it become more popular to be used for analysis the HPI by many countries.

You are most welcome.

Which database?

You are most welcome. You are right, it is just a linear regression. One definition of the word Hedonic is "characterized by"; we call this type of linear regression hedonic because we decompose the characteristics of a certain product/object, and see their independent contributions to the formation of prices.

@@AliNasserEddine Thank you for your answer. I have a second question. Isn't it a problem that the two dummy variables RL and CL are complementary? Or the multicolinearity is not a problem with hedonic regression?

You are most welcome. Yes, multicollinearity represents a problem within the hedonic regression framework and with all other linear regression models. However, we must keep in mind that the example used here is strictly for illustrative purpose: to make it simple and clear. More importantly, the correlation between RL and CL doesn’t affect the prediction we are after, and this is observable. Please have a look at this article bit.ly/2XwWCPc; particularly, look at the third point the author raises about fixing multicollinearity. On the other hand, multicollinearity becomes a real issue when we can’t isolate the effect of an independent variable on the regressand; and this is not the case in our example.

I have no idea. Actually, KZbin doesn't notify me of comments at all, I only see them if I check manually in the comment section of YT studio. Regarding your question, the S&P has been always updating. The method used here is the one discussed in their original paper.

​@@AliNasserEddine Many thanks. My replies keep getting auto-deleted as soon as I post, so I'm keeping this short! The methodology they describe now was the same in 2015, I know because I put it in a powerpoint. It's very odd - nowhere does it refer to Bailey Muth and Nourse, just to a Shiller book which I have ordered from Amazon. Anyways thanks for your video and response!

You are most welcome. Regarding comments, please feel free to drop me an email; the address is located in the about section. As for the methodology, please refer to Case & Shiller (1987). I wrote a paper on art investment a while ago, where the difference between Bailey et al. and Case & Shiller was discussed. I can share it with you. Here is a little from it: after the seminal work of Bailey et al. (1963), various extensions to the standard framework have been suggested. Probably, the proposition of Case-Shiller is the second major milestone in the repeat-sales regression framework. In 1987, Case and Shiller proposed an extension that assumes a non-constant variance of the random component. They suggested that this variance is related to the timespan between sales . To account for heteroscedasticity in the error terms, they proposed a three-step generalized least squares (GLS) regression. In the first step, the procedure of Bailey et al. (1963) is followed exactly. In the second stage, the squared residuals from the first stage are regressed on the respective holding periods plus a constant. In the last stage, the first step is repeated; however, the dependent variable becomes the log price difference divided by the square root of the fitted value from the second stage.

Because average price is an average, whether there are duplicates or not.

You are most welcome. Unfortunately and as per the email, I can't share the original data. The sources of this data were auction houses. You can create an automation script, or use an available program to collect the data in a fast way.

You re-create the index with every additional year. That is, you run the regression again by including the last year. Also, there is an alternative technique for constructing the hedonic index. They call it chain, adjacent or pair-wise hedonic model. Instead of putting all years in one regression in one go, you put just two at a time. For example, 2005-2006 in one regression. Then, 2006-2007, etc. You keep all the other variables exactly the same as in the main model. In this case, for every regression, you would have 1 time coefficient (for the second year in a pair). Then, at the end, you chain the coefficients. Using this method, with the addition of another year, you don't need to run the regression for all years, but only for the additional year with the year before. For further information on this method, please refer to Palmquist 1980, "Alternative Techniques for Developing Real Estate Price Indexes".

​@@AliNasserEddine Very interesting thanks also good hint for the further information. I have a follow up question. Many indices call their index laspeyres. Do you think it's the same result like a laspreyres index if you only take the dummytimevariable to create the index ((vt1-vt0)/vt0) or do you have to fill in all characteristics of the mean and then calculate ((p1-p0)/p0)?.

You are welcome. The Hedonic and Laspeyres indices are completely different concepts. The latter only considers the quantity and price, while the former considers the price and characteristics; the quantity here is implicitly included because we consider the prices of all sales.

i hope you like discussing stuff like this otherwise i won't ask more

I like discussing what is useful of any kind, and always happy if I can be of any help. You are most welcome to ask/discuss.

You treat them the same way you treat the yearly dummy variables. For example, if you have houses with 1, 2 and 3 rooms only, you create three dummy variables, one for each. Then, you give it a value of 1 if it matches, and 0 otherwise. For example, if the house has two rooms, the dummy variable of "2 rooms" would be 1, and 0 for the other two.

@@AliNasserEddine okay but is it not also possible to add dummy variables and normal variables like rooms and then type the number of rooms or will that create wrong results?

I would add it as a dummy variables set as explained. This is the common practice. You may try using the number of rooms as a numerical variable, one for all; I assume this should lead to similar results. If you used the log price as the dependent variable, you should use the log of the number of rooms.

Hi Minyan, my email address is [email protected]. Sorry about that, I have just updated the link in the description of the video. Sorry for the late reply. Please let me know if you have any further questions.

You are most welcome. It can have both. It depends on the type of the variable. For example, for years, we use dummy variables, but for an area on the other hand, we use a numerical variable. Sorry for the late reply, please let me know if you have any further questions.

You are most welcome. This is right, you have to drop this pair unless you desire to construct a semi-annual index, or monthly or quarterly, and the pair doesn't fall within the same time period. Sorry for the late reply, please let me know if you have any further questions.

Hello Aliraza, the password is the description of the video. Sorry of the late reply.

Of course I can. Please let me know what you need. Sorry for the late reply.

Hi Le, Since I am replying late, I don't know if you still need assistance with this. Sorry for the late reply. However, I don't believe dividing the average salary by the HPI could be reliable for producing a new salary. Regarding the model, yes, you can do that: setting the wage as dependent variable, and the rest as independent variables. As per my opening statement, please let me know if you have any further questions.

Hi Conor, I haven't used such an interaction term before. By adding the distance variables by themselves (by cluster), you could get the result you desire. You just add the distance clusters as variables in the same manner we do for years. Sorry for the late reply. If you have any further questions, please let me know.

You are most welcome. The required fund is a calculated field. It is the sum of the total costs you set in the "assumptions" sheet. Sorry for the late reply. Please let me know if you have any further questions.

You are most welcome brother.

You are most welcome. The link to the template is available in the description of the video. Sorry for the late reply. Please let me know if you have any further questions.

Hi Olivier, the base year is normally set to 1 by default. If you desire to set it to 100, you just multiply the produced coefficients of the yearly variables by 100. Sorry for the late reply. Please let me know if you have any further questions.

You are most welcome.

Hi Augustine, All software is good. I advise you to use the one that you are most familiar with. Sorry for the late reply.

Of course I can. Copy of this exact study is not available because it is for educational purpose, but you have better ones in my channel. Please have a look. Sorry for the late reply. Please let me know if you have any further questions.

You are most welcome.

Yes, it is the same.

Sorry; I misunderstood your question previously. They are totally different. What is in common is that we use them for heterogenous assets, and both are linear regression models.